"Inference" is just software running. It has always cost money to run software, it's just that it is generally too cheap to matter. If a client makes a regular API call to your server, you pay for that compute, probably in the form of a flat hosting fee. If too many calls come in and workload goes up, you pay for a more expensive hosting tier to handle it (or do dynamic scaling which is per-unit of compute).
Right now the "hosting" cost for inference is per-unit because it's new and expensive, but that won't last.
There is a lot of inefficiency right now keeping prices elevated. That will change very fast and soon paying for inference will likely resemble paying for hosting your app.
The bigger problem for SaaS is that the floor has risen - people can build their own solutions for things that they used to buy SaaS for. So the industry needs to level up and solve harder problems.
> There is a lot of inefficiency right now keeping prices elevated. That will change very fast and soon paying for inference will likely resemble paying for hosting your app.
I don't know about "very fast" or "soon" unless you're speaking in geological terms.
SOTA models like Kimi 3 require thousands of GB of RAM/VRAM to run at speeds that are real-time useful. Manufacturing the memory necessary for that quantity to be available at app-hosting prices will take decades. Software efficiency solutions might drop needed memory by an order of magnitude in that time...but a tenth of an enormous amount is still pretty darn big so won't get us there "soon".
They claim to support 24/7 agent coding with no VC subsidizations or money lost on a subscription, instead relying on optimizations on agent selection. I think it'll shift sooner rather than later.
Exactly my point in another comment. Just to illustrate this further: a rough ballpark of how the cost of intelligence fell since 2022 could be about 1000x, and continues to fall. Unfortunately, it’s really hard to measure.
It’s so cheap that companies choose to spend more on AI inference (more reasoning, more capabilities, longer context), not less - see Jevons paradox.
> That will change very fast and soon paying for inference will likely resemble paying for hosting your app.
The article have talked about this:
> You cannot, however, build a business model today on a cost breakthrough that has not happened yet. Even if costs do fall, Jevons paradox[5] tends to kick in: historically, when a resource gets cheaper, we find more uses for it rather than consuming less. Cheaper inference does not get banked as margin improvement; it gets consumed by deeper integration, more calls per interaction, more autonomous agents running in the background. The cost per call drops but the calls per user multiply.
I don't think it's at all certain this won't land back on the same unit economics as the old way. The cost of serving a user doesn't have to be free - it never has been - it just has to not be the dominating factor in your costs. I'm guessing there are still quite a lot of per-user costs that aren't easily visible. Like how many of your users are logging support requests, or suing you, or demanding bug fixes or custom integrations or a myriad of other things. And how much are you having to invest in security updates, regulatory compliance, marketing etc. Not to mention, users are getting well acclimatised to the idea of quotas and paying for increased limits.
I think part of the new equation may also become; "Why even pay for the SaaS in the first place if you can just forge the service exactly how you want it?" The benefits of unlimited access to the tool you forge are still there, its just a lot easier to make whatever tool you want.
Are there any examples of products containing ai inference that are successful? Products that are beyond just direct access to frontier LLM's, I mean.
As if making all the decisions that go into that product, hosting it, and maintaining it are free! Now do that for all the SaaS you subscribe to. Would you even get any real work done?
Hosting is cheap if its abstracted away. Same with the technical aspect of maintainance
About design and the product side of maintainance, what i have felt is that, deep down inside, the user does know exactly what they need. If they can learn to communicate that in a way another person or agent can understand, then design and maintainance will be fine. Thats a big if, but if the economics align for people to benefit from developing that skill, maybe they do.
There wouldn't be a product. Consumers would just use the underlying foundational models directly to solve their problem. ChatGPT and Claude and Gemini are the products, not your business.
Products are only viable when you do a lot of work to solve a problem, which is then shared by many potential customers. It makes sense to amortize the high costs of solving the problem across all the different customers to reap economies of scale. Businesses then pay for product design, hosting, and maintenance because those costs can also be amortized, and they are cheaper than a bespoke solution for each customer. But if the bespoke solution becomes cheaper than that, because it's generated by an LLM that doesn't need to be paid a living wage, there's no reason to have the product in the first place. Just solve your damn problem and let other people solve theirs.
This is an underrated factor in the market structure of the AI bubble going on now. Anecdotally, we're not seeing new AI-based products other than the foundational models and coding assistants gain traction. Why? Because AI makes it so easy to customize the solution to your particular needs that everybody is just solving their particular needs directly. It's the opposite of the Internet boom, where the network greatly expanded the potential market, reduced the cost of reaching them, and made it economical to spend large amounts of money building a software product that had a TAM of billions. The AI boom instead enables extremely cheap customization, which shrinks the market to a single customer who uses AI to directly solve their problem rather than building a product that's generally applicable.
> Consumers would just use the underlying foundational models directly to solve their problem
Whether this could happen is a critical question.
What ive found currently is that its not possible because the non-determinism of the model means you cant trust it, and these tasks are such that people expect determinism. Even if they dont then they dont tolerate the kind of stupid mistakes these models make
What could happen is that the model becomes good enough to be near deterministic, or as deterministic as humans. I dont know if the current next token prediction foundation is sufficient for that. But who knows, maybe it is
If you want determinism, you ask Claude Code to write you a computer program to the desired specs, and then run that program over and over again, instead of renting the software from a SaaS vendor. I've done that a bunch - I'll show Claude the inputs, I'll tell it what output format I want, tell it the tech stack, and then have it write the program. Then I'll ask for a systemd service or crontab entry to run it repeatedly.
For a lot of tasks, classical deterministic computer programming is better than an LLM. It's just that LLMs are very good at classical deterministic computer programming now.
> Consumers would just use the underlying foundational models directly to solve their problem. ChatGPT and Claude and Gemini are the products, not your business.
How am I going to use ChatGPT instead of, say, Figma or an office suite with my coworkers?? It's no use for it to spin up a clone for me if my coworkers can't collaborate. And if they can, what it will serve is a product.
How can it be cheaper for every company to re-invent the wheel?
> Products are only viable when you do a lot of work to solve a problem, which is then shared by many potential customers.
Actual example: my wife wanted a logo for the facepainting business she's starting with our 5yo, so she asked Gemini "I want a logo with these objects in it, drawn in these colors, with this business name." And it spits back a perfectly passable logo in 5 seconds. Done. No designer needed. No Figma needed. It cut out the entire value chain.
Is it as good as a professional could've done with Photoshop? No. But it's about 90-95% of the way there, which is good enough that neither her nor any customers would care.
A great deal of business is precisely this, tasks that need to be done but where you only care about "good enough" solutions. After all, a key principle of business is you don't outsource your competitive advantages. You pay people for the table stakes, the things that everybody needs but that they only need to be "good enough". LLMs can generate "good enough" facsimiles of a wide variety of fields.
Your wife would never have used Figma for that before LLMs anyway. What is she using to manage the accounts; the bread and butter of her business? Did she conjure up her own accounting and CRM software? If she vibe coded it what did she do to ensure there are no bugs? And in any case, she works alone, I presume? I'm talking about normal companies where people work together; the typical case.
My wife wouldn't have considered this a business before LLMs. She's mostly just doing friends' parties, and oftentimes getting paid in-kind (i.e. instead of bringing a gift, she just facepaints all the neighborhood kids).
Remember that in the mid-00s, this is what YouTube and the social web was. Folks doing little, productive economic activity, usually unpaid, which can now reach an audience through technology. And now the top career aspiration in young kids is "Be a social media influencer." That's how disruptive innovations start: with little transactions that aren't considered business at all, which then grow until they subsume the economy that everybody knows of.
There are already free, open source clones of numerous things but companies still pay for commercial versions. Do you think LLMs that you have to pay for are going to change that?
If my manager prefers to pay for SaaS while I sit on my ass rather than have me develop an in-house solution, how do LLMs drastically change the equation?
>Users began expecting something fundamentally different from software: not just tools that store and retrieve, but products that reason, generate, and respond.
Not really.
>Every inference call costs money.
Not really, either. If you buy your own GPU, rack it, and run an open model, there is no unit cost. This is just expensive hosting infra. You also pay unit costs for SaaS that your software uses (things like SMS etc).
> If you buy your own GPU, rack it, and run an open model, there is no unit cost.
No. There is economic opportunity cost (borrowing), energy cost, infra cost, depreciation / risk of failure with each unit of work, bandwidth, maintenance, and lots more. Small, but not zero, and often overlooked - especially the opportunity cost.
These are basically all fixed costs, not unit costs. You buy the GPU once and use it for as many calls as you have traffic for, and depreciate it over a fixed lifetime. You have to power it regardless of whether it's fully utilized or not. You have to maintain it by virtue of owning it, not really based on how many queries it has served. Bandwidth is the only one that really scales as a unit cost.
Open question whether this model is actually more economical than using the cloud AI service. The whole reason the industry moved to cloud computing in the first place was because computing had very high fixed costs, and the more these could be amortized over a fully-loaded query stream, the lower the unit costs.
You have to power it all the time, but the amount of power it uses while it’s on will change by up to a few orders of magnitude depending on the gpu. It’s not uncommon for a gpu to be pulling just a couple of watts at idles and several hundred at full tilt.
So the only way it’s a fixed cost is if you don’t pay for power. If you only consider the cost of the power, it might still be cheaper paying for an api.
Good points. Speaking from experience, it’s really hard to make it more economical than using a cloud AI service - even if you utilize the GPU to its fullest. There are, of course, other benefits, such as privacy/control/compliance/security, which should be the real reasons to do this, not cost.
Spoken like a guy who has never had to maintain bare metal infrastructure ops at scale. These things break, need re-imaged, have parts that break, have to be configured (now you need provisioning pipelines, monitoring, alarms, etc), have to be maintained when something goes wrong (swapping hardware and software, debugging the alarms into actually figuring out which bits are broken and/or misconfigured), have to be catalogued, have to be planned for, deprecated, and the finances accounted for through complicated accounting to show investors the Capex at quarterly meetings.
Then you have to make fleet decisions on how much of each type of server you will want to buy, expanded storage, how long you will support each generation of server, when you will order new hardware, how to order new hardware, lag for real world installs, hiring actual humans to fly around to all of your datacenters and do the actual installations / maintainance / etc. Then you will have to do contracts with individual datacenter operators for margins, electricity rates, hosting contracts, white glove ops hourly rates, etc.
Businesses that own their own hardware tend to have a lot of employees whose jobs are maintaining it and running the business side of that.
How is this different from hosting infra or just SaaS in general?
TFA contends that there is some fundamental shift in the economics of software, but it doesn't look to be very different from either a new SaaS dependency or racking new hardware.
> Build the product once, distribute it to a million users for roughly the same cost as distributing it to one. Every incremental customer flowed largely to the bottom line.
common misconception about software unit economics. With enterprise software (one that costs real $$$) there always more costs attached post shipping. Before client/server it was support. Then it was security and the constant threat of cyber attacks.
Distributing software is nothing like distributing books.
"Users began expecting something fundamentally different from software: not just tools that store and retrieve, but products that reason, generate, and respond."
Absolutely not. Customers want systems for sales, reservations, accounting, and taking stock. That's where almost all the SaaS money is and none of it benefits from AI - and never will.
This is one reason I've been trying to figure out tasks (and products based on those tasks) that can be pushed to the edge, either via small specialized models or small general purpose open models. I suspect the same desire to keep unit costs low is part of why Google is falling behind on the "frontier", but seemingly at the lead, or near it, on models that run on-device. I think they're just focused on making models for tasks that don't require boiling the ocean.
But, it's a hard problem. The models that run locally on normal computers/phones are pretty terrible compared to the frontier, without specialization and fine-tuning. And, even with specialization and fine-tuning, often a high-end general purpose model is going to do a better job and people don't need a bunch of local tools installed to do their various tasks.
> And, even with specialization and fine-tuning, often a high-end general purpose model is going to do a better job and people don't need a bunch of local tools installed to do their various tasks.
I think this is the critical point that would be interesting to see if it holds. Technology seemingly tends towards increased specialization.
The bitter lesson is rather specific and IMO vastly over-applied. It basically says that you can beat hand-tuning by waiting a few years and using "more of the same" generic method.
It's not all that different from saying "Don't optimize software, just wait for faster hardware". Yet highly optimized software exist, and performance on currently available stacks is a competitive advantage.
What gives? I say: Staying ahead of the "Don't optimize, just wait" curve can absolutely make sense. At worst, your advantage decays after a few years. At best, you stay ahead by n number of years and keep increasing the gap as you invest more.
If you make a specialist model and harness that takes six months to train and launch, will you have enough time to recoup your costs and make a profit before the generalists get good enough and cheap enough in your area of specialty and become the simpler option?
We already see "skills" repos containing a bunch of specialist stuff being pushed to general models via prose instructions for how to do various things and use various tools; and those models will do the tasks much more expensively and slowly, but without needing the user to find/buy/learn something specific to do the job.
The bitter lesson may be narrow, but the fact is people have been using spreadsheets for everything for 50 years, and only sometimes can a specialist tool effectively displace it in an organization. Very powerful generalist models are starting to feel like spreadsheets: The default way to solve problems in organizations. You might be able to pick off some of the specialist cases, but you might also miss.
I think the bitter lesson (from 2019, what is now seen as the tail end of an AI winter - probably wasn't at the time) made the most sense in academia during AI winter. There is no academic fame to be had by making something a little better for a limited time. When billions are at stake, there is potentially a lot of money (and utility for customers) to be gained by doing some extra work even if the work is thrown away soon after. Most commercial products contain a ton of extra work to be slightly better than the competition in one area or another.
Perhaps we can call this "The sweet dollars lesson".
The author misses one potential future of AI software where people use use their own subscriptions / API keys for their AI usage.
In this scenario the user would either sign into the platform via their Anthropic / OpenAI / Gemini account or use their API keys, and any of their usage would be billed to them directly.
The company then doesn't have to worry about the increasing costs from the AI usage.
Of course, in this future, AI providers become the new "Facebooks" of the world.
There's an interesting bellwether in mobile games right now. There are two easy statistics to track when trying random free-to-start games: time to first ad (TTFA), time to paywall (TTP).
Most LLM-using mobile games are already at incredibly weak TTP scores and perhaps the only current stratum of mobile games where TTP is almost always ahead of TTFA. TTFA before TTP is almost nonexistent because they claim to need a monthly subscription as soon as immediately after install, despite being advertised as free to start. It's also one of the few types of games where the paywall explicitly does not include "no ads". Some of these games running on monthly subscriptions still need ads for unit costs.
(The bulk of mobile games try for a sweet spot of TTFA in the order of hours of gameplay and TTP in the order of days of gameplay. Easier to get people hooked on your game if you can give them a few hours of uninterrupted fun up front.)
It seems pretty condemning of software economics with LLMs involved.
> Meeting that expectation means making LLM calls, and LLM calls cost money.
Of course, and so does everything in the software world. The point is getting the cost so low that it’s basically free. The new DS V4 Flash or the smaller Qwen3.6 models are still really expensive compared to what we were used to in the economics of software, but it’s not unreasonable to expect these costs to continue falling down.
Rough chatgpt estimate says 3-5 orders of magnitude of difference compared to a typical user interaction with a SPA (db/cache lookup, CDN…)
Yeah, my point is that LLMs can reach that point too. Especially if you do it clientside. Copying was also way more expensive back in the 60s (accounting just for machine time and electricity, not storage cost, about 100 million times more expensive than today). Everything has a cost.
Definitely, of course. But I think the logic of the author’s article is based on there being a huge difference between the two, or rather a high cost of inference in absolute terms. And that can change and we have seen it change. Which breaks the entire premise of the article going forward, no?
Inference will be always more expensive than db operations or copying, sure. But how much more expensive is the question.
If I'm any kind of indicator of where Youtube users are headed, their AI chatbots in the video pages are going to kill their business model.
There are so many videos with hooks/teasers/'you won't believe what we discovered!!1', and now I just pause the video in the first second, ask "what's the tldr" and get the value from the video without a single ad impression (and likely racking up far more opex for Youtube than if I just streamed the video).
It depends on the solution. If AI is generating value, you can charge for the value.
Most SaaS already works this way. M365 or Adobe Creative Cloud are great examples. They value it like a life insurance policy and find ways to make you sticky. It’s easier to just buy it.
The first round of AI products suck because they are not well defined. Copilot only makes sense if you do shit in office and SharePoint isn’t a dumpster fire. In my large O365 environment the bottom 50% of users use less storage than the top 2%. So why would i buy copilot for my janitor?
When M365 E9 reconciles invoices automatically with Excel, I’ll pay $150/mo and fire a bunch of people.
I strongly predict this article is mostly pointless very soon.
Much as people may not want to like it, "software" as a product to buy and sell, even as a subscription, is probably going away, and will make about as much sense as "math" as a product.
We were already headed in this direction, but AI's going to rapidly accelerate this.
The inference cost discussion is interesting but I think it misses the more consequential shift. The real change in unit economics isn't what it costs to run the model — it's that the marginal cost of building a new feature dropped by an order of magnitude.
Previously the bottleneck was engineering time. Now a competent person with a frontier model can prototype in hours what used to take a team weeks. That compresses the cost side but it also compresses the moat. If your product can be rebuilt by a motivated person in a weekend, your pricing power evaporates regardless of your inference costs.
The SaaS companies that survive this will be the ones whose value comes from network effects, proprietary data, or integration depth — not from code complexity that used to be expensive to replicate.
I don't think the bottleneck was engineering time alone. There is context, brand, customer support, etc. I am not sure big complex products are particularly vulnerable in the way you state. However I have been thinking of a one smallish problem in a niche space that I could easily code for. There will be more of that.
"Inference" is just software running. It has always cost money to run software, it's just that it is generally too cheap to matter. If a client makes a regular API call to your server, you pay for that compute, probably in the form of a flat hosting fee. If too many calls come in and workload goes up, you pay for a more expensive hosting tier to handle it (or do dynamic scaling which is per-unit of compute).
Right now the "hosting" cost for inference is per-unit because it's new and expensive, but that won't last.
There is a lot of inefficiency right now keeping prices elevated. That will change very fast and soon paying for inference will likely resemble paying for hosting your app.
The bigger problem for SaaS is that the floor has risen - people can build their own solutions for things that they used to buy SaaS for. So the industry needs to level up and solve harder problems.
> There is a lot of inefficiency right now keeping prices elevated. That will change very fast and soon paying for inference will likely resemble paying for hosting your app.
I don't know about "very fast" or "soon" unless you're speaking in geological terms.
SOTA models like Kimi 3 require thousands of GB of RAM/VRAM to run at speeds that are real-time useful. Manufacturing the memory necessary for that quantity to be available at app-hosting prices will take decades. Software efficiency solutions might drop needed memory by an order of magnitude in that time...but a tenth of an enormous amount is still pretty darn big so won't get us there "soon".
Speaking of soon, tangentially related this just got announced: https://standardcode.ai/
They claim to support 24/7 agent coding with no VC subsidizations or money lost on a subscription, instead relying on optimizations on agent selection. I think it'll shift sooner rather than later.
Exactly my point in another comment. Just to illustrate this further: a rough ballpark of how the cost of intelligence fell since 2022 could be about 1000x, and continues to fall. Unfortunately, it’s really hard to measure.
It’s so cheap that companies choose to spend more on AI inference (more reasoning, more capabilities, longer context), not less - see Jevons paradox.
> That will change very fast and soon paying for inference will likely resemble paying for hosting your app.
The article have talked about this:
> You cannot, however, build a business model today on a cost breakthrough that has not happened yet. Even if costs do fall, Jevons paradox[5] tends to kick in: historically, when a resource gets cheaper, we find more uses for it rather than consuming less. Cheaper inference does not get banked as margin improvement; it gets consumed by deeper integration, more calls per interaction, more autonomous agents running in the background. The cost per call drops but the calls per user multiply.
I don't think it's at all certain this won't land back on the same unit economics as the old way. The cost of serving a user doesn't have to be free - it never has been - it just has to not be the dominating factor in your costs. I'm guessing there are still quite a lot of per-user costs that aren't easily visible. Like how many of your users are logging support requests, or suing you, or demanding bug fixes or custom integrations or a myriad of other things. And how much are you having to invest in security updates, regulatory compliance, marketing etc. Not to mention, users are getting well acclimatised to the idea of quotas and paying for increased limits.
I think part of the new equation may also become; "Why even pay for the SaaS in the first place if you can just forge the service exactly how you want it?" The benefits of unlimited access to the tool you forge are still there, its just a lot easier to make whatever tool you want.
Are there any examples of products containing ai inference that are successful? Products that are beyond just direct access to frontier LLM's, I mean.
As if making all the decisions that go into that product, hosting it, and maintaining it are free! Now do that for all the SaaS you subscribe to. Would you even get any real work done?
Hosting is cheap if its abstracted away. Same with the technical aspect of maintainance
About design and the product side of maintainance, what i have felt is that, deep down inside, the user does know exactly what they need. If they can learn to communicate that in a way another person or agent can understand, then design and maintainance will be fine. Thats a big if, but if the economics align for people to benefit from developing that skill, maybe they do.
There wouldn't be a product. Consumers would just use the underlying foundational models directly to solve their problem. ChatGPT and Claude and Gemini are the products, not your business.
Products are only viable when you do a lot of work to solve a problem, which is then shared by many potential customers. It makes sense to amortize the high costs of solving the problem across all the different customers to reap economies of scale. Businesses then pay for product design, hosting, and maintenance because those costs can also be amortized, and they are cheaper than a bespoke solution for each customer. But if the bespoke solution becomes cheaper than that, because it's generated by an LLM that doesn't need to be paid a living wage, there's no reason to have the product in the first place. Just solve your damn problem and let other people solve theirs.
This is an underrated factor in the market structure of the AI bubble going on now. Anecdotally, we're not seeing new AI-based products other than the foundational models and coding assistants gain traction. Why? Because AI makes it so easy to customize the solution to your particular needs that everybody is just solving their particular needs directly. It's the opposite of the Internet boom, where the network greatly expanded the potential market, reduced the cost of reaching them, and made it economical to spend large amounts of money building a software product that had a TAM of billions. The AI boom instead enables extremely cheap customization, which shrinks the market to a single customer who uses AI to directly solve their problem rather than building a product that's generally applicable.
> Consumers would just use the underlying foundational models directly to solve their problem
Whether this could happen is a critical question.
What ive found currently is that its not possible because the non-determinism of the model means you cant trust it, and these tasks are such that people expect determinism. Even if they dont then they dont tolerate the kind of stupid mistakes these models make
What could happen is that the model becomes good enough to be near deterministic, or as deterministic as humans. I dont know if the current next token prediction foundation is sufficient for that. But who knows, maybe it is
If you want determinism, you ask Claude Code to write you a computer program to the desired specs, and then run that program over and over again, instead of renting the software from a SaaS vendor. I've done that a bunch - I'll show Claude the inputs, I'll tell it what output format I want, tell it the tech stack, and then have it write the program. Then I'll ask for a systemd service or crontab entry to run it repeatedly.
For a lot of tasks, classical deterministic computer programming is better than an LLM. It's just that LLMs are very good at classical deterministic computer programming now.
> Consumers would just use the underlying foundational models directly to solve their problem. ChatGPT and Claude and Gemini are the products, not your business.
How am I going to use ChatGPT instead of, say, Figma or an office suite with my coworkers?? It's no use for it to spin up a clone for me if my coworkers can't collaborate. And if they can, what it will serve is a product.
How can it be cheaper for every company to re-invent the wheel?
> Products are only viable when you do a lot of work to solve a problem, which is then shared by many potential customers.
Yes, that's what businesses pay services for.
Actual example: my wife wanted a logo for the facepainting business she's starting with our 5yo, so she asked Gemini "I want a logo with these objects in it, drawn in these colors, with this business name." And it spits back a perfectly passable logo in 5 seconds. Done. No designer needed. No Figma needed. It cut out the entire value chain.
Is it as good as a professional could've done with Photoshop? No. But it's about 90-95% of the way there, which is good enough that neither her nor any customers would care.
A great deal of business is precisely this, tasks that need to be done but where you only care about "good enough" solutions. After all, a key principle of business is you don't outsource your competitive advantages. You pay people for the table stakes, the things that everybody needs but that they only need to be "good enough". LLMs can generate "good enough" facsimiles of a wide variety of fields.
Your wife would never have used Figma for that before LLMs anyway. What is she using to manage the accounts; the bread and butter of her business? Did she conjure up her own accounting and CRM software? If she vibe coded it what did she do to ensure there are no bugs? And in any case, she works alone, I presume? I'm talking about normal companies where people work together; the typical case.
My wife wouldn't have considered this a business before LLMs. She's mostly just doing friends' parties, and oftentimes getting paid in-kind (i.e. instead of bringing a gift, she just facepaints all the neighborhood kids).
Remember that in the mid-00s, this is what YouTube and the social web was. Folks doing little, productive economic activity, usually unpaid, which can now reach an audience through technology. And now the top career aspiration in young kids is "Be a social media influencer." That's how disruptive innovations start: with little transactions that aren't considered business at all, which then grow until they subsume the economy that everybody knows of.
> How can it be cheaper for every company to re-invent the wheel?
Companies not matching their prices to current reality, mostly.
There are already free, open source clones of numerous things but companies still pay for commercial versions. Do you think LLMs that you have to pay for are going to change that?
So many pivot opportunities, we’ll live remotely from office.
>"Why even pay for the SaaS in the first place if you can just forge the service exactly how you want it?"
This is the part I'm least convinced by.
Corporates used to develop their own tools and they moved away from it for many reasons. Cost of development was only one of those reasons.
If my manager prefers to pay for SaaS while I sit on my ass rather than have me develop an in-house solution, how do LLMs drastically change the equation?
[dead]
>Users began expecting something fundamentally different from software: not just tools that store and retrieve, but products that reason, generate, and respond.
Not really.
>Every inference call costs money.
Not really, either. If you buy your own GPU, rack it, and run an open model, there is no unit cost. This is just expensive hosting infra. You also pay unit costs for SaaS that your software uses (things like SMS etc).
> If you buy your own GPU, rack it, and run an open model, there is no unit cost.
No. There is economic opportunity cost (borrowing), energy cost, infra cost, depreciation / risk of failure with each unit of work, bandwidth, maintenance, and lots more. Small, but not zero, and often overlooked - especially the opportunity cost.
These are basically all fixed costs, not unit costs. You buy the GPU once and use it for as many calls as you have traffic for, and depreciate it over a fixed lifetime. You have to power it regardless of whether it's fully utilized or not. You have to maintain it by virtue of owning it, not really based on how many queries it has served. Bandwidth is the only one that really scales as a unit cost.
Open question whether this model is actually more economical than using the cloud AI service. The whole reason the industry moved to cloud computing in the first place was because computing had very high fixed costs, and the more these could be amortized over a fully-loaded query stream, the lower the unit costs.
You have to power it all the time, but the amount of power it uses while it’s on will change by up to a few orders of magnitude depending on the gpu. It’s not uncommon for a gpu to be pulling just a couple of watts at idles and several hundred at full tilt.
So the only way it’s a fixed cost is if you don’t pay for power. If you only consider the cost of the power, it might still be cheaper paying for an api.
Good points. Speaking from experience, it’s really hard to make it more economical than using a cloud AI service - even if you utilize the GPU to its fullest. There are, of course, other benefits, such as privacy/control/compliance/security, which should be the real reasons to do this, not cost.
You sweet sweet summer child.
Spoken like a guy who has never had to maintain bare metal infrastructure ops at scale. These things break, need re-imaged, have parts that break, have to be configured (now you need provisioning pipelines, monitoring, alarms, etc), have to be maintained when something goes wrong (swapping hardware and software, debugging the alarms into actually figuring out which bits are broken and/or misconfigured), have to be catalogued, have to be planned for, deprecated, and the finances accounted for through complicated accounting to show investors the Capex at quarterly meetings.
Then you have to make fleet decisions on how much of each type of server you will want to buy, expanded storage, how long you will support each generation of server, when you will order new hardware, how to order new hardware, lag for real world installs, hiring actual humans to fly around to all of your datacenters and do the actual installations / maintainance / etc. Then you will have to do contracts with individual datacenter operators for margins, electricity rates, hosting contracts, white glove ops hourly rates, etc.
Businesses that own their own hardware tend to have a lot of employees whose jobs are maintaining it and running the business side of that.
You do realize that a fully utilised GPU uses about 10 to 50 times more power than an idling one?
How is this different from hosting infra or just SaaS in general?
TFA contends that there is some fundamental shift in the economics of software, but it doesn't look to be very different from either a new SaaS dependency or racking new hardware.
> opportunity cost
We need a better word for this because the things you didn't do are not a cost
The things you would have done, but now can't do, are a loss to you
opportunity lost
> Build the product once, distribute it to a million users for roughly the same cost as distributing it to one. Every incremental customer flowed largely to the bottom line.
common misconception about software unit economics. With enterprise software (one that costs real $$$) there always more costs attached post shipping. Before client/server it was support. Then it was security and the constant threat of cyber attacks.
Distributing software is nothing like distributing books.
"Users began expecting something fundamentally different from software: not just tools that store and retrieve, but products that reason, generate, and respond."
Absolutely not. Customers want systems for sales, reservations, accounting, and taking stock. That's where almost all the SaaS money is and none of it benefits from AI - and never will.
This is one reason I've been trying to figure out tasks (and products based on those tasks) that can be pushed to the edge, either via small specialized models or small general purpose open models. I suspect the same desire to keep unit costs low is part of why Google is falling behind on the "frontier", but seemingly at the lead, or near it, on models that run on-device. I think they're just focused on making models for tasks that don't require boiling the ocean.
But, it's a hard problem. The models that run locally on normal computers/phones are pretty terrible compared to the frontier, without specialization and fine-tuning. And, even with specialization and fine-tuning, often a high-end general purpose model is going to do a better job and people don't need a bunch of local tools installed to do their various tasks.
> And, even with specialization and fine-tuning, often a high-end general purpose model is going to do a better job and people don't need a bunch of local tools installed to do their various tasks.
I think this is the critical point that would be interesting to see if it holds. Technology seemingly tends towards increased specialization.
> Technology seemingly tends towards increased specialization.
The bitter lesson says the exact opposite.
The bitter lesson is rather specific and IMO vastly over-applied. It basically says that you can beat hand-tuning by waiting a few years and using "more of the same" generic method.
It's not all that different from saying "Don't optimize software, just wait for faster hardware". Yet highly optimized software exist, and performance on currently available stacks is a competitive advantage.
What gives? I say: Staying ahead of the "Don't optimize, just wait" curve can absolutely make sense. At worst, your advantage decays after a few years. At best, you stay ahead by n number of years and keep increasing the gap as you invest more.
The question is: How soon?
If you make a specialist model and harness that takes six months to train and launch, will you have enough time to recoup your costs and make a profit before the generalists get good enough and cheap enough in your area of specialty and become the simpler option?
We already see "skills" repos containing a bunch of specialist stuff being pushed to general models via prose instructions for how to do various things and use various tools; and those models will do the tasks much more expensively and slowly, but without needing the user to find/buy/learn something specific to do the job.
The bitter lesson may be narrow, but the fact is people have been using spreadsheets for everything for 50 years, and only sometimes can a specialist tool effectively displace it in an organization. Very powerful generalist models are starting to feel like spreadsheets: The default way to solve problems in organizations. You might be able to pick off some of the specialist cases, but you might also miss.
I think the bitter lesson (from 2019, what is now seen as the tail end of an AI winter - probably wasn't at the time) made the most sense in academia during AI winter. There is no academic fame to be had by making something a little better for a limited time. When billions are at stake, there is potentially a lot of money (and utility for customers) to be gained by doing some extra work even if the work is thrown away soon after. Most commercial products contain a ton of extra work to be slightly better than the competition in one area or another.
Perhaps we can call this "The sweet dollars lesson".
The author misses one potential future of AI software where people use use their own subscriptions / API keys for their AI usage.
In this scenario the user would either sign into the platform via their Anthropic / OpenAI / Gemini account or use their API keys, and any of their usage would be billed to them directly.
The company then doesn't have to worry about the increasing costs from the AI usage.
Of course, in this future, AI providers become the new "Facebooks" of the world.
There's an interesting bellwether in mobile games right now. There are two easy statistics to track when trying random free-to-start games: time to first ad (TTFA), time to paywall (TTP).
Most LLM-using mobile games are already at incredibly weak TTP scores and perhaps the only current stratum of mobile games where TTP is almost always ahead of TTFA. TTFA before TTP is almost nonexistent because they claim to need a monthly subscription as soon as immediately after install, despite being advertised as free to start. It's also one of the few types of games where the paywall explicitly does not include "no ads". Some of these games running on monthly subscriptions still need ads for unit costs.
(The bulk of mobile games try for a sweet spot of TTFA in the order of hours of gameplay and TTP in the order of days of gameplay. Easier to get people hooked on your game if you can give them a few hours of uninterrupted fun up front.)
It seems pretty condemning of software economics with LLMs involved.
> Meeting that expectation means making LLM calls, and LLM calls cost money.
Of course, and so does everything in the software world. The point is getting the cost so low that it’s basically free. The new DS V4 Flash or the smaller Qwen3.6 models are still really expensive compared to what we were used to in the economics of software, but it’s not unreasonable to expect these costs to continue falling down.
Rough chatgpt estimate says 3-5 orders of magnitude of difference compared to a typical user interaction with a SPA (db/cache lookup, CDN…)
> and so does everything in the software world.
Well, no. Copying is free, or so near free it makes zero sense to charge. LLMs are just papering over the damage caused by profit.
Yeah, my point is that LLMs can reach that point too. Especially if you do it clientside. Copying was also way more expensive back in the 60s (accounting just for machine time and electricity, not storage cost, about 100 million times more expensive than today). Everything has a cost.
LLMs will never be as cheap or reliable as copying.
Definitely, of course. But I think the logic of the author’s article is based on there being a huge difference between the two, or rather a high cost of inference in absolute terms. And that can change and we have seen it change. Which breaks the entire premise of the article going forward, no?
Inference will be always more expensive than db operations or copying, sure. But how much more expensive is the question.
If I'm any kind of indicator of where Youtube users are headed, their AI chatbots in the video pages are going to kill their business model.
There are so many videos with hooks/teasers/'you won't believe what we discovered!!1', and now I just pause the video in the first second, ask "what's the tldr" and get the value from the video without a single ad impression (and likely racking up far more opex for Youtube than if I just streamed the video).
Fortunately, SaaS vendors have already conditioned users to accept usage limits within seat-based plans: “Upgrade to Pro for 50 GB of storage.”
How do we make subscribers become equally comfortable paying for AI usage? Tokens, credits, inference calls?
It depends on the solution. If AI is generating value, you can charge for the value.
Most SaaS already works this way. M365 or Adobe Creative Cloud are great examples. They value it like a life insurance policy and find ways to make you sticky. It’s easier to just buy it.
The first round of AI products suck because they are not well defined. Copilot only makes sense if you do shit in office and SharePoint isn’t a dumpster fire. In my large O365 environment the bottom 50% of users use less storage than the top 2%. So why would i buy copilot for my janitor?
When M365 E9 reconciles invoices automatically with Excel, I’ll pay $150/mo and fire a bunch of people.
Customers expect more, so they'll pay more. It really isn't more complicated than that.
Wonder how much this cost the author to write, as it’s just AI slop…
I strongly predict this article is mostly pointless very soon.
Much as people may not want to like it, "software" as a product to buy and sell, even as a subscription, is probably going away, and will make about as much sense as "math" as a product.
We were already headed in this direction, but AI's going to rapidly accelerate this.
The inference cost discussion is interesting but I think it misses the more consequential shift. The real change in unit economics isn't what it costs to run the model — it's that the marginal cost of building a new feature dropped by an order of magnitude.
Previously the bottleneck was engineering time. Now a competent person with a frontier model can prototype in hours what used to take a team weeks. That compresses the cost side but it also compresses the moat. If your product can be rebuilt by a motivated person in a weekend, your pricing power evaporates regardless of your inference costs.
The SaaS companies that survive this will be the ones whose value comes from network effects, proprietary data, or integration depth — not from code complexity that used to be expensive to replicate.
I don't think the bottleneck was engineering time alone. There is context, brand, customer support, etc. I am not sure big complex products are particularly vulnerable in the way you state. However I have been thinking of a one smallish problem in a niche space that I could easily code for. There will be more of that.