The current commitment by hyperscalers is around 1.7T USD, reported liabilities 1.3T and this year global debt related to AI is 570B. So that’s around 3T total. For this to make sense AI must generate 2T in new revenue per year by the end of the decade. And that would be only a 10% ROIC. For context ROIC for big tech is around 35% so at 10% they will be barely breaking even. The SP500 gives 10-12%. With 10% ROIC from AI the only thing investors will be celebrating is that the whole thing didn’t trigger a financial crisis. Data centers are NOT real estate. Buildings and power lines usually last 30-50 years. GPUs become obsolete in 5 years. If hyperscalers need to refinance and their interest rate goes up there’s zero margin for error.
H100 is nearing five years and costs more to buy a used one now than a new one when it was released :)
You are completely missing the bet these companies are making.
They think can outlast their competitors and capture a larger portion of the pie while the cost of inference keeps going down dramatically.
If you haven't been paying attention, the cost is about 1/100th of what it was in 2024. This is the trajectory pretty much every technology has followed.
Of course there will be market crashes and corrections and things like that and most companies won't survive, but the bet is that whoever survives ends up doing pretty well.
The fact that supply constrained GPUs holding value is negative indicator. It's like the tulips mania except bubble enthusiasts are buying wilted/dead tulips because live tulips supply constrained due to irrational demand. Now GPUs has more gross utility than tulips but seems like at current revenue/capex spend, every GPU is still negative net financial yield - they lose money - literally buying tulips and watching it wilt. Economically, better off simply not buying and losing more. That's the level of economic irrationality at place sustaining bubble, at least for hyperscaler/big tech balance sheet - small operators logic different and antagonistic to big operator demand/business model.
If cost of inference goes down 100x, would need 100x more demand. This makes overspending on GPU even more irrational. Jevons this, Jevons that but ultimately irrelevant. At end of day, leading players, hungergame winner candidates is saddling themselves with so much debt, even if they survive, post crash they are immediately uncompetitive against new entrant with blank slate and newer gen, more efficient GPUs that will be cheaper to buy/operate post crash when hardware prices will revert to mean.
It doesn't matter if some of the current players survive, they've basically stabbed and weakened themselves so much any healthy upstart in the future can wipe them out unless they lock in legislative protection... safety regulations, ban open source models etc.
That is the new bet, regulatory capture moat, because economic bet is entirely lost, especially with open models eroding mote.
You're making a classic philosophical error here by accidentally anthropomorphizing companies :)
A company losing out on a risky bet and failing, letting a new upstart rise up is pretty natural. A large fraction of experts from the failed companies continue at the new ones, business as usual. There are engineering teams at $BIGTECH now full of OS, database, or compiler experts from XP, Sun, HP etc.
This natural ability of companies to take risky bets is what made silicon valley successful.
Companies are run by people, 100 billion dollar companies are ran by people who want to stay billionaires and will street accordingly. But I'm not sure what we're disagreeing on, yes the talent will migrate, AI industry will eventually settle on some none bubble equilibrium, but that doesn't mean current AI economics is sensible, or inflated hardware costs beyond yield is not danger indicator. Like yeah, risky bets are burning, most people are going to move and carry their technical expertise on instead of unalive themselves or flip burgers, but that's independent of whether business models and broader economy is going to explode.
not to miss, future models will be more compute hungry too. Current hardware prices are still goin up and no it's not cheaper to run your AI for like %99 of the people because of lots of costs, it's not just hardware.
I think this fails to take into account how many people are fine with "fast enough" vs "fastest".
I've seen people happily use AI that takes several minutes to generate text or edit an image because to them they already aren't using their computer when they tell it to start; they just grab their phone and walk away and come back only to check in on it.
I feel like people here and on other technology discussions -- although it's worse here -- don't seem to parse what being the minority means.
They know they're one of the few to have access to such incredible hardware -- whether it be rented or purchased for way too much cash -- but they only see their own kin; their own ilk. They only compare themselves to the best.
The reality is that nobody expects data centre speed nor power in their own home and are satisfied to just go "haha its thinking" and let their computer quietly tick in the background as opposed to paying outragious prices for subscriptions or hardware.
That is completely discounting future capabilities and new use cases. Sure in 10 years you will have current SOTA locally, but in no way is it obvious we are anywhere near the limit of marginal value from improved capability
Unless you're a developer or doing complicated research you do not need much to use an LLM at home with consumer-grade hardware.
I'm editing photos with 32GB of DDR4 RAM at 3200mhz alongside a 3050 with a nearly decade-old mobo and ryzen 5800XT and at most? It takes 30 seconds and that's allowing the gpu to use my actual RAM as 'fallback' memory.
There are photoshop filters -now- that take longer than that before the LLM craze even began. Hell; I can train a lora in an hour or two.
This obsession with "more specs more data faster and faster" isn't going to win; it already isn't winning.
The per token costs plummet with more concurrents. A box that can do 100 tps at request depth 1 might be able to do 3000 tps at request depth 64. Less per thread, but massively more per GPU/joule/etc. That’s the economy of scale of running in a DC rather than locally that they were referring to.
> It worked out for Uber, they are wildly profitable now after spending a decade losing money.
Uber have a 10% margin, which is definitely not what I'd consider wildly profitable. (Their post tax numbers look better, because of accumulated losses).
Uber is a two sided marketplace and was still famously unprofitable for at the same time.
Over 14 years, Uber burnt ~31 Billion dollars and in nearly whereas the amount invested within AI seems to be within Trillions at this point and in near future with a product which doesn't have much moat and shaky financials on profit.
Is this true? Hardware costs have only gone up during this time. Are you referring to electricity cost to serve these models? (i.e. compute got more efficient?)
> They think can outlast their competitors and capture a larger portion of the pie while the cost of inference keeps going down dramatically.
We are also within an arms race of training newer larger models with more speed while discontinuing older models.
Gemini/Chatgpt have already discontinued their models from 2024 (iirc) because they are using all their compute in serving/training newer models. Being quite frank, nobody is serving a model from 2024 as the intended use-case while having very little moat as open source models are catching up.
> Of course there will be market crashes and corrections and things like that and most companies won't survive, but the bet is that whoever survives ends up doing pretty well.
How so, by raising the prices? because the current prices aren't sustainable and I feel as if there would certainly be companies which will try for one reason or other to be cheaper to capture the market share because of the larger promise of whoever is able to get as market share. I had once thought about it and I don't think that even in an ideal world, they would end up doing pretty well given no moat.
Also even if a company survives and ends up being one of the survivors and makes profit in the ideal scenario you mention, then within some years other companies will try again and construct more datacenters and end up driving the prices down for everyone, so nobody knows how things might look down for 2-3 years let alone a decade, so I remain a bit skeptic currently so.
I had actually thought some on the economics of datacenters and I found it to be very related to power. The only ones which seems to be making money might be the power generators actually because power is the actual bottleneck rather than GPU's in datacenters from my understanding.
Though the power is raised at the cost of electricity bill increases for everybody including people living in houses. The job prospects are minimal as well, as a nation, aside from just getting investment just for the sake of it because AI's trendy right now, I feel like its a net negative deal for people living there.
The “contents” represent the majority of the cost and meaningful functionality of what we call a “datacenter”. Those contents will not last for “real estate” debt timelines.
Right but their payback period is insanely short - a $5M GB200 NVL72 cluster is expected to generate $75M in revenue over 3 years for inference providers. That's a 3 month payback period.
AND - they're operating well past their estimated service lifetime.
Chances they're planning on replacing personal, local compute with time-sharing on data center hardware that's too outmoded for AI...? You know, since they sunk the consumer component market for the next half-decade.
I would be more careful before assuming 5 years depreciation schedule. Currently price tags are attached to computing power, not the production cost. Computing is not getting meaningfully cheaper with newer GPUs but it only allows better scaling, which makes older hardware more relevant for many use cases. This is why A100 is still selling like hotcakes. I don't think this trend will change soon.
Apple can enjoy because it controls a significant fraction of consumer computing platform so they can simply collect tax from everyone else. This is not true for the rest of big tech. Only Google has Android but it cannot sit and enjoy because they still don't control hardware and AI is an existential problem for their search business.
The GPUs are far from worthless after 5 years. E.g. the A100 80GB PCIe version cost around $15k when it was introduced in 2021 and now sells for $10k used.
Things might be slightly worse for the data center servers, but I am sure they will find find buyers.
They only reason that they are retaining value is there was not so much demand for GPUs in 2021 as there is today. Once the demand drops you will find then in dumpsters across our barren, burning dystopia.
They hold value as there is insane demand. The same reason a consumer RTX4090 costs more today than bew in 2021. Once the tide drops enough for hardware lead times to shorten to weeks, they will go the way of other used DC hardware - written off after 5 years.
For what its worth, SK Hynix CEO has every incentive to show that RAM prices will remain high for as long as possible because that is the only thing which is floating their evaluation to such astronomical amounts.
Independent estimates sort of show around 2027-28 from what I remember.
I remember reading some article which said that RAM prices are already going down from its peak slowly (IIRC I can be wrong, I usually am but 3-5% month from its absolute peak) but the current RAM prices are still astronomical given past rates but the RAM prices will slow down hopefully sooner rather than later.
No one knows. There's a known bullwhip effect in supply chain [0], and DRAM makers are pretty far out along the supply chain. Just like it ramped up wildly it will stop even faster.
Not only that, but they're typically amortized over 5 years, where the actual lifespan usually falls far shorter (1-3 years), adding to the artificial subsidy conditions we see today. So they're gaming the lenders into deferring interest payments as much as possible today so that new competitors don't have the same cheap financing advantage.[0]
If they're deliberately inflating the likely useful economic life of their assets to get a lower interest rate, it's hard to see how that wouldn't be classed as fraud.
It's the sort of behaviour that really does end up with people going to prison.
You’re all getting some concepts mixed up here. That five year amortization rate is the IRS’ usual amortization rate for computers. GPUs are classed as computers for asset depreciation purposes. But GPUs are part of 168(k) so they’re eligible for a 100% bonus depreciation the year of purchase.
There’s nothing fraudulent at all here just people using terms they really aren’t comfortable with.
That sounds reasonable, it's "just" $1k/yr for 2B workers (there are about 1.2B total "knowledge workers" in the world including gig drivers), or $10k/yr for 200M workers (there are 70M office and technical workers in the US). /s
In 4 years it better be 10x more important to have than a cell phone is today, or 10x more important than having internet/monitor/pc/printer is for an office worker today.
The math looks good on paper, but in reality enterprise AI is hard, most companies are realizing they are actually not seeing ROI from AI. One of my customers took about 8 months to rollout an AI initiative that by the time it launched and people got trained on it, it was already legacy. Also if you are 10x more productive with AI that doesn’t necessarily increases your billable output. There could be some super models like Mythos aimed at very specific hard tasks like drug development, but we have not seen any of that yet, and the clock is ticking.
Yes, I'm agreeing with you. There need to be 200 companies willing to pay $10B/yr for this. What is the ROI? That's the pay roll of ~half the work force of the largest 200 companies. Unless you can fire %50 your employees, everything else is a sunk cost you already own.
I just don’t understand this view. This is the most significant technology ever developed. The uncertainty currently is whether it 1) has massive impact, completely altering society and the making world significantly significantly better or 2) if we go into a fast takeoff/rsi loop. Personally I’ve always been highly skeptical of the later, but that seems like a genuine possibility now. It’s not ‘are we going to be able to generate 10% roic on compute’ the answer to that is yes.
It will obviously be a large part of the economy like online shopping is today. The companies that built up a lot of debt to be brand names in online shopping primarily went bankrupt because new companies had no debt (and perhaps no negative sentiment from early customer experiences.)
the difference is all the current capex is going to durable, hard to get physical assets + things like PPAs. In your online shopping analogy, the hyperscalars are acting like Amazon in 1998
What destroyed hardware manufacturers after the first bubble burst was all the old but overpowered server hardware ending up in resale after bankruptcies. Hard to get hardware is a bad investment that is also potentially obsolete after new optimizations make the next generation much more efficient. The companies that think their individual optimizations are going to outpace industry wide optimization are delusional, historically speaking.
Looking at cash burn is looking at the wrong end of the horse. Some companies, like Meta, have burned huge piles of cash in pursuit of, for example, the Metaverse and they've got nothing to show for it, not even a slight increment in ad tech, and yet they earned enough to shrug it off.
There's a big difference between Google spending tens of billions on AI infrastructure and what Oracle is doing. Oracle is spending to get on a bandwagon. Google is transforming their business, so far seemingly correctly. If AI flops big-time, Google will be left with some stranded assets, but it won't be existential the way it would be to Oracle.
If AI flops, they’re be left holding large pools of useful datacenter/compute capacity and “revert” to one of the most profitable businesses of all time.
Smart glasses are about a 15 million unit per year business. Smart watches are about 150 million units per year. Meta owns about half the smart glasses business, but it hasn't provided them a platform they can control that has enough penetration to make a difference.
But wait it gets worse: since Meta hasn't got a platform comparable to a mainstream PC or mobile device OS, the Meta glasses business is vulnerable and subscale, as impressive as owning half the TAM is. Since Meta glasses aren't a companion to an existing platform (not even AndroidAR, even though they run Android) they will become a second tier choice as soon as Apple or Samsung ship a smart glasses product.
I'm thinking Apple has been really smart in their AI strategy here.
It seems a mistake to make unprecedentedly large capital expenditures, in a very very crowded space, without much evidence of a moat. Presumably people thought the moat would be singularity-like self-improvement of AI, but the singularity is merely a religious concept, and nobody should take religious myth as fact, it's merely narrative for orientation and inspiration.
Their strategy to let Siri stagnate for 15 years and let everyone else take that market? Their strategy to put a bunch of not ready for consumer use AI features on their devices and then roll them back?
They just have such a strong hardware + os ecosystem that they can sit on the sidelines. They'll be able to negotiate with some LLM provider at a good discount when the time is right and put harnesses around it for actual useful features.
I was talking more about massive investments in compute and model development. Voice assistants are barely AI, but real AI might help make better ones in the future. Similarly, I think Apple's failed attempts to add AI to macOS and iOS are a huge justification of avoiding capital investments in compute and AI frontier models.
There's two separate things here: 1) the underlying models, and 2) how they are applied to real world tasks. Both are changing at tremendous speed. The underlying models, 1), can be purchased as a component, so there's little advantage to vertical integration, as long as there's a robust competitive market, and boy is there ever. And Apple was never going to be the big B2B provider for LLM compute and models, such sales are just not in their corporate DNA.
For 2), application of AI in Apple's products, nobody knows how to do that correctly yet, and it's going to a wild few years of people trying things and failing before something comes to pass. Waiting to see where to actually find a competitive edge totally makes sense. Microsoft's Copilot bungling was far worse than Apple AI's bungling, for example. I can't even get access to Microsoft Office anymore, it seems, I'm instead confronted with a dumb chat interface that doesn't even know how to launch a new document or open existing ones. At least Apple never destroyed their brand like that...
I think that's precisely what the previous commenter is saying. Sit on the side lines, and let other people bloody themselves up.
Similar in a way to dot com. It's not to say ML won't have practical application in the future, but the likelihood that it will have specifically this form is low and worth waiting until the dust settles and a more commonly accepted utility presents itself.
If AI/ML were monstrously useful in its current form the companies pushing it would not need to be hawking products; people would be bashing their doors down. I think that's why in areas where it's more directly applied to a known problem set (like Pharma research, and I'm hoping someone with Pharma expertise can pipe up here) there has been more natural pickup.
Coming from trading and markets, ML has been a part of the mix in quantitative strategies for...well, nearly 20 years (by definition I suppose). Spaces with obvious utility will see rapid adoption. Worth waiting that out, honestly.
The point was that one rational approach is that it's OK to not be top ranked in a market that has significant medium-term profitability issues. Apple has no problem creating new markets, but not if being top ranked requires years and years and years without return on investment after launch, and large subsidies. Not to say Apple hasn't failed at creating new markets either, or anything like that.
Claude is a far superior _model_ to whatever Siri uses, but in terms of the Claude app correctly transcribing speech, in my experience it's as bad or worse than Siri (which is already bad). Doesn't matter how good the model is if it misinterprets what you say.
Google and Meta's moat will be their ability to set untold billions on fire. OpenAI and Anthropic can do it for now but once the bubble pops they'll be fucked.
i dont understand the concern. they are putting up great financials. you have to invest ahead of the outcome. this is just classic quarterly public company earnings BS, where public markets dont reward innovation investment. they just want crank the handle financials.
The bigger issue is on the model front, can Google compete; Gemini doesnt seem to be able to compete on the heavy expert end; they are doing well on lighter faster models.
People bought Google for the torrential free cashflow, that looks like its gone forever with this new capital intensive model. If that the case then it needs to be valued like a heavy industrial rather than a capital light tech company.
I agree Gemini's value is not at the frontier, but they are making very useful smaller models. 3.5 flash lite is super fast, cheap and token efficient and accepts any kind of input. For large, price-sensitive processing, in consolidated workflows that don't need the latest improvement, it is the best. Perhaps they should focus on improving their dominance on this business, I don't see gemini reaching claude/gpt in coding. But they could be the ones that make it possible, and economically viable, to roll out llms in large scale automation.
I've given up on Gemini. It sounds smart but most of what it tells me ends up being wrong or misleading. I might actually hand $20/mo to OpenAI. It's been far more helpful with the random collection of legal and health problems I've thrown at it. My recent comment history is going to make me come across like a shill for them but, holy crap, GPT has been doing amazing things for me at work as well.
I don't get it either... Google has so much talent yet they just can't seem to get it right.
At work it's paid but in my own personal queries it's free vs free and gemini seems to give erroneous answers with confidence frequently and they will completely flip flop in subsequent queries of the same text. It seems very unstable.
> this is just classic quarterly public company earnings BS, where public markets dont reward innovation investment
Genuine question; but aren't these treating stocks as speculative and on vibes? One can say that these comments could be true for the first signs of cracking of dot com bubble. Sure, Web eventually succeeded but many tech giants from dot com era (AOL/Yahoo and so many more) eventually went to dust for spending too much time on the innovative bandwagon.
During the Dot-com bubble really tried to give this example but IIRC there were companies like pets.com who lost 2$ for every 1$ of sale so how a company treats its financials do matter a lot.
The market doesn't seem to reward innovation sometimes because there have been times the first persons to innovative have actually really failed to capitalize on that innovation and many extremely innovative businesses like Airlines (We can literally fly speak of innovation!) have been terrible businesses investment-wise generally speaking.
They just raised $85 billion and they're sitting on a mountain of cash - if their spending didn't increase in this context, it'd be bad management. The real story here is that they have decided to spend that mountain of cash on AI CapEx.
I'm referring to the equity offering that started in June and has a second component that starts in 2026Q3. Most of this raise came from the sale of Class A and Class C stock. A fraction came from the sale of convertible stock. To my knowledge, none of this raise came from the sale of bonds.
Also, looks like I got it wrong and they've only raised $45B to date. The rest will come as part of the ATM offering program that begins in Q3.
All these big tech companies are fighting over the basics eventually like power and transformers and don't like to do anything dirty that would hurt their ESG score like getting into any sort of industrial business. Thus, the default is all that stuff that heavily bottlenecks American AI gets done in China.
If you listen to Tesla's recent conference call they are going to making solar panels all the way back to making the silicon ingots and totally vertically integrate. Elon lamented on a previous call that nobody wants to get involved in these primary industries and he has to do it all himself unless he puts his whole supply chain in China. For example, Tesla recently opened a state of the art lithium refinery in Texas cause nobody outside of China does that anymore. He's opening a new fab, because everyone else is too hesitant to expand to meet the capacity he needs.
The alarms in this case are that the profits and margins won’t be as high as we’ve come to expect from cloud companies.
Other than Oracle’s questionable spending spree, these big tech companies are still in very good financial positions. The enormous R&D and infrastructure spends are just feeling unusual to investors who got comparable with the unusually high margins and low costs for SaaS companies. Now they have to put a lot of that money back into the business like more normal companies.
If the margins aren't as high then there will be a repricing for all the massive cloud companies, which means several trillions worth of valuations to be cut from the companies.
AWS/Azure/GCP/Oracle/SpaceX/etc neoclouds... are worth a combined 10+Trillion. That going down by 50-70% is going to be insane.
This is a good chart that shows historical CAPEX spending. Hyperscalers have been through a couple CAPEX cycles like this, they all know what they are doing.
Thanks for sharing.
Agreed it's a good chart. But I draw a different conclusion. M$ looks pretty iffy: capex/revenue 10% -> 37% in the last 5y, scary trajectory.
Why? GPUs are replaced every 3 to 5 years. This is going to be an ongoing operational cost forever. It will probably increase more if larger models require bigger VRAM sizes.
We have probably hit a limit to scaling LLMs through raw parameter count alone, at least we're not seeing the exponential pace. I personally think we'll end up with a nice sigmoid curve plateauing in the sub 10T parameter regime. The amount of tokens processed (in inference) is increasing exponentially though (I've been following open router usage stats for years and it's always been exponential). We will of course make technological advances in hardware efficiency, and model parameter efficiency, but I think a much more plausible future is that VRAM needed for loading and serving individual models will slow down or even stop. We will need more chips, and more power, as demand continues to grow of course, but the operational lifetime of GPUs today will be a lot longer than the SoTA cards from 5 years ago.
They are building new datacenters for the AI demand, so around half of this CAPEX is not for the GPU-s, and those will not be replaced every 3-5 years.
Also, TPUv2 was introduced in 2018, and still not completely retired in all regions, from accounting pov, it has been written down to 0, but they are still working.
The upfront cost of facility is ~30%, network infra 10-15%, land/utilities is small percentage, power could be significant for an AI DC. The servers are ~50-60% only.
That cost has always been there and allowed for their lucrative margins. It's the upfront cost of building/populating their datacenters (many more than before) that is eating those margins.
I mean the issue is scaling, the worlds for cloud never kept getting bigger and bigger and compute scaling had stopped a while ago in the CPU space.
With AI every new generation with both massive hardware and software stack changes from Nvidia makes prior chips extremely inefficient to run, basically we are comparing an ASIC industry to a general purpose compute industry where all work loads are the same shape and size and so on.
Margins for ASIC based mining companies or ASIC solutions providers were never high, Optane and other weird solutions are niche and great for a specific category or moment in time, but they become obsolete pretty quickly.
The fear is we don't know if this Capex can stop.
The worst type of fear is if this Capex will stop then what? Someone is very overpriced in this market, the cloud companies, the hardware providers or both.
I don't see how we reconcile this without a massive wave of repricing, ofc markets can stay irrational and we don't see the actual books but AI doesn't have so much revenue. Suddenly the AI token/cloud revenue won't 100x in a year or two...
Especially when intelligence will continue to get cheaper, the margin compression is a massive risk.
All the data centers for hyper scalers were a miniscule part of their story the real moat was the software layer on top otherwise Hetzner would be priced like an Amazon as well.
Something is shaky with this market I don't know what it's very opaque even as an insider working on for big tech and startups. I have no clue who falls first and which bottleneck cracks but there is not enough revenue for tokens, we will see a strong 2-3x growth in the next few years, from here which is absurd, but it's not enough, not nearly enough. If the capex keeps high and increasing.
Ofc they can stop the capex and the otherside gets repriced it's not like nvidia, micron and co aren't worth trillions.
Cannot hear what you’re saying with all those alarms blaring non stop since a year. Someone should do something about them, maybe turn them off, I don’t know
> Everyone is in too deep to now admit that there’s a problem
I'm not sure how to square this with the dramatic improvement in LLM capabilities in the last 8-9 months. If anything, it makes the earlier investments look prescient?
"Anthropic and OpenAI generate a lot of revenue with relatively few employees – an estimated $9M and $5.5M in revenue per employee (RPE), respectively. If either company were to go public, it would have a higher RPE than any public tech company on Forbes’ Global 2000 list." https://epoch.ai/data-insights/revenue-per-employee-ai-compa...
Maybe, but that's a different claim. You wrote that the improvements are "not translating to a dramatic increase in revenue", but going from about $1B to about $30B run rate in 16 months seems like a pretty dramatic increase to me!
No, I wasn't claiming that revenue scales with headcount, though it probably does to some extent. The point is that these companies' revenue is large and growing quickly, which is what the comment above mine denied.
This assumes they do not have to increase prices to be profitable, and that they will continue to have customers when customers can switch to open models at similar performance.
As an analogy, Uber could crank up rates after the VC growth play was over to stoke revenue and profits because they have a duopoly with Lyft. LLM consumers can switch to Kimi models fairly trivially today, and whatever the frontier open model landscape looks like later. Model training and development is expensive, self hosted inference on open models not so much.
(a component of my work is currently building scaffolding so our organization can swap out commercial inference providers for on prem inference infra to derisk against the eventual rug pull when the math gets icky for LLM providers, while consuming as much subsidized tokens as we can until then, when it makes sense to use tokens for work)
Those seem like reasonable questions about future margins and moats, but I was making the narrower point that revenue is in fact growing quickly, contra the comment above mine
The question will be whether customers can switch.
Can you install a near-SOTA model on a cluster in a data center? Of course. Compliance and operations are the sticking points. I work in healthcare IT, and it's amazing how tight the data compliance requirements are. I can't have someone in Canada look at prod data. If we told hospitals that we were handing off PHI/PII to Chinese models, they'd end our relationship due to the long history China has of hacking Western networks and computers. They don't care how open and cheap things are.
Then, you have to keep up-to-date on the latest technology and right-size things in a very fluid market. If you sign a contract for hosting the model on a data center that's running what the SOTA is now in hardware, and someone comes through with a data center hardware or software product that makes that data center contract a disadvantage (maybe it's too expensive and the other party won't budge on the price), you might have to factor that into your offering's price, and that could put you at a disadvantage in your marketplace.
Google, MS, etc. all want to leverage the cloud model to make this be less of an issue for you, for a price. They have the ability to update you with the SOTA stuff in the data centers, because they're the ones driving that SOTA. They can say they host in the US and develop most of their stuff in the US.
Will that be enough of a moat?
Probably not for the levels of spending that are happening now, but over the long term, probably.
How long will a SOTA model be necessary? If day to day work can be achieved on an open weight model, the most evaporates overnight.
Look at any computer in a big company. It isn't the fastest on the market, nor will it have the most RAM or largest monitor or fanciest keyboard. It is good enough at a good enough price point. Once it becomes possible and cheaper to host your own good enough open weight models, with all the benefits of keeping data internal to the company, then the big providers are cooked, so to speak.
> How long will a SOTA model be necessary? If day to day work can be achieved on an open weight model, the most evaporates overnight.
Depends on the advantage it gives people and marketing of that advantage. You'd be surprised at how overpowered the average workplace laptop is. Each company I've been at has had at least some people who do non-technical roles using high-end hardware. Why? Because the account executive wants the fast machine and they get what they want.
You can apply the same to GenAI. Humans are notoriously bad at estimating actual needs when it comes to resource consumption. Best to have it and not need it than need it and not have it, especially if your competition just shelled out for SOTA.
And that's not even taking into consideration regulatory and customer concerns about where the AI you're serving requests with came from.
Execs getting what they want doesn't mean they let everyone have the same thing. It's far more likely that everyone else is using lesser equipment.
We have already seen tech workers at big name companies get whiplash from "leaderboards showing people using the most tokens!" as a good thing one month to being pressured to using fewer tokens a month later.
> Execs getting what they want doesn't mean they let everyone have the same thing. It's far more likely that everyone else is using lesser equipment.
You'd be surprised. I've seen people request upgrades and get them. Not even execs, just employees. Even the average devices are probably overpowered these days. Chromebooks could do the vast majority of work in a corporate environment. Try getting workers to accept them, though.
Regardless, there are probably enough reasons to use proprietary Western AI for the time being, particularly in regulated industries, that poop won't completely hit the fan. Particularly if the CTOs start searching the phrase "Operation Aurora".
> We have already seen tech workers at big name companies get whiplash from "leaderboards showing people using the most tokens!" as a good thing one month to being pressured to using fewer tokens a month later.
Indeed. What do you do when you want a group of people to do something that they'd be otherwise adverse to doing at all? You rank them on it and let them get in a competition over who can do it the most. That's what tokenmaxxing was about: getting people to use the AI at all. Now that the people are using AI, we move onto another objective, which is getting them to use the tokens efficiently. The trick is measuring that efficiency.
My primary role is cybersecurity in a regulated entity in a regulated industry, I am highly confident it is straightforward to do so based on work accomplished in only a couple of weeks. Stand up a router, stand up a Kubernetes cluster if you don't have one, stand up the necessary VMs and compute for serving inference. Two pizza team, in my experience.
Customers can switch (although we can argue the speed and pain of doing so), and the speed at which they do will be a function of cost efficiency and demonstrable value (imho). A recent example of this is Broadcom and VMware [1], for example. When motivated, it can be done. If there is no objective, measured value being delivered, the spend will be cut. If the value delivered is measured, it will be enabled at a lower cost through cost optimization measures (ie self hosting) [2].
This is all to say: there is no moat, the revenue of inference providers is volatile and not assured in any measure. Caveat emptor.
Meanwhile, in real companies, you have to wait 2 months or more to access an API endpoint in preprod.
To setup a cross business kubernetes cluster will take 2 years with unknown results.
On Cloud, in Switzerland, you need to call Microsoft when you need new resources, so much for agility and minute infrastructure provisioning, and I heard the same for AWS.
> Meanwhile, in real companies, you have to wait 2 months or more to access an API endpoint in preprod.
> To setup a cross business kubernetes cluster will take 2 years with unknown results.
Do you seriously believe those times will not go down 95% if the CEO pushes for it to get done yesterday because it will save the company millions in expenses?
You get it. Given sufficient incentives, processes and systems become potentially more malleable, and hard requirements can become optional. Speed is a function of appetite, will, and resources.
There's a difference between "It's straightforward to do" and "I can convince a customer to sign a contract allowing us to do it."
If the second one were as easy as the first, I wouldn't have to be online at 9:00 to deploy stuff to prod tonight; the team in India would handle it. But customers write into the contracts that only US-based employees interact with prod systems. No amount of cajoling will get them to change their minds; they have data sovereignty, international telecommunications treaties, and HIPAA compliance to worry about. So I'll be pressing buttons tonight.
Could you swap out Anthropic or OpenAI or Google or whoever's models for Kimi? Yes. They're like other software these days, they're modular. What isn't modular is regulatory and geopolitical concern.
Right, and the comment I replied to was about revenue, not profit. (That said, while I don't think Anthropic is already profitable, it reportedly expects its first operating profit later this year.)
The technology is too hard to capitalize on. It’s far more democratic than, say, an iPhone, or a search engine. Anyone can download a model to their computer and start toying with it, how do you profit off of that? Even if everyone was constantly tokenmaxxing (which we cannot, since the process gets fucked up if you let it run entirely on its own), it probably still wouldn’t be marginally profitable.
Ans so far, the dramatic improvements have come with an increase in API costs.
Even if, hypothetically, Fable or a Fable-class model could seriously replace some headcount, it'll only gain further traction of it's actually cheaper than hiring humans. $50/MTok is expensive. Wouldn't be unreasonable to expect somewhere between ~$3k-$5k/month/developer in spend. Cheaper than a Junior in the HCoL areas (in the US), but not much cheaper in lower-to-average COL areas. Most acceleration will come from having the headcount + giving said headcount $3k-$5k/month in token budget, so now it just becomes a very expensive dev tool rather than a headcount replacement tool.
The idea that a $30k/year API bill will replace 2 $100k developers falls part outside of SFC/NYC. No CFO of a mid-market company in a LCOL area is signing off on $3k/month/dev API bills. They'll just hire juniors and cap their spend at $200/month.
(The claim felt so wild I wanted to check, and indeed, the private Google Cloud for the $125bn Australian pension fund was accidentally deleted by a provisioning misconfiguration. Any others?)
IIRC, the files for Toy Story 2 were accidentally deleted during production, and the film was only saved because someone on maternity leave had a backup at home.
Yes, Google randomly deleted UniSuper for basically the same reason they randomly ban individual customers: they don't care. Relying on them for anything is a huge mistake.
Source? Has Anthropic's annualized revenue not quadrupled in the last 7 months? And OpenAI's annualized revenue quadrupled since January 2025? Which is only unimpressive by comparison to Anthropic's meteoric revenue growth
I'd be with you if you claimed that the revenue hasn't translated into substantial profits. Being able to spend a lot of money to get less money back is not that impressive. But revenue by itself is on a dramatic rise as capabilities improve
Moreover there’s no guarantee that eventual AI profits (if any) will go to the companies investing all this cash. If the worst case scenario of Chinese labs building and serving frontier-level models on 2nd tier nvidia hardware comes to be then what will be left of all the “hyperscalers”?
Exponential growth when you're starting from zero is neither difficult nor sufficient in this case. The title of the linked thread is "Dramatic cash burn." So clearly, the revenue did not grow anywhere fast enough.
I’m saying there is no proof that companies _paying for AI_ are seeing a positive effect to their ROI. If you have such a proof, please share, that would be a massive news
I know it's easy to forget, and InsideOutSanta kind of anchored the conversation on "revenue", but profit is ultimately what matters. Back when Silicon Valley was merely insane rather than bat-guano crazy insane, it was commonly observed that it's not hard to build a business around selling a dollar for 95 cents. The point being that it doesn't necessarily mean much when you have a business doing that, because of course the demand will be insane. It doesn't mean you have a viable business. You don't know you have a viable business until you transition to selling a dollar for $1.03. Many a VC-funded business that looked successful, even wildly successful, has run aground on that transition, or at least, suddenly stopped looking so wildly successful.
If AI-related expenses are also growing exponentially, and they are growing exponentially faster, it doesn't matter that revenue is growing exponentially.
The AI funding has also now absolutely baked in exponential growth of expenses, because that's how debt works. A slow exponential, hopefully, but an exponential none-the-less.
Something Hacker News needs to be periodically reminded of is that we are the field getting the most out of AI, and it's not even close. That's great for us. But the stocks aren't priced for "a pretty nice coding tool". They're priced for every field in the world getting even more value out of this than our field is getting now. That is, frankly, not happening anywhere near fast enough for the spending and stock valuations. When you don't have all the engineering guardrails that are present in software engineering [1], suddenly the AI is, ahem, exponentially less useful.
As I say in that post, watch your AI actually doing something, even the frontier models. Watch the thinking traces. Watch how many times they bang into a guardrail of some sort; a failing test, a failing compile, a linter failure, a bash script that doesn't work, all those things. How much value would you get out of an AI coding assistant if the first time it banged into a guard rail it was done and you had to stop using it for that task? How much value would you get out of an AI coding assistant if instead it silently failed and just proceeded forward with errors that you lack the infrastructure to easily detect? In the first case, it would be fairly modest, almost certainly not worth the money, and in the second, it would be worth paying to not use.
Even in our field, while the rate of code output has increased substantially, the rate of value generation increase has been quite a bit more modest. I have observed, and heard from a number of other places, that while my own output has increased somewhat we still generally can't plan on being able to work with other teams at much faster a rate than we used to.
There's a viable business here but I can't see how all these companies expect to be returning all this revenue in any financially sensible period of time. They're all spending like if only they spend enough they can own about %900 of the market in three years. They can't all do that, even accounting for "AI makes the market bigger".
And they're wildly vulnerable to some new solution coming out that obsoletes all this spending, like an ASIC that starts running a popular model directly (especially if model capabilities plateau), meaning that all this nVidia GPU spending is so much dead silicon. Or someone comes out with a much more efficient way to train models. There has to be some insight we're missing; humans do not learn what they do by having the entire contents of the Internet poured through their head hundreds of times over. We are far more efficient with our training data. What if someone works out a solution to that and we don't need to spend billions on GPUs but only millions? The whole spending proposition could collapse overnight and the companies that suddenly have three orders of magnitude too much hardware and the debt to match would be up a creek without a paddle.
I very much agree with this. Even the top tier models today, without the unit tests, without integration tests, and domain experts reviewing the code would flounder for 50% of the work they do. Sure they can write the unit tests and integration tests themselves, but at that point you aren't in need of a specific system being built, but rather an out of the box solution would probably fit your needs. It does speed up the grunt boilerplate work of development quite a bit, it does help with gnarly bugs and the like, but expertise is still needed. And we as engineers/programmers have systems in place that make using AI easier, we have the human context windows to be able to parse the technical jargon the AI spits out. Will AI for the masses be akin to slightly better automation?
> we are the field getting the most out of AI, and it's not even close.
Just emphasizing that as, due to spending far too much time online the past week, I've been seeing a fair bit of this. "AI is definitely gaining popularity because all the software companies I know are going all in on it."
For certain values of ‘dramatic improvement’. Is lots more important work being done with LLMs? Not much sign of it yet, they’ve been helpful for experts at times (e.g. vuln research or maths research) but that hardly justifies the vast sums for Google investors.
Presumably at some point you need a measurable productivity return yea? Maybe organizations are not built around skill and aptitude so much as liability, which LLMs cannot provide barring (very welcome and also very unlikely) legislation in the US.
The infamous 2025 MIT study that found almost all AI pilots in companies were failing, also found that virtually every worker was using AI many times a week if not daily.
Turns out people just use their personal AI accounts rather than company ones. Which would make sense if you want to claim the work the AI does as your own.
>Presumably at some point you need a measurable productivity return yea?
At what point? This technology is brand new. Did you think we were going to double productivity in 3 years?
Capacity is being built. It's hard to build data centres, there are no chips, there is no memory, it's hard to get talent, we don't have the energy to power the facilities.
No one knows where this is going. We are scratching the surface. There is an absolute boom happening, and yet every day I have log onto Hacker News and read this nonsense about everything falling apart. Are we living in the same universe??? So-called "technologists" saying, "meh, it's not that cool". Okay.
Guess what? You're not Michael Burry. Nobody cares or will care that you "called it". Look around this place: you aren't even slightly contrarian.
"We would be profitable if we had the resources but we don't," isn't the smackdown argument you seem to believe it is.
There used to be a thing where successful tech companies were profitable right out of the gate, and very successful companies doubled those profits for years, and companies who bought and used the tech could point to clear, actioned, benefits and cost savings.
Now it's all "This will be really, really profitable one day, probably, if the omens align and we can deal with all of the problems."
No, because the LLMs will keep getting more efficient and capable. Distillation and quantization will mean firms spending trillions on giant data centres are left holding the bag. I suspect Apple ends up laughing all the way to the bank.
Tech is real, impact is gigantic, long term winners hard to predict, capex spending hard to recoup soon, if ever.
And differently than internet or rails, you don't build once and maintain later, but enter a loop of ever increased spending to keep on top of the arms race and ever exploding usage.
I'm not sure I've seen what I would call dramatic improvement since maybe GPT4?
Sure, things got better. But I'd call it iterative more than revolutionary. I still wouldn't trust any of the models to do anything meaningful unattended. They all still do dumb shit all the time.
Plus, even if they were genuinely dramatically better, the businesses sure as hell aren't. They're burning money left and right, they have no moat, Chinese open models are basically equivalent these days. What's the path to profitability, or hell, break-even? How do you envision this being anything but a giant money pit?
I mostly use anthropic models, but there was a big step function when claude code came out, and it’s been incremental or a plateau since then.
Opus 4.6 and 4.8 are basically indistinguishable from Fable and Sonnet 5. 4.7 was a hot mess. The guardrails on 4.8 and 5.0 make them worse than 4.6 for many tasks. So, even if Fable is theoretically better, refusals/downgrades make it a worse product in practice. Who cares if it outperforms on 1-2% of real world tasks if 5-10% of tasks are blocked?
I’d bet most people could be downgraded to a 12 month old frontier model, and not notice for a week or so.
Anthropic’s big problem is that open weight models are 0-6 months behind. So, their product is commoditized and margins are never going to be good.
They've literally rated the debt as too big to fail in order to get foreign sovereign wealth funds (mostly gulf states) to agree to put up the money for loans. This has been happening this entire time.
What would be the best thing to do with ones investments considering these alarms?
Say you had some money in cash rn, what should one do? Wait for a crash and buy stuff up cheap? Put it in some safe category?
This stuff is stressing me out and I do believe it's gonna come crashing down sooner or later, but I don't know enough about investments to know how to best come out unscathed.
Diversify! Historically, the average length of a recession has been 12-24 months. So set up a system whereby you won’t screw’s yourself over by selling when things are low, but instead you can weather the storm.
Build a rainy day fund. Determine how much cash you will need if you are out of a job and how long you think that will last, allocate some portion of that amount into low risk bonds. Russ way if you need cash you aren’t selling investments at a big loss.
If you have enough liquidity put some in real estate as a forced savings vehicle as it’s harder to liquidate than stocks. Then just sit out any coming storm.
If there's a big AI bust, there will be no escaping it. Like 2008, the entire economy will slow down. This time it might even end in a war. A well diversified portfolio e.g. index funds will weather the storm and recover.
Build your emergency fund first if you don't have one. 6-12 months of salary in cash or CDs. Then dollar-cost average into well diversified equities. Don't watch them day-to-day. You're concerned about their value in 20-30 years, not tomorrow.
Nobody can tell you what to diversify into without information about what you are concentrated in.
Sure, spread investments across stocks and bonds and treasuries from different markets.
But you can also diversify more broadly beyond economic capital to cultural and social capital. Learn new skills and build networks of generalized reciprocity with others before you (or they) need help.
I see everyone around me doing way more work, of way more depth, than they ever did before using AI models. I see my company and friends of mine all paying large sums of money to Anthropic, Google, OpenAI to use AI models, and do more work than we did before.
So Google is investing in infrastructure which is HIGHLY in demand, there is much more demand than supply, and then they are making money from this infrastructure...
That's a good thing for Google, and as an investor in Google, I am glad they are making these investments.
> The search giant now expects to spend between $195 billion and $205 billion in capital expenditures, its finance chief Anat Ashkenazi said on a conference call with analysts. The company said last quarter that it planned to spend between $180 billion and $190 billion this year.
> "If there is a huge demand for shipping goods internationally,
> investing in ships and planes isn't burning money.
> There is massive demand for compute in the world right now"
Emphasis on right now. CapEx makes sense if the demand is forecast to deliver enough profit over the expected lifespan of the investment to recoup the cost and margin.
There's enough hype and exuberance in the AI market that it's likely some players are going to be left holding the bag with a write-down on assets.
Serious investors look at balance sheets, less then what CEOs say. Elon Musk -- as an example-- says all kinds of things that don't really happen. Mark Zuckerberg is arguably less grandiose. When FB changed their name to Meta, said they were committed to the metaverse the stock didn't dump. When the really big investments in consumer VR hit Meta's balance sheet, there was a big drop.
Think of it as the difference between the waiter describing dishes with ingredients you don't really understand (or maybe even taste) vs presenting the bill for the meal.
If a company’s value was completely representated within their balance sheet, you would just run a computer program and be done. The problem is 1) balance sheets can be manipulated in legal ways to support a specific narrative 2) growth is governed by vision + strategy + execution.
For example, Apple the year before the iPhone got launched isn’t an attractive investment. They’re a one hit wonder with the iPod saving them from bankruptcy and the market has been fully saturated. The year the iPhone gets released their balanced sheet hasn’t really changed.
Only google serves its own model - increasing its cloud revenue. The growth chart shows linear increase over time, indicating exponential growth if cloud revenue for google.
Meta at least has a theory that it will transform their ad business. No guarantees but it seems like a pretty decent theory, with direct connections to revenue.
This revenue growth in Search is artificial & extremely unhealthy for Google’s business long term
Search volumes are declining as legacy search is being increasingly cannibalized by non-monetized LLM queries
Google’s response?
Manufacture revenue growth via short-sighted, highly extractive, customer-hostile tactics. I.e. charge advertisers more for lower quality clicks, including clicks they do not want and explicitly did not approve Google to charge them for
A few examples to illustrate:
For all of its history until recently, Google operated on a 2nd price auction model
I.e. if you bid $5 CPC and the next highest bidder bids $1 CPC, Google charged you $1.01 for the click (one penny more than the 2nd highest bidder) rather than the $5 you bid
This was a genius move by Google early on as it incentivizes advertisers to input their true maximum willingness to pay rather than trying to play the game of bidding low and constantly adjusting to try to stay just ahead of the next highest bidder while still not paying too much
However recently, Google silently deprecated the 2nd price auction and began charging advertisers as much as their bid and budget caps allow, regardless of what anyone else is bidding
It’s a short-sighted cash grab at the expense of the long term health of the advertiser ecosystem
Making thing worse, Google also recently nerfed keyword targeting precision
Google previously had precise keyword targeting settings that allowed advertisers pick individual search phrases to bid on, defined down to the character w/ exact match or phrase match targeting
This was one of the core features that made search advertising magic, enabling advertisers to run extremely precise campaigns based on exactly what their target customer typed
But now, even if you bid on a specific term or phrase using the strictest exact
-match targeting settings, Google will show your ad across 1000’s of unrelated keywords, labeling them as as “exact match (close variant)”
The definition of “close variant” means whatever they want it to and changes constantly. The result is advertisers get billed for clicks that are totally irrelevant to their business and that their targeting settings explicitly forbid Google from targeting. Google does it anyway and there’s no ability to turn this off
So now exact match is broad match, and broad match is just meaningless spam
This is all very bad for advertisers, but for Google, it allows them to show your ad and bill you for clicks across 1000x more searches that were previously going unmonetized (mainly because they’re garbage queries no one wants)
This is how you grow revenue atop declining search volumes
Lastly, and perhaps most egregiously, Google quietly stopped respecting budget caps by a factor of 2x. For example campaigns we’ve been running for years with $1000 daily budget caps suddenly began spending $2000+ per day
And the extra spend is entirely on the garbage keywords Google arbitrarily throws in as “exact match (close variants)” which have no value to our business, but can’t be turned off
Google offers no refunds nor any recourse for overspend or spend on keywords you explicitly did not target
These are not the actions of a healthy business. These are the actions of company whose core business is in decline but desperately needs to pump quarterly earnings so Wall Street will continue to fund insane capex while hopefully looking through their rapidly deteriorating negative free cash flow
Google operated a benevolent monopoly for the better part of 25 yrs
Meaning the value Google captured from Search was but a small fraction of the value it created, and that spread produced a potential energy that justified expectations of high earnings growth far, far into the future
This is now no longer the case
At the alter of AI capex, Google is sacrificing the golden goose
> Lastly, and perhaps most egregiously, Google quietly stopped respecting budget caps by a factor of 2x. For example campaigns we’ve been running for years with $1000 daily budget caps suddenly began spending $2000+ per day
When I worked on Google Ads (I left in 2020), I remember this one tripping a lot of people. As I remember it, the limit for a single day is indeed 2x daily budget, but over a month it will average to it. This is supposed to give more flexibility to the auto bidder.
Basically all of Big Tech is betting it all on Red that this whole AI business pays off before they end up losing everything. And I get it, it would be unwise to stay behind and ignore what could very easily turn out to be humanity's greatest invention since pizza. But still, is there seriously no other way to go about it instead of collectively running head first, hands behind at a breakneck pace, while risking the complete collapse of ... well, everything? I suppose not, especially considering it's a technology with potentially massive military and social impact on a global scale, or even beyond that if we're being particularly delusional. Though one has to wonder who will end up paying the tab, and I think that we all know the answer to that.
It could absolutely harm their long term value but keep in mind Alphabet and the other hyperscalers are generally flush with cash. Is this a lot of debt? Absolutely but the businesses are generating a lot of cash too.
Sounds like Google is the only honest company in the market. Cash burn alone doesn’t mean much, it’s free cash flow vs return on invested capital that’s the real signal imo.
Google does have a positional advantage in Android and Apple picking up Gemini as the baked-in AI assistant. It will be interesting to see what kind of impact that has in their services and profitability for AI.
There was an excellent article about AI DC value and depreciation yesterday [1] (discussion [2]). The effective life of GPUs in paticular is a huge unknown. One of my big questions has always been "what will happen to existing GPUs when new GPUs come out?" My guess is that the life of these things isn't as long as the depreciation schedules for some of these companies would have you believe. IIRC Meta was using an 8 year schedule whereas Google is using 4-6, which seems more realistic.
I believe that performance-per-Watt is going to be the only metric that matters. We already have 6 year old hardware (A100) that cannot run the latest models. There will also be new capabilities (eg quantization methods).
I'm not concerned with Alphabet's cash burn rate to be honest. These tech companies are typically shielding themselves from the consequences of this by using Special Purpose Vehicles ("SPVs") where the GPUs themselves are the secured assets for the loans. Even the physical buildings and infrastructure isn't owned by the SPV. Those are rented from another vehicle. So investors are pouring money in to buy GPUs for Google, Amazon, etc. Even SpaceX is partly-insulated by using an xAI SPV.
All of this is I think is a huge risk for OpenAI and Anthropic. The risk to SpaceX is a stock collapse because the AI aspect was always overstated (IMHO).
I think Google will be fine. What is funny is that this is almost using Private Equity type tactics against other investors. Things like the structcures in which the real estate and physical buildings are held in separate entities and the SPVs end up off balance sheet.
Are you asserting that there exists a time period to analyze that is not arbitrary and meaningless? If so, which?
The reason why 2026 specifically is interesting is because it wasn't until late December of last year that AI models started to demonstrate particularly interesting capabilities, while we finally got IPO announcements for OpenAI and Anthropic. Assuming that the market works at all, it should be pricing in these events.
If what they're referring to is coding agents, I can say they were pretty crappy as of November last year, and they've come a long way. Much more useful and usable now.
Amusing how volatile these stocks are. In any case, my intent was not to boost Alphabet but to denounce the overheated state of the tech sector in general, so having the entirety of the Mag 7 fall below the baseline is no skin off my nose.
The current commitment by hyperscalers is around 1.7T USD, reported liabilities 1.3T and this year global debt related to AI is 570B. So that’s around 3T total. For this to make sense AI must generate 2T in new revenue per year by the end of the decade. And that would be only a 10% ROIC. For context ROIC for big tech is around 35% so at 10% they will be barely breaking even. The SP500 gives 10-12%. With 10% ROIC from AI the only thing investors will be celebrating is that the whole thing didn’t trigger a financial crisis. Data centers are NOT real estate. Buildings and power lines usually last 30-50 years. GPUs become obsolete in 5 years. If hyperscalers need to refinance and their interest rate goes up there’s zero margin for error.
H100 is nearing five years and costs more to buy a used one now than a new one when it was released :)
You are completely missing the bet these companies are making.
They think can outlast their competitors and capture a larger portion of the pie while the cost of inference keeps going down dramatically.
If you haven't been paying attention, the cost is about 1/100th of what it was in 2024. This is the trajectory pretty much every technology has followed.
Of course there will be market crashes and corrections and things like that and most companies won't survive, but the bet is that whoever survives ends up doing pretty well.
The fact that supply constrained GPUs holding value is negative indicator. It's like the tulips mania except bubble enthusiasts are buying wilted/dead tulips because live tulips supply constrained due to irrational demand. Now GPUs has more gross utility than tulips but seems like at current revenue/capex spend, every GPU is still negative net financial yield - they lose money - literally buying tulips and watching it wilt. Economically, better off simply not buying and losing more. That's the level of economic irrationality at place sustaining bubble, at least for hyperscaler/big tech balance sheet - small operators logic different and antagonistic to big operator demand/business model.
If cost of inference goes down 100x, would need 100x more demand. This makes overspending on GPU even more irrational. Jevons this, Jevons that but ultimately irrelevant. At end of day, leading players, hungergame winner candidates is saddling themselves with so much debt, even if they survive, post crash they are immediately uncompetitive against new entrant with blank slate and newer gen, more efficient GPUs that will be cheaper to buy/operate post crash when hardware prices will revert to mean.
It doesn't matter if some of the current players survive, they've basically stabbed and weakened themselves so much any healthy upstart in the future can wipe them out unless they lock in legislative protection... safety regulations, ban open source models etc.
That is the new bet, regulatory capture moat, because economic bet is entirely lost, especially with open models eroding mote.
You're making a classic philosophical error here by accidentally anthropomorphizing companies :)
A company losing out on a risky bet and failing, letting a new upstart rise up is pretty natural. A large fraction of experts from the failed companies continue at the new ones, business as usual. There are engineering teams at $BIGTECH now full of OS, database, or compiler experts from XP, Sun, HP etc.
This natural ability of companies to take risky bets is what made silicon valley successful.
Companies are run by people, 100 billion dollar companies are ran by people who want to stay billionaires and will street accordingly. But I'm not sure what we're disagreeing on, yes the talent will migrate, AI industry will eventually settle on some none bubble equilibrium, but that doesn't mean current AI economics is sensible, or inflated hardware costs beyond yield is not danger indicator. Like yeah, risky bets are burning, most people are going to move and carry their technical expertise on instead of unalive themselves or flip burgers, but that's independent of whether business models and broader economy is going to explode.
> that's independent of whether business models and broader economy is going to explode.
It's independent of the former, but not the later. That is my point.
Businesses and business models fail all the time, does not mean the 'broader economy is going to explode'.
It could, sure. But that has been predicted several hundred times and happened only a few times.
You just stated yourself that it costs more now used than when they were new.
If everyone's running local then why are these larger companies dumping cash into data centres?
Economies of scale.
You need a cluster of 8-12 H100s to run the largest models locally.
It doesn't make sense to run these locally yet unless your use case also involves making it available for several dozen concurrent users.
not to miss, future models will be more compute hungry too. Current hardware prices are still goin up and no it's not cheaper to run your AI for like %99 of the people because of lots of costs, it's not just hardware.
I think this fails to take into account how many people are fine with "fast enough" vs "fastest".
I've seen people happily use AI that takes several minutes to generate text or edit an image because to them they already aren't using their computer when they tell it to start; they just grab their phone and walk away and come back only to check in on it.
I feel like people here and on other technology discussions -- although it's worse here -- don't seem to parse what being the minority means.
They know they're one of the few to have access to such incredible hardware -- whether it be rented or purchased for way too much cash -- but they only see their own kin; their own ilk. They only compare themselves to the best.
The reality is that nobody expects data centre speed nor power in their own home and are satisfied to just go "haha its thinking" and let their computer quietly tick in the background as opposed to paying outragious prices for subscriptions or hardware.
That is completely discounting future capabilities and new use cases. Sure in 10 years you will have current SOTA locally, but in no way is it obvious we are anywhere near the limit of marginal value from improved capability
Unless you're a developer or doing complicated research you do not need much to use an LLM at home with consumer-grade hardware.
I'm editing photos with 32GB of DDR4 RAM at 3200mhz alongside a 3050 with a nearly decade-old mobo and ryzen 5800XT and at most? It takes 30 seconds and that's allowing the gpu to use my actual RAM as 'fallback' memory.
There are photoshop filters -now- that take longer than that before the LLM craze even began. Hell; I can train a lora in an hour or two.
This obsession with "more specs more data faster and faster" isn't going to win; it already isn't winning.
Deal with it or get wrecked.
The per token costs plummet with more concurrents. A box that can do 100 tps at request depth 1 might be able to do 3000 tps at request depth 64. Less per thread, but massively more per GPU/joule/etc. That’s the economy of scale of running in a DC rather than locally that they were referring to.
> If everyone's running local
Who's running local? Image generation can make sense to run locally, but frontier LLM make no sense to run on your own hardware.
Google's doing a attempt to answer that (while still firmly hiding who their customers are) here: https://blog.google/innovation-and-ai/technology/research/un...
They promise updates.
>H100 is nearing five years and costs more to buy a used one now than a new one when it was released :)
Because everyone is buying as they want to run their own models and not pay for a cloud service?
Because demand for inference tokens is above supply
Because there's no supply, data centers with these GPUs are running reasonably well.
This is absolutely the calculus. There is no moat. It is survival of the best financed. Open AI and Anthropic are in very precarious situations.
Isn’t that the same bet that famously profitable companies like Uber did in the ride share market?
It's not the exact same bet, but it rhymes.
It worked out for Uber, they are wildly profitable now after spending a decade losing money.
Also worked out as Amazon managed to outlast all the dotcom era e-commerce competitors while being unprofitable.
> It worked out for Uber, they are wildly profitable now after spending a decade losing money.
Uber have a 10% margin, which is definitely not what I'd consider wildly profitable. (Their post tax numbers look better, because of accumulated losses).
Uber is a two sided marketplace and was still famously unprofitable for at the same time.
Over 14 years, Uber burnt ~31 Billion dollars and in nearly whereas the amount invested within AI seems to be within Trillions at this point and in near future with a product which doesn't have much moat and shaky financials on profit.
What costs are 1/100th?
Of serving a (approximately) gpt4 sized model.
What made the cost go down? Can't be cheaper used H100, can't be cheaper RAM. A revolutionary breakthrough in hardware use per query?
Is this true? Hardware costs have only gone up during this time. Are you referring to electricity cost to serve these models? (i.e. compute got more efficient?)
So the number is irrelevant. No one wants yesterdays newspaper.
The only relevant number is the price to serve a frontier or near-frontier model.
Does that include the capital costs of spinning up to the current models/scale or is it just running costs?
Also, lost revenue from other services being degraded by shifting resources to supporting training/serving models (Google Search...)?
> They think can outlast their competitors and capture a larger portion of the pie while the cost of inference keeps going down dramatically.
We are also within an arms race of training newer larger models with more speed while discontinuing older models.
Gemini/Chatgpt have already discontinued their models from 2024 (iirc) because they are using all their compute in serving/training newer models. Being quite frank, nobody is serving a model from 2024 as the intended use-case while having very little moat as open source models are catching up.
> Of course there will be market crashes and corrections and things like that and most companies won't survive, but the bet is that whoever survives ends up doing pretty well.
How so, by raising the prices? because the current prices aren't sustainable and I feel as if there would certainly be companies which will try for one reason or other to be cheaper to capture the market share because of the larger promise of whoever is able to get as market share. I had once thought about it and I don't think that even in an ideal world, they would end up doing pretty well given no moat.
Also even if a company survives and ends up being one of the survivors and makes profit in the ideal scenario you mention, then within some years other companies will try again and construct more datacenters and end up driving the prices down for everyone, so nobody knows how things might look down for 2-3 years let alone a decade, so I remain a bit skeptic currently so.
I had actually thought some on the economics of datacenters and I found it to be very related to power. The only ones which seems to be making money might be the power generators actually because power is the actual bottleneck rather than GPU's in datacenters from my understanding.
Though the power is raised at the cost of electricity bill increases for everybody including people living in houses. The job prospects are minimal as well, as a nation, aside from just getting investment just for the sake of it because AI's trendy right now, I feel like its a net negative deal for people living there.
> The SP500 gives 10-12%
the historical average is closer to 7%. sustained 12% would be excellent growth for any mature firm
In real or nominal dollars?
real
historical did not have free money printer this big
there's always something. technology has come a long way. AI is disruptive, but so were railways, electricity, the transistor, etc...
> Data centers are NOT real estate. Buildings and power lines usually last 30-50 years. GPUs become obsolete in 5 years.
Data centers are real estate. One of the big players in carrier neutral data centers even calls themselves Digitial Realty.
The contents of the DC is not real estate. But neither is the an office or a house or a warehouse.
The “contents” represent the majority of the cost and meaningful functionality of what we call a “datacenter”. Those contents will not last for “real estate” debt timelines.
Right but their payback period is insanely short - a $5M GB200 NVL72 cluster is expected to generate $75M in revenue over 3 years for inference providers. That's a 3 month payback period.
AND - they're operating well past their estimated service lifetime.
Chances they're planning on replacing personal, local compute with time-sharing on data center hardware that's too outmoded for AI...? You know, since they sunk the consumer component market for the next half-decade.
I would be more careful before assuming 5 years depreciation schedule. Currently price tags are attached to computing power, not the production cost. Computing is not getting meaningfully cheaper with newer GPUs but it only allows better scaling, which makes older hardware more relevant for many use cases. This is why A100 is still selling like hotcakes. I don't think this trend will change soon.
Everyone that has invested even a dollar to AI believes the revenue will easily surpass the most optimistic predictions. Ask them.
Wonder if you can pay margin calls with belief.
What's the risk of NOT doing this?
That's the problem. That's the risk that few (if any) hyperscalers want to take.
Apple might be a good counter example of what happens if you don't focus entirely on AI. Right now it seems to be doing ok.
Apple can enjoy because it controls a significant fraction of consumer computing platform so they can simply collect tax from everyone else. This is not true for the rest of big tech. Only Google has Android but it cannot sit and enjoy because they still don't control hardware and AI is an existential problem for their search business.
> GPUs become obsolete in 5 years
The GPUs are far from worthless after 5 years. E.g. the A100 80GB PCIe version cost around $15k when it was introduced in 2021 and now sells for $10k used.
Things might be slightly worse for the data center servers, but I am sure they will find find buyers.
How much of that is due to inflated RAM prices though? I wouldn't assume the current trend is going to continue.
It really depends on how you view the future demand for compute, which really is the crux of the question of whether the capex is rational or not...
They only reason that they are retaining value is there was not so much demand for GPUs in 2021 as there is today. Once the demand drops you will find then in dumpsters across our barren, burning dystopia.
Why are you assuming the demand will drop?
Because AI has a negative ROI for the user.
They hold value as there is insane demand. The same reason a consumer RTX4090 costs more today than bew in 2021. Once the tide drops enough for hardware lead times to shorten to weeks, they will go the way of other used DC hardware - written off after 5 years.
> Once the tide drops enough for hardware lead times to shorten to weeks
Which will not be any time soon according to SK Hynix CEO:
> We still forecast that customer demand will remain higher than our supply capacity even beyond 2030
https://www.reuters.com/world/asia-pacific/sk-hynix-ceo-sees...
For what its worth, SK Hynix CEO has every incentive to show that RAM prices will remain high for as long as possible because that is the only thing which is floating their evaluation to such astronomical amounts.
Independent estimates sort of show around 2027-28 from what I remember.
I remember reading some article which said that RAM prices are already going down from its peak slowly (IIRC I can be wrong, I usually am but 3-5% month from its absolute peak) but the current RAM prices are still astronomical given past rates but the RAM prices will slow down hopefully sooner rather than later.
No one knows. There's a known bullwhip effect in supply chain [0], and DRAM makers are pretty far out along the supply chain. Just like it ramped up wildly it will stop even faster.
[0] https://en.wikipedia.org/wiki/Bullwhip_effect
So you're short the market, right?
Shorting the market always has a greater risk even if you are fairly confident something is true. You also have to be fairly confident of the timing.
> GPUs become obsolete in 5 years.
Not only that, but they're typically amortized over 5 years, where the actual lifespan usually falls far shorter (1-3 years), adding to the artificial subsidy conditions we see today. So they're gaming the lenders into deferring interest payments as much as possible today so that new competitors don't have the same cheap financing advantage.[0]
0: https://blog.citp.princeton.edu/2025/10/15/lifespan-of-ai-ch...
If they're deliberately inflating the likely useful economic life of their assets to get a lower interest rate, it's hard to see how that wouldn't be classed as fraud.
It's the sort of behaviour that really does end up with people going to prison.
You’re all getting some concepts mixed up here. That five year amortization rate is the IRS’ usual amortization rate for computers. GPUs are classed as computers for asset depreciation purposes. But GPUs are part of 168(k) so they’re eligible for a 100% bonus depreciation the year of purchase.
There’s nothing fraudulent at all here just people using terms they really aren’t comfortable with.
That sounds reasonable, it's "just" $1k/yr for 2B workers (there are about 1.2B total "knowledge workers" in the world including gig drivers), or $10k/yr for 200M workers (there are 70M office and technical workers in the US). /s
https://www.dpeaflcio.org/factsheets/the-professional-and-te...
In 4 years it better be 10x more important to have than a cell phone is today, or 10x more important than having internet/monitor/pc/printer is for an office worker today.
It's super-intelligence or bust.
The math looks good on paper, but in reality enterprise AI is hard, most companies are realizing they are actually not seeing ROI from AI. One of my customers took about 8 months to rollout an AI initiative that by the time it launched and people got trained on it, it was already legacy. Also if you are 10x more productive with AI that doesn’t necessarily increases your billable output. There could be some super models like Mythos aimed at very specific hard tasks like drug development, but we have not seen any of that yet, and the clock is ticking.
Yes, I'm agreeing with you. There need to be 200 companies willing to pay $10B/yr for this. What is the ROI? That's the pay roll of ~half the work force of the largest 200 companies. Unless you can fire %50 your employees, everything else is a sunk cost you already own.
I just don’t understand this view. This is the most significant technology ever developed. The uncertainty currently is whether it 1) has massive impact, completely altering society and the making world significantly significantly better or 2) if we go into a fast takeoff/rsi loop. Personally I’ve always been highly skeptical of the later, but that seems like a genuine possibility now. It’s not ‘are we going to be able to generate 10% roic on compute’ the answer to that is yes.
It will obviously be a large part of the economy like online shopping is today. The companies that built up a lot of debt to be brand names in online shopping primarily went bankrupt because new companies had no debt (and perhaps no negative sentiment from early customer experiences.)
the difference is all the current capex is going to durable, hard to get physical assets + things like PPAs. In your online shopping analogy, the hyperscalars are acting like Amazon in 1998
What destroyed hardware manufacturers after the first bubble burst was all the old but overpowered server hardware ending up in resale after bankruptcies. Hard to get hardware is a bad investment that is also potentially obsolete after new optimizations make the next generation much more efficient. The companies that think their individual optimizations are going to outpace industry wide optimization are delusional, historically speaking.
> the hyperscalars are acting like Amazon in 1998
I hope you remember furniture.com, pets.com, webvan.com, kozmo.com and many others.
Amazon.com (during 1998) in some sense is the exception, not the norm from the bloodbath in stock markets during the dot com bubble.
(I highly recommend the book How the internet happened for more knowledge about things before, during and after the dot com bubble.)
> It’s not ‘are we going to be able to generate 10% roic on compute’ the answer to that is yes.
Based on what? No AI company has ever made a cent in profit (exept for Nvidia lmao).
Yeah but space data center Econs are going to revolutionize the tulip marketplace
Looking at cash burn is looking at the wrong end of the horse. Some companies, like Meta, have burned huge piles of cash in pursuit of, for example, the Metaverse and they've got nothing to show for it, not even a slight increment in ad tech, and yet they earned enough to shrug it off.
There's a big difference between Google spending tens of billions on AI infrastructure and what Oracle is doing. Oracle is spending to get on a bandwagon. Google is transforming their business, so far seemingly correctly. If AI flops big-time, Google will be left with some stranded assets, but it won't be existential the way it would be to Oracle.
If AI flops, they’re be left holding large pools of useful datacenter/compute capacity and “revert” to one of the most profitable businesses of all time.
Meta entering cloud at scale would see huge competition and lowering of profit margins.
useful for what? my unemployed neighbours wordpress website?
Welcome to the world of too big to fail and the rich are beyond mortal judgement.
Meta glasses are a direct result of this investment, and they're selling like hotcakes
Smart glasses are about a 15 million unit per year business. Smart watches are about 150 million units per year. Meta owns about half the smart glasses business, but it hasn't provided them a platform they can control that has enough penetration to make a difference.
But wait it gets worse: since Meta hasn't got a platform comparable to a mainstream PC or mobile device OS, the Meta glasses business is vulnerable and subscale, as impressive as owning half the TAM is. Since Meta glasses aren't a companion to an existing platform (not even AndroidAR, even though they run Android) they will become a second tier choice as soon as Apple or Samsung ship a smart glasses product.
Great, let's keep expanding privacy-destroying products and hyper-surveillance in the name of corporate profits!
>and yet they earned enough to shrug it off.
Zuck has 60% voting power, otherwise he would have been fired over metaverse and then model delays
I'm thinking Apple has been really smart in their AI strategy here.
It seems a mistake to make unprecedentedly large capital expenditures, in a very very crowded space, without much evidence of a moat. Presumably people thought the moat would be singularity-like self-improvement of AI, but the singularity is merely a religious concept, and nobody should take religious myth as fact, it's merely narrative for orientation and inspiration.
Their strategy to let Siri stagnate for 15 years and let everyone else take that market? Their strategy to put a bunch of not ready for consumer use AI features on their devices and then roll them back?
They just have such a strong hardware + os ecosystem that they can sit on the sidelines. They'll be able to negotiate with some LLM provider at a good discount when the time is right and put harnesses around it for actual useful features.
> Their strategy to let Siri stagnate for 15 years and let everyone else take that market?
What have they lost by doing this? Did anyone switch from iPhones to something else?
Google's Assistant was and is better than Siri's. How many switched to Android because of it?
I was talking more about massive investments in compute and model development. Voice assistants are barely AI, but real AI might help make better ones in the future. Similarly, I think Apple's failed attempts to add AI to macOS and iOS are a huge justification of avoiding capital investments in compute and AI frontier models.
There's two separate things here: 1) the underlying models, and 2) how they are applied to real world tasks. Both are changing at tremendous speed. The underlying models, 1), can be purchased as a component, so there's little advantage to vertical integration, as long as there's a robust competitive market, and boy is there ever. And Apple was never going to be the big B2B provider for LLM compute and models, such sales are just not in their corporate DNA.
For 2), application of AI in Apple's products, nobody knows how to do that correctly yet, and it's going to a wild few years of people trying things and failing before something comes to pass. Waiting to see where to actually find a competitive edge totally makes sense. Microsoft's Copilot bungling was far worse than Apple AI's bungling, for example. I can't even get access to Microsoft Office anymore, it seems, I'm instead confronted with a dumb chat interface that doesn't even know how to launch a new document or open existing ones. At least Apple never destroyed their brand like that...
I think that's precisely what the previous commenter is saying. Sit on the side lines, and let other people bloody themselves up.
Similar in a way to dot com. It's not to say ML won't have practical application in the future, but the likelihood that it will have specifically this form is low and worth waiting until the dust settles and a more commonly accepted utility presents itself.
If AI/ML were monstrously useful in its current form the companies pushing it would not need to be hawking products; people would be bashing their doors down. I think that's why in areas where it's more directly applied to a known problem set (like Pharma research, and I'm hoping someone with Pharma expertise can pipe up here) there has been more natural pickup.
Coming from trading and markets, ML has been a part of the mix in quantitative strategies for...well, nearly 20 years (by definition I suppose). Spaces with obvious utility will see rapid adoption. Worth waiting that out, honestly.
The point was that one rational approach is that it's OK to not be top ranked in a market that has significant medium-term profitability issues. Apple has no problem creating new markets, but not if being top ranked requires years and years and years without return on investment after launch, and large subsidies. Not to say Apple hasn't failed at creating new markets either, or anything like that.
Claude is a far superior _model_ to whatever Siri uses, but in terms of the Claude app correctly transcribing speech, in my experience it's as bad or worse than Siri (which is already bad). Doesn't matter how good the model is if it misinterprets what you say.
> They just have such a strong hardware
"just" doing a very heavy lifting here
Google and Meta's moat will be their ability to set untold billions on fire. OpenAI and Anthropic can do it for now but once the bubble pops they'll be fucked.
i dont understand the concern. they are putting up great financials. you have to invest ahead of the outcome. this is just classic quarterly public company earnings BS, where public markets dont reward innovation investment. they just want crank the handle financials.
The bigger issue is on the model front, can Google compete; Gemini doesnt seem to be able to compete on the heavy expert end; they are doing well on lighter faster models.
People bought Google for the torrential free cashflow, that looks like its gone forever with this new capital intensive model. If that the case then it needs to be valued like a heavy industrial rather than a capital light tech company.
I agree Gemini's value is not at the frontier, but they are making very useful smaller models. 3.5 flash lite is super fast, cheap and token efficient and accepts any kind of input. For large, price-sensitive processing, in consolidated workflows that don't need the latest improvement, it is the best. Perhaps they should focus on improving their dominance on this business, I don't see gemini reaching claude/gpt in coding. But they could be the ones that make it possible, and economically viable, to roll out llms in large scale automation.
Their bigger positive in my opinion is that they have massive amounts of data and are working to vertically integrate with stuff like TPUs.
I've given up on Gemini. It sounds smart but most of what it tells me ends up being wrong or misleading. I might actually hand $20/mo to OpenAI. It's been far more helpful with the random collection of legal and health problems I've thrown at it. My recent comment history is going to make me come across like a shill for them but, holy crap, GPT has been doing amazing things for me at work as well.
I don't get it either... Google has so much talent yet they just can't seem to get it right.
Are you using Gemini Flash or Gemini Pro?
Are you about to compare the free tier of Gemini to the paid tier of OpenAI?
At work it's paid but in my own personal queries it's free vs free and gemini seems to give erroneous answers with confidence frequently and they will completely flip flop in subsequent queries of the same text. It seems very unstable.
> It's been far more helpful with the random collection of legal and health problems I've thrown at it.
As in convincing?
Or accurate?
Some of both sadly, but where I've checked, accuracy.
> you have to invest ahead of the outcome
> this is just classic quarterly public company earnings BS, where public markets dont reward innovation investment
Genuine question; but aren't these treating stocks as speculative and on vibes? One can say that these comments could be true for the first signs of cracking of dot com bubble. Sure, Web eventually succeeded but many tech giants from dot com era (AOL/Yahoo and so many more) eventually went to dust for spending too much time on the innovative bandwagon.
During the Dot-com bubble really tried to give this example but IIRC there were companies like pets.com who lost 2$ for every 1$ of sale so how a company treats its financials do matter a lot.
The market doesn't seem to reward innovation sometimes because there have been times the first persons to innovative have actually really failed to capitalize on that innovation and many extremely innovative businesses like Airlines (We can literally fly speak of innovation!) have been terrible businesses investment-wise generally speaking.
They just raised $85 billion and they're sitting on a mountain of cash - if their spending didn't increase in this context, it'd be bad management. The real story here is that they have decided to spend that mountain of cash on AI CapEx.
That $85 billion was bonds, and requires ongoing repayments of billions every year in interest payments
And then the $85bn to be repaid too.
I'm referring to the equity offering that started in June and has a second component that starts in 2026Q3. Most of this raise came from the sale of Class A and Class C stock. A fraction came from the sale of convertible stock. To my knowledge, none of this raise came from the sale of bonds.
Also, looks like I got it wrong and they've only raised $45B to date. The rest will come as part of the ATM offering program that begins in Q3.
more details here: https://www.sec.gov/Archives/edgar/data/1652044/000119312526...
sorry, my error. Bonds was $31.5 billion
https://www.reuters.com/business/alphabet-sells-bonds-worth-...
wow, they really are exhausting every avenue to raise as much money as possible.
All these big tech companies are fighting over the basics eventually like power and transformers and don't like to do anything dirty that would hurt their ESG score like getting into any sort of industrial business. Thus, the default is all that stuff that heavily bottlenecks American AI gets done in China.
If you listen to Tesla's recent conference call they are going to making solar panels all the way back to making the silicon ingots and totally vertically integrate. Elon lamented on a previous call that nobody wants to get involved in these primary industries and he has to do it all himself unless he puts his whole supply chain in China. For example, Tesla recently opened a state of the art lithium refinery in Texas cause nobody outside of China does that anymore. He's opening a new fab, because everyone else is too hesitant to expand to meet the capacity he needs.
These alarms have been going off for a long time now. Everyone is already in too deep to admit that there’s a problem.
The alarms in this case are that the profits and margins won’t be as high as we’ve come to expect from cloud companies.
Other than Oracle’s questionable spending spree, these big tech companies are still in very good financial positions. The enormous R&D and infrastructure spends are just feeling unusual to investors who got comparable with the unusually high margins and low costs for SaaS companies. Now they have to put a lot of that money back into the business like more normal companies.
If the margins aren't as high then there will be a repricing for all the massive cloud companies, which means several trillions worth of valuations to be cut from the companies.
AWS/Azure/GCP/Oracle/SpaceX/etc neoclouds... are worth a combined 10+Trillion. That going down by 50-70% is going to be insane.
That would only happen if they need to invest like this forever, otherwise it's just a short-term dent in their margins while they re-calibrate.
This is a good chart that shows historical CAPEX spending. Hyperscalers have been through a couple CAPEX cycles like this, they all know what they are doing.
https://eco3min.fr/en/big-tech-capex-revenue-ratio-quarterly...
Thanks for sharing. Agreed it's a good chart. But I draw a different conclusion. M$ looks pretty iffy: capex/revenue 10% -> 37% in the last 5y, scary trajectory.
Why? GPUs are replaced every 3 to 5 years. This is going to be an ongoing operational cost forever. It will probably increase more if larger models require bigger VRAM sizes.
We have probably hit a limit to scaling LLMs through raw parameter count alone, at least we're not seeing the exponential pace. I personally think we'll end up with a nice sigmoid curve plateauing in the sub 10T parameter regime. The amount of tokens processed (in inference) is increasing exponentially though (I've been following open router usage stats for years and it's always been exponential). We will of course make technological advances in hardware efficiency, and model parameter efficiency, but I think a much more plausible future is that VRAM needed for loading and serving individual models will slow down or even stop. We will need more chips, and more power, as demand continues to grow of course, but the operational lifetime of GPUs today will be a lot longer than the SoTA cards from 5 years ago.
> We have probably hit a limit to scaling LLMs through raw parameter count alone, at least we're not seeing the exponential pace.
Source?
They are building new datacenters for the AI demand, so around half of this CAPEX is not for the GPU-s, and those will not be replaced every 3-5 years.
Also, TPUv2 was introduced in 2018, and still not completely retired in all regions, from accounting pov, it has been written down to 0, but they are still working.
The cost of the land, building, mechanical equipment, etc is a very small fraction of the total cost of a DC.
The upfront cost of facility is ~30%, network infra 10-15%, land/utilities is small percentage, power could be significant for an AI DC. The servers are ~50-60% only.
That was not my experience, but that was pre-Covid
That cost has always been there and allowed for their lucrative margins. It's the upfront cost of building/populating their datacenters (many more than before) that is eating those margins.
I mean the issue is scaling, the worlds for cloud never kept getting bigger and bigger and compute scaling had stopped a while ago in the CPU space.
With AI every new generation with both massive hardware and software stack changes from Nvidia makes prior chips extremely inefficient to run, basically we are comparing an ASIC industry to a general purpose compute industry where all work loads are the same shape and size and so on.
Margins for ASIC based mining companies or ASIC solutions providers were never high, Optane and other weird solutions are niche and great for a specific category or moment in time, but they become obsolete pretty quickly.
The fear is we don't know if this Capex can stop. The worst type of fear is if this Capex will stop then what? Someone is very overpriced in this market, the cloud companies, the hardware providers or both.
I don't see how we reconcile this without a massive wave of repricing, ofc markets can stay irrational and we don't see the actual books but AI doesn't have so much revenue. Suddenly the AI token/cloud revenue won't 100x in a year or two...
Especially when intelligence will continue to get cheaper, the margin compression is a massive risk.
All the data centers for hyper scalers were a miniscule part of their story the real moat was the software layer on top otherwise Hetzner would be priced like an Amazon as well.
Something is shaky with this market I don't know what it's very opaque even as an insider working on for big tech and startups. I have no clue who falls first and which bottleneck cracks but there is not enough revenue for tokens, we will see a strong 2-3x growth in the next few years, from here which is absurd, but it's not enough, not nearly enough. If the capex keeps high and increasing.
Ofc they can stop the capex and the otherside gets repriced it's not like nvidia, micron and co aren't worth trillions.
And then they'll be valued like more normal companies as well. Which will mean a drastic re-rating.
Google's P/E is 25, which normal for "tech", and comparable to S&P overall current, average, which is 50-100% of historical average.
Also all these companies went from buybacks to dilution and debts again.
Cannot hear what you’re saying with all those alarms blaring non stop since a year. Someone should do something about them, maybe turn them off, I don’t know
Sink rate! Sink rate! Pull up! Pull up! Too low; terrain. Too low; terrain. Wind shear! Stall! Stall!
> Everyone is in too deep to now admit that there’s a problem
I'm not sure how to square this with the dramatic improvement in LLM capabilities in the last 8-9 months. If anything, it makes the earlier investments look prescient?
The problem is that the dramatic improvement in capabilities is not translating to a dramatic increase in revenue.
"Anthropic and OpenAI generate a lot of revenue with relatively few employees – an estimated $9M and $5.5M in revenue per employee (RPE), respectively. If either company were to go public, it would have a higher RPE than any public tech company on Forbes’ Global 2000 list." https://epoch.ai/data-insights/revenue-per-employee-ai-compa...
Each one of those employees maps to several fold times more spending on compute.
True, but the comment I was replying to was about revenue, not costs
The guy who sells $20 bills for $10 also generates a lot of revenue
That's a point about margins, not revenue
They've replaced employees with compute, so RPE is irrelevant.
The point is that revenue is large and growing quickly, which is what the comment above mine denied
The revenue needs to be way, way higher than this to warrant the investment.
Maybe, but that's a different claim. You wrote that the improvements are "not translating to a dramatic increase in revenue", but going from about $1B to about $30B run rate in 16 months seems like a pretty dramatic increase to me!
How is RPE relevant if they are spending hundreds of billions on compute and data centers?
It's relevant to the claim I was replying to, which was that revenue isn't growing, not to the question of whether the spending will pay off
So they can add employees endlessly? And still make same revenue? Increasing employees only scale so far at those numbers.
No, I wasn't claiming that revenue scales with headcount, though it probably does to some extent. The point is that these companies' revenue is large and growing quickly, which is what the comment above mine denied.
Isn't that just saying CapEx is a bigger part of their costs as if that is a positive thing?
I wasn't saying anything about costs, only that revenue is growing quickly
This assumes they do not have to increase prices to be profitable, and that they will continue to have customers when customers can switch to open models at similar performance.
As an analogy, Uber could crank up rates after the VC growth play was over to stoke revenue and profits because they have a duopoly with Lyft. LLM consumers can switch to Kimi models fairly trivially today, and whatever the frontier open model landscape looks like later. Model training and development is expensive, self hosted inference on open models not so much.
https://www.wheresyoured.at/the-openai-bubble/ has the math.
(a component of my work is currently building scaffolding so our organization can swap out commercial inference providers for on prem inference infra to derisk against the eventual rug pull when the math gets icky for LLM providers, while consuming as much subsidized tokens as we can until then, when it makes sense to use tokens for work)
Those seem like reasonable questions about future margins and moats, but I was making the narrower point that revenue is in fact growing quickly, contra the comment above mine
The question will be whether customers can switch.
Can you install a near-SOTA model on a cluster in a data center? Of course. Compliance and operations are the sticking points. I work in healthcare IT, and it's amazing how tight the data compliance requirements are. I can't have someone in Canada look at prod data. If we told hospitals that we were handing off PHI/PII to Chinese models, they'd end our relationship due to the long history China has of hacking Western networks and computers. They don't care how open and cheap things are.
Then, you have to keep up-to-date on the latest technology and right-size things in a very fluid market. If you sign a contract for hosting the model on a data center that's running what the SOTA is now in hardware, and someone comes through with a data center hardware or software product that makes that data center contract a disadvantage (maybe it's too expensive and the other party won't budge on the price), you might have to factor that into your offering's price, and that could put you at a disadvantage in your marketplace.
Google, MS, etc. all want to leverage the cloud model to make this be less of an issue for you, for a price. They have the ability to update you with the SOTA stuff in the data centers, because they're the ones driving that SOTA. They can say they host in the US and develop most of their stuff in the US.
Will that be enough of a moat?
Probably not for the levels of spending that are happening now, but over the long term, probably.
How long will a SOTA model be necessary? If day to day work can be achieved on an open weight model, the most evaporates overnight.
Look at any computer in a big company. It isn't the fastest on the market, nor will it have the most RAM or largest monitor or fanciest keyboard. It is good enough at a good enough price point. Once it becomes possible and cheaper to host your own good enough open weight models, with all the benefits of keeping data internal to the company, then the big providers are cooked, so to speak.
> How long will a SOTA model be necessary? If day to day work can be achieved on an open weight model, the most evaporates overnight.
Depends on the advantage it gives people and marketing of that advantage. You'd be surprised at how overpowered the average workplace laptop is. Each company I've been at has had at least some people who do non-technical roles using high-end hardware. Why? Because the account executive wants the fast machine and they get what they want.
You can apply the same to GenAI. Humans are notoriously bad at estimating actual needs when it comes to resource consumption. Best to have it and not need it than need it and not have it, especially if your competition just shelled out for SOTA.
And that's not even taking into consideration regulatory and customer concerns about where the AI you're serving requests with came from.
Execs getting what they want doesn't mean they let everyone have the same thing. It's far more likely that everyone else is using lesser equipment.
We have already seen tech workers at big name companies get whiplash from "leaderboards showing people using the most tokens!" as a good thing one month to being pressured to using fewer tokens a month later.
> Execs getting what they want doesn't mean they let everyone have the same thing. It's far more likely that everyone else is using lesser equipment.
You'd be surprised. I've seen people request upgrades and get them. Not even execs, just employees. Even the average devices are probably overpowered these days. Chromebooks could do the vast majority of work in a corporate environment. Try getting workers to accept them, though.
Regardless, there are probably enough reasons to use proprietary Western AI for the time being, particularly in regulated industries, that poop won't completely hit the fan. Particularly if the CTOs start searching the phrase "Operation Aurora".
> We have already seen tech workers at big name companies get whiplash from "leaderboards showing people using the most tokens!" as a good thing one month to being pressured to using fewer tokens a month later.
Indeed. What do you do when you want a group of people to do something that they'd be otherwise adverse to doing at all? You rank them on it and let them get in a competition over who can do it the most. That's what tokenmaxxing was about: getting people to use the AI at all. Now that the people are using AI, we move onto another objective, which is getting them to use the tokens efficiently. The trick is measuring that efficiency.
My primary role is cybersecurity in a regulated entity in a regulated industry, I am highly confident it is straightforward to do so based on work accomplished in only a couple of weeks. Stand up a router, stand up a Kubernetes cluster if you don't have one, stand up the necessary VMs and compute for serving inference. Two pizza team, in my experience.
Customers can switch (although we can argue the speed and pain of doing so), and the speed at which they do will be a function of cost efficiency and demonstrable value (imho). A recent example of this is Broadcom and VMware [1], for example. When motivated, it can be done. If there is no objective, measured value being delivered, the spend will be cut. If the value delivered is measured, it will be enabled at a lower cost through cost optimization measures (ie self hosting) [2].
This is all to say: there is no moat, the revenue of inference providers is volatile and not assured in any measure. Caveat emptor.
[1] https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
[2] Microsoft considers replacing ChatGPT and Claude with Kimi K3 to save $600M - https://news.ycombinator.com/item?id=49022984 - July 2026
Meanwhile, in real companies, you have to wait 2 months or more to access an API endpoint in preprod.
To setup a cross business kubernetes cluster will take 2 years with unknown results.
On Cloud, in Switzerland, you need to call Microsoft when you need new resources, so much for agility and minute infrastructure provisioning, and I heard the same for AWS.
> Meanwhile, in real companies, you have to wait 2 months or more to access an API endpoint in preprod.
> To setup a cross business kubernetes cluster will take 2 years with unknown results.
Do you seriously believe those times will not go down 95% if the CEO pushes for it to get done yesterday because it will save the company millions in expenses?
You get it. Given sufficient incentives, processes and systems become potentially more malleable, and hard requirements can become optional. Speed is a function of appetite, will, and resources.
> Stand up a router.
it has to be some amazing router and while the models are open-weights, the knowhow to run them efficiently surely is not?
There's a difference between "It's straightforward to do" and "I can convince a customer to sign a contract allowing us to do it."
If the second one were as easy as the first, I wouldn't have to be online at 9:00 to deploy stuff to prod tonight; the team in India would handle it. But customers write into the contracts that only US-based employees interact with prod systems. No amount of cajoling will get them to change their minds; they have data sovereignty, international telecommunications treaties, and HIPAA compliance to worry about. So I'll be pressing buttons tonight.
Could you swap out Anthropic or OpenAI or Google or whoever's models for Kimi? Yes. They're like other software these days, they're modular. What isn't modular is regulatory and geopolitical concern.
Good luck explaining to an exec that the locally hosted Chinese model definitely doesn't have a backdoor or hidden trained-in intentions.
Meanwhile the cost/benefit analysis doesn't move much even if you are paying 2x for tokens, and you don't need anything on prem.
Revenue isn't profit though. Anthropic is already profitable OpenAI financials have looked doomed for the past year
Right, and the comment I replied to was about revenue, not profit. (That said, while I don't think Anthropic is already profitable, it reportedly expects its first operating profit later this year.)
I’ve never seen anything point to Anthropic being profitable.
Ah well we just need to convert our entire economy into an MLM and then I'm sure we'll be set
Lots of people paying for a product they use is more or less the opposite of an MLM
The technology is too hard to capitalize on. It’s far more democratic than, say, an iPhone, or a search engine. Anyone can download a model to their computer and start toying with it, how do you profit off of that? Even if everyone was constantly tokenmaxxing (which we cannot, since the process gets fucked up if you let it run entirely on its own), it probably still wouldn’t be marginally profitable.
Ans so far, the dramatic improvements have come with an increase in API costs.
Even if, hypothetically, Fable or a Fable-class model could seriously replace some headcount, it'll only gain further traction of it's actually cheaper than hiring humans. $50/MTok is expensive. Wouldn't be unreasonable to expect somewhere between ~$3k-$5k/month/developer in spend. Cheaper than a Junior in the HCoL areas (in the US), but not much cheaper in lower-to-average COL areas. Most acceleration will come from having the headcount + giving said headcount $3k-$5k/month in token budget, so now it just becomes a very expensive dev tool rather than a headcount replacement tool.
The idea that a $30k/year API bill will replace 2 $100k developers falls part outside of SFC/NYC. No CFO of a mid-market company in a LCOL area is signing off on $3k/month/dev API bills. They'll just hire juniors and cap their spend at $200/month.
The article notes Google Cloud revenue grew 82% YoY
How much did Google spend to get that increase?
Why do people choose the cloud with a history of randomly deleting billion-dollar accounts?
UniSuper?
(The claim felt so wild I wanted to check, and indeed, the private Google Cloud for the $125bn Australian pension fund was accidentally deleted by a provisioning misconfiguration. Any others?)
IIRC, the files for Toy Story 2 were accidentally deleted during production, and the film was only saved because someone on maternity leave had a backup at home.
Turns out you can fuck up self hosting too.
Yes, Google randomly deleted UniSuper for basically the same reason they randomly ban individual customers: they don't care. Relying on them for anything is a huge mistake.
Source? Has Anthropic's annualized revenue not quadrupled in the last 7 months? And OpenAI's annualized revenue quadrupled since January 2025? Which is only unimpressive by comparison to Anthropic's meteoric revenue growth
I'd be with you if you claimed that the revenue hasn't translated into substantial profits. Being able to spend a lot of money to get less money back is not that impressive. But revenue by itself is on a dramatic rise as capabilities improve
Moreover there’s no guarantee that eventual AI profits (if any) will go to the companies investing all this cash. If the worst case scenario of Chinese labs building and serving frontier-level models on 2nd tier nvidia hardware comes to be then what will be left of all the “hyperscalers”?
Except it did get translated to a dramatic increase in revenue. "Dramatic increase" is a ridiculous understatement here, by the way.
>not translating to a dramatic increase in revenue.
Completely false.
AI and AI related revenues are growing exponentially.
Expenditure on compute is growing even more exponentially.
Exponential growth when you're starting from zero is neither difficult nor sufficient in this case. The title of the linked thread is "Dramatic cash burn." So clearly, the revenue did not grow anywhere fast enough.
Not for the companies using the LLMs…
Are you denying that AI revenues are growing?
Or are you just adding nonsense about "yeah but yeah but no value"?
I’m saying there is no proof that companies _paying for AI_ are seeing a positive effect to their ROI. If you have such a proof, please share, that would be a massive news
...source?
Please try and provide one for such strong claims.
I know it's easy to forget, and InsideOutSanta kind of anchored the conversation on "revenue", but profit is ultimately what matters. Back when Silicon Valley was merely insane rather than bat-guano crazy insane, it was commonly observed that it's not hard to build a business around selling a dollar for 95 cents. The point being that it doesn't necessarily mean much when you have a business doing that, because of course the demand will be insane. It doesn't mean you have a viable business. You don't know you have a viable business until you transition to selling a dollar for $1.03. Many a VC-funded business that looked successful, even wildly successful, has run aground on that transition, or at least, suddenly stopped looking so wildly successful.
If AI-related expenses are also growing exponentially, and they are growing exponentially faster, it doesn't matter that revenue is growing exponentially.
The AI funding has also now absolutely baked in exponential growth of expenses, because that's how debt works. A slow exponential, hopefully, but an exponential none-the-less.
Something Hacker News needs to be periodically reminded of is that we are the field getting the most out of AI, and it's not even close. That's great for us. But the stocks aren't priced for "a pretty nice coding tool". They're priced for every field in the world getting even more value out of this than our field is getting now. That is, frankly, not happening anywhere near fast enough for the spending and stock valuations. When you don't have all the engineering guardrails that are present in software engineering [1], suddenly the AI is, ahem, exponentially less useful.
As I say in that post, watch your AI actually doing something, even the frontier models. Watch the thinking traces. Watch how many times they bang into a guardrail of some sort; a failing test, a failing compile, a linter failure, a bash script that doesn't work, all those things. How much value would you get out of an AI coding assistant if the first time it banged into a guard rail it was done and you had to stop using it for that task? How much value would you get out of an AI coding assistant if instead it silently failed and just proceeded forward with errors that you lack the infrastructure to easily detect? In the first case, it would be fairly modest, almost certainly not worth the money, and in the second, it would be worth paying to not use.
Even in our field, while the rate of code output has increased substantially, the rate of value generation increase has been quite a bit more modest. I have observed, and heard from a number of other places, that while my own output has increased somewhat we still generally can't plan on being able to work with other teams at much faster a rate than we used to.
There's a viable business here but I can't see how all these companies expect to be returning all this revenue in any financially sensible period of time. They're all spending like if only they spend enough they can own about %900 of the market in three years. They can't all do that, even accounting for "AI makes the market bigger".
And they're wildly vulnerable to some new solution coming out that obsoletes all this spending, like an ASIC that starts running a popular model directly (especially if model capabilities plateau), meaning that all this nVidia GPU spending is so much dead silicon. Or someone comes out with a much more efficient way to train models. There has to be some insight we're missing; humans do not learn what they do by having the entire contents of the Internet poured through their head hundreds of times over. We are far more efficient with our training data. What if someone works out a solution to that and we don't need to spend billions on GPUs but only millions? The whole spending proposition could collapse overnight and the companies that suddenly have three orders of magnitude too much hardware and the debt to match would be up a creek without a paddle.
[1]: https://jerf.org/iri/post/2026/programming_is_engineering/
I very much agree with this. Even the top tier models today, without the unit tests, without integration tests, and domain experts reviewing the code would flounder for 50% of the work they do. Sure they can write the unit tests and integration tests themselves, but at that point you aren't in need of a specific system being built, but rather an out of the box solution would probably fit your needs. It does speed up the grunt boilerplate work of development quite a bit, it does help with gnarly bugs and the like, but expertise is still needed. And we as engineers/programmers have systems in place that make using AI easier, we have the human context windows to be able to parse the technical jargon the AI spits out. Will AI for the masses be akin to slightly better automation?
> we are the field getting the most out of AI, and it's not even close.
Just emphasizing that as, due to spending far too much time online the past week, I've been seeing a fair bit of this. "AI is definitely gaining popularity because all the software companies I know are going all in on it."
For certain values of ‘dramatic improvement’. Is lots more important work being done with LLMs? Not much sign of it yet, they’ve been helpful for experts at times (e.g. vuln research or maths research) but that hardly justifies the vast sums for Google investors.
Presumably at some point you need a measurable productivity return yea? Maybe organizations are not built around skill and aptitude so much as liability, which LLMs cannot provide barring (very welcome and also very unlikely) legislation in the US.
The infamous 2025 MIT study that found almost all AI pilots in companies were failing, also found that virtually every worker was using AI many times a week if not daily.
Turns out people just use their personal AI accounts rather than company ones. Which would make sense if you want to claim the work the AI does as your own.
AI was garbage quality or OK but unimportant (Grammarly-esque) in early 2025.
A new study is needed.
>Presumably at some point you need a measurable productivity return yea?
At what point? This technology is brand new. Did you think we were going to double productivity in 3 years?
Capacity is being built. It's hard to build data centres, there are no chips, there is no memory, it's hard to get talent, we don't have the energy to power the facilities.
No one knows where this is going. We are scratching the surface. There is an absolute boom happening, and yet every day I have log onto Hacker News and read this nonsense about everything falling apart. Are we living in the same universe??? So-called "technologists" saying, "meh, it's not that cool". Okay.
Guess what? You're not Michael Burry. Nobody cares or will care that you "called it". Look around this place: you aren't even slightly contrarian.
"We would be profitable if we had the resources but we don't," isn't the smackdown argument you seem to believe it is.
There used to be a thing where successful tech companies were profitable right out of the gate, and very successful companies doubled those profits for years, and companies who bought and used the tech could point to clear, actioned, benefits and cost savings.
Now it's all "This will be really, really profitable one day, probably, if the omens align and we can deal with all of the problems."
No, because the LLMs will keep getting more efficient and capable. Distillation and quantization will mean firms spending trillions on giant data centres are left holding the bag. I suspect Apple ends up laughing all the way to the bank.
https://github.com/microsoft/BitNet
Everyone who initially failed at this stumbled backward into victory.
I suppose there is some limit, but it’s a bit hard to believe that Google won’t find a good use for more data centers.
It's an internet/railroad issue again.
Tech is real, impact is gigantic, long term winners hard to predict, capex spending hard to recoup soon, if ever.
And differently than internet or rails, you don't build once and maintain later, but enter a loop of ever increased spending to keep on top of the arms race and ever exploding usage.
> not sure how to square this with the dramatic improvement in LLM capabilities
A good tech demo doesn’t matter to the business if the products don’t become profitable at the scale the investment chased.
I'm not sure I've seen what I would call dramatic improvement since maybe GPT4?
Sure, things got better. But I'd call it iterative more than revolutionary. I still wouldn't trust any of the models to do anything meaningful unattended. They all still do dumb shit all the time.
Plus, even if they were genuinely dramatically better, the businesses sure as hell aren't. They're burning money left and right, they have no moat, Chinese open models are basically equivalent these days. What's the path to profitability, or hell, break-even? How do you envision this being anything but a giant money pit?
> I'm not sure I've seen what I would call dramatic improvement since maybe GPT4?
LLM conversations online are so weird. Whenever I read things like this it’s like I’m living in a different world than the other person.
GPT4 was almost useless compared to what we have available today.
I mostly use anthropic models, but there was a big step function when claude code came out, and it’s been incremental or a plateau since then.
Opus 4.6 and 4.8 are basically indistinguishable from Fable and Sonnet 5. 4.7 was a hot mess. The guardrails on 4.8 and 5.0 make them worse than 4.6 for many tasks. So, even if Fable is theoretically better, refusals/downgrades make it a worse product in practice. Who cares if it outperforms on 1-2% of real world tasks if 5-10% of tasks are blocked?
I’d bet most people could be downgraded to a 12 month old frontier model, and not notice for a week or so.
Anthropic’s big problem is that open weight models are 0-6 months behind. So, their product is commoditized and margins are never going to be good.
> I’d bet most people could be downgraded to a 12 month old frontier model, and not notice for a week or so.
This is another unbelievable claim. I actually used frontier models from 12 months ago and they were completely different.
It sure is funny how everyone claims the current model is a "dramatic improvement" over the models from X months ago.
You'd think if there had been that many dramatic improvements I'd have to babysit an LLM less frequently.
Too big to fail now, so everything is fine.
They've literally rated the debt as too big to fail in order to get foreign sovereign wealth funds (mostly gulf states) to agree to put up the money for loans. This has been happening this entire time.
Wow. So plain. No "AAA" euphemisms this time round eh!
Sergey Brin said he would rather Google go bankrupt instead of losing the AI race. That is where the bar was set.
I see eventuality here as job cuts or salary cuts.
Don't think that day is far when "software people" are paid as if they were taxi drivers.
You're probably right but honestly why would anyone stay in software if that happens?
My company is remaking its career ladder to emphasize agentic coding just in time for this.
What would be the best thing to do with ones investments considering these alarms?
Say you had some money in cash rn, what should one do? Wait for a crash and buy stuff up cheap? Put it in some safe category?
This stuff is stressing me out and I do believe it's gonna come crashing down sooner or later, but I don't know enough about investments to know how to best come out unscathed.
Diversify! Historically, the average length of a recession has been 12-24 months. So set up a system whereby you won’t screw’s yourself over by selling when things are low, but instead you can weather the storm.
Build a rainy day fund. Determine how much cash you will need if you are out of a job and how long you think that will last, allocate some portion of that amount into low risk bonds. Russ way if you need cash you aren’t selling investments at a big loss.
If you have enough liquidity put some in real estate as a forced savings vehicle as it’s harder to liquidate than stocks. Then just sit out any coming storm.
But diversify into what?
If we assume this takes down the US economy and bonds, what then? International bonds/stocks? Won't those also be too entangled? Precious metals?
If there's a big AI bust, there will be no escaping it. Like 2008, the entire economy will slow down. This time it might even end in a war. A well diversified portfolio e.g. index funds will weather the storm and recover.
Build your emergency fund first if you don't have one. 6-12 months of salary in cash or CDs. Then dollar-cost average into well diversified equities. Don't watch them day-to-day. You're concerned about their value in 20-30 years, not tomorrow.
Nobody can tell you what to diversify into without information about what you are concentrated in.
Sure, spread investments across stocks and bonds and treasuries from different markets.
But you can also diversify more broadly beyond economic capital to cultural and social capital. Learn new skills and build networks of generalized reciprocity with others before you (or they) need help.
> without information about what you are concentrated in.
They said cash.
Those gold guys have been decrying the collapse the US economy for 25 years now, so you'll be in good company.
The top will be when Jim Cramer loudly proclaims there is no problem at Oracle and gives a buy rating.
https://www.youtube.com/watch?v=gUkbdjetlY8
What problem? What alarms?
I see everyone around me doing way more work, of way more depth, than they ever did before using AI models. I see my company and friends of mine all paying large sums of money to Anthropic, Google, OpenAI to use AI models, and do more work than we did before.
So Google is investing in infrastructure which is HIGHLY in demand, there is much more demand than supply, and then they are making money from this infrastructure...
That's a good thing for Google, and as an investor in Google, I am glad they are making these investments.
Haven't they announced the spending like, years ago? Is the market deaf and blind now too?
They raised their forecast a bit:
> The search giant now expects to spend between $195 billion and $205 billion in capital expenditures, its finance chief Anat Ashkenazi said on a conference call with analysts. The company said last quarter that it planned to spend between $180 billion and $190 billion this year.
Why does it raise alarm? Pretty sure all this spending was planned.
I'm pretty sure they didn't plan to just spend cash without any return. It raises an alarm because there is no end in sight for the money burning
> they didn't plan to just spend cash without any return.
No return? Annual earnings have kept increasing at 20-40% for the last 4 years.
Plus there's this:
https://www.theregister.com/paas-and-iaas/2026/07/22/google-...
> Google Cloud is killing it
> It's Alphabet's fastest-growing business and now makes up more than a fifth of the juggernaut's revenue and operating profit
There is pretty clear return as of now. And half a trillion in backlog
Also the ~4% drop is really not a big swing for earnings. This looks like a non story
Since when is investing in infrastructure burning money?
If there is a huge demand for shipping goods internationally, investing in ships and planes isn't burning money.
There is massive demand for compute in the world right now, Google is investing in that area. That's a good thing.
There's enough hype and exuberance in the AI market that it's likely some players are going to be left holding the bag with a write-down on assets.
Serious investors look at balance sheets, less then what CEOs say. Elon Musk -- as an example-- says all kinds of things that don't really happen. Mark Zuckerberg is arguably less grandiose. When FB changed their name to Meta, said they were committed to the metaverse the stock didn't dump. When the really big investments in consumer VR hit Meta's balance sheet, there was a big drop.
Think of it as the difference between the waiter describing dishes with ingredients you don't really understand (or maybe even taste) vs presenting the bill for the meal.
If a company’s value was completely representated within their balance sheet, you would just run a computer program and be done. The problem is 1) balance sheets can be manipulated in legal ways to support a specific narrative 2) growth is governed by vision + strategy + execution.
For example, Apple the year before the iPhone got launched isn’t an attractive investment. They’re a one hit wonder with the iPod saving them from bankruptcy and the market has been fully saturated. The year the iPhone gets released their balanced sheet hasn’t really changed.
thats how i justify my vacation spending
Only google serves its own model - increasing its cloud revenue. The growth chart shows linear increase over time, indicating exponential growth if cloud revenue for google.
Meta, Microsoft, Amazon also serve their own models, though these models are not frontier models.
meta does too?
Meta at least has a theory that it will transform their ad business. No guarantees but it seems like a pretty decent theory, with direct connections to revenue.
I've been seeing quite a few companies juicing short term margins and quarter to quarter maxxing even more than before, one such example:
https://x.com/MaxAnderson/status/2080229375773941871 https://xcancel.com/MaxAnderson/status/2080229375773941871 --- As someone who has personally spent $500k / mo+ on Google Ads for years, I can tell you with certainty:
This revenue growth in Search is artificial & extremely unhealthy for Google’s business long term
Search volumes are declining as legacy search is being increasingly cannibalized by non-monetized LLM queries
Google’s response?
Manufacture revenue growth via short-sighted, highly extractive, customer-hostile tactics. I.e. charge advertisers more for lower quality clicks, including clicks they do not want and explicitly did not approve Google to charge them for
A few examples to illustrate:
For all of its history until recently, Google operated on a 2nd price auction model
I.e. if you bid $5 CPC and the next highest bidder bids $1 CPC, Google charged you $1.01 for the click (one penny more than the 2nd highest bidder) rather than the $5 you bid
This was a genius move by Google early on as it incentivizes advertisers to input their true maximum willingness to pay rather than trying to play the game of bidding low and constantly adjusting to try to stay just ahead of the next highest bidder while still not paying too much
However recently, Google silently deprecated the 2nd price auction and began charging advertisers as much as their bid and budget caps allow, regardless of what anyone else is bidding
It’s a short-sighted cash grab at the expense of the long term health of the advertiser ecosystem
Making thing worse, Google also recently nerfed keyword targeting precision
Google previously had precise keyword targeting settings that allowed advertisers pick individual search phrases to bid on, defined down to the character w/ exact match or phrase match targeting
This was one of the core features that made search advertising magic, enabling advertisers to run extremely precise campaigns based on exactly what their target customer typed
But now, even if you bid on a specific term or phrase using the strictest exact -match targeting settings, Google will show your ad across 1000’s of unrelated keywords, labeling them as as “exact match (close variant)”
The definition of “close variant” means whatever they want it to and changes constantly. The result is advertisers get billed for clicks that are totally irrelevant to their business and that their targeting settings explicitly forbid Google from targeting. Google does it anyway and there’s no ability to turn this off
So now exact match is broad match, and broad match is just meaningless spam
This is all very bad for advertisers, but for Google, it allows them to show your ad and bill you for clicks across 1000x more searches that were previously going unmonetized (mainly because they’re garbage queries no one wants)
This is how you grow revenue atop declining search volumes
Lastly, and perhaps most egregiously, Google quietly stopped respecting budget caps by a factor of 2x. For example campaigns we’ve been running for years with $1000 daily budget caps suddenly began spending $2000+ per day
And the extra spend is entirely on the garbage keywords Google arbitrarily throws in as “exact match (close variants)” which have no value to our business, but can’t be turned off
Google offers no refunds nor any recourse for overspend or spend on keywords you explicitly did not target
These are not the actions of a healthy business. These are the actions of company whose core business is in decline but desperately needs to pump quarterly earnings so Wall Street will continue to fund insane capex while hopefully looking through their rapidly deteriorating negative free cash flow
Google operated a benevolent monopoly for the better part of 25 yrs
Meaning the value Google captured from Search was but a small fraction of the value it created, and that spread produced a potential energy that justified expectations of high earnings growth far, far into the future
This is now no longer the case
At the alter of AI capex, Google is sacrificing the golden goose
> Lastly, and perhaps most egregiously, Google quietly stopped respecting budget caps by a factor of 2x. For example campaigns we’ve been running for years with $1000 daily budget caps suddenly began spending $2000+ per day
When I worked on Google Ads (I left in 2020), I remember this one tripping a lot of people. As I remember it, the limit for a single day is indeed 2x daily budget, but over a month it will average to it. This is supposed to give more flexibility to the auto bidder.
Thanks for this informative post. Many have been puzzled as to why Google keeps claiming search isn't affected by chat apps, when clearly it is.
Curious how you are responding to this? Are there viable alternatives you are moving budget to or are you just hostage to their new tactics?
> “exact match (close variant)”
I have to laugh to keep from crying.
Basically all of Big Tech is betting it all on Red that this whole AI business pays off before they end up losing everything. And I get it, it would be unwise to stay behind and ignore what could very easily turn out to be humanity's greatest invention since pizza. But still, is there seriously no other way to go about it instead of collectively running head first, hands behind at a breakneck pace, while risking the complete collapse of ... well, everything? I suppose not, especially considering it's a technology with potentially massive military and social impact on a global scale, or even beyond that if we're being particularly delusional. Though one has to wonder who will end up paying the tab, and I think that we all know the answer to that.
It could absolutely harm their long term value but keep in mind Alphabet and the other hyperscalers are generally flush with cash. Is this a lot of debt? Absolutely but the businesses are generating a lot of cash too.
Are you clicking on more ads now or less. Are you using google search more now or less.
I'm using it a lot less.
Don't think Google can point to past revenue an indicator of future revenue, they need to establish new streams of revenue.
Have you checked the numbers? I thought for most of the ad driven companies the numbers are up.
Sounds like Google is the only honest company in the market. Cash burn alone doesn’t mean much, it’s free cash flow vs return on invested capital that’s the real signal imo.
Google does have a positional advantage in Android and Apple picking up Gemini as the baked-in AI assistant. It will be interesting to see what kind of impact that has in their services and profitability for AI.
How does this spend affect Google CEO's $692 Million potential pay? Is it meeting the required goals or taking him away from them?
https://fortune.com/2026/03/10/google-ceo-sundar-pichai-692-...
There are so many news about this topics nowadays, that I think they are going to blow the horn of the final attack shortly after.
Profit is up 20% YoY. Google is a money printing machine, and they printed over $40B last quarter. Are you kidding me.
Non-paywalled MSN link: https://www.msn.com/en-us/money/general/alphabet-s-cash-burn...
(still somewhat outraged Reuters has a paywall now. Also BBC, CNN...)
wow
There was an excellent article about AI DC value and depreciation yesterday [1] (discussion [2]). The effective life of GPUs in paticular is a huge unknown. One of my big questions has always been "what will happen to existing GPUs when new GPUs come out?" My guess is that the life of these things isn't as long as the depreciation schedules for some of these companies would have you believe. IIRC Meta was using an 8 year schedule whereas Google is using 4-6, which seems more realistic.
I believe that performance-per-Watt is going to be the only metric that matters. We already have 6 year old hardware (A100) that cannot run the latest models. There will also be new capabilities (eg quantization methods).
I'm not concerned with Alphabet's cash burn rate to be honest. These tech companies are typically shielding themselves from the consequences of this by using Special Purpose Vehicles ("SPVs") where the GPUs themselves are the secured assets for the loans. Even the physical buildings and infrastructure isn't owned by the SPV. Those are rented from another vehicle. So investors are pouring money in to buy GPUs for Google, Amazon, etc. Even SpaceX is partly-insulated by using an xAI SPV.
All of this is I think is a huge risk for OpenAI and Anthropic. The risk to SpaceX is a stock collapse because the AI aspect was always overstated (IMHO).
I think Google will be fine. What is funny is that this is almost using Private Equity type tactics against other investors. Things like the structcures in which the real estate and physical buildings are held in separate entities and the SPVs end up off balance sheet.
[1]: https://ciphertalk.substack.com/p/nobody-knows-what-a-used-g...
[2]: https://news.ycombinator.com/item?id=48917135
Some more discussion on source: https://news.ycombinator.com/item?id=49012630
please let me buy some ram and storage
So tired of media doomposting and exaggerating everything.
It's not all roses out there, to be sure. But there are some roses, I think. A lot of roses actually. Can we talk about the roses once in a while?
GOTTA BUY THOSE TULIPS!!
Keeping in mind that Alphabet is the only one of the Mag 7 stocks that has managed to outperform the S&P 500 in 2026.
Short term stock price is a popularity machine, not an indicator of value.
Apple stock is up 18% in 2026
Keeping in mind that Jan 1 2026 to Jul 22 2026 is an arbitrary and meaningless time period to analyze.
Are you asserting that there exists a time period to analyze that is not arbitrary and meaningless? If so, which?
The reason why 2026 specifically is interesting is because it wasn't until late December of last year that AI models started to demonstrate particularly interesting capabilities, while we finally got IPO announcements for OpenAI and Anthropic. Assuming that the market works at all, it should be pricing in these events.
> wasn't until late December of last year that AI models started to demonstrate particularly interesting capabilities
What are you referring to here?
If what they're referring to is coding agents, I can say they were pretty crappy as of November last year, and they've come a long way. Much more useful and usable now.
No they haven't. SPY YTD: 9.40%, GOOG YTD: 3.33%
They have (massively) outperformed it in 2025 though.
I'm seeing 8.43% for GOOG.
Not sure where you got 3.33%, looks to me like GOOGL is +9.44% YTD while GOOG is +9.1%.
Might be related to the massive drop this morning. According to yahoo finance, YTD GOOG is +1.81% and GOOGL +2.33%
Amusing how volatile these stocks are. In any case, my intent was not to boost Alphabet but to denounce the overheated state of the tech sector in general, so having the entirety of the Mag 7 fall below the baseline is no skin off my nose.
Is this why Google decided to release their article explaining how AI spend makes sense to the plebes?
I missed that, link please?
sure, here you go https://news.ycombinator.com/item?id=49021006
Oh no a company spending money is bad for the economy… especially since they are spending it on … the most advanced tech humanity has ever created…
Photonic computing?