I think all in all, a short term disruption in RAM price is acceptable considering most people don't upgrade their computers that often. In fact, it's a common sentiment on HN that since the introduction of Apple silicon, a number of people haven't found a need for new hardware yet.
RAM prices will eventually come down within 2-3 years.
My local Microcenter has tons of RAM at inflated prices, with even more inflated “list prices” to make it appear like it is on sale. Checking just now - they’ve lowered prices on some of it since I last looked 2 weeks ago.
So either sales have collapsed or the shortage doesn’t exist (yet?)
> Capital that once sustained the hardware and software ecosystem of the digital economy is being siphoned into subsidized “AI factories,” chasing artificial general intelligence instead of cheaper, more efficient investments in narrow AI.
The author seems to believe, wrongly, that these "AI factories" don't run on the same hardware or software as the digital economy.
> But they...don't. Look at the CPU to GPU ratio in any business vs an "AI business".
First off, the language of the article implies that the hardware itself, regardless of CPU/GPU ratio (which isn't mentioned at all), is inherently incapable of anything but serving AI. Its very assumption is a false dichotomy between "AI hardware" and "general purpose technology", an assumption that the hardware that runs "AI" can't serve YouTube videos, store cloud backups, or serve "narrow AI," whatever the author's definition of that is.
Second, it's incorrect to claim "nobody had GPUs in servers." Virtual desktop infrastructure, HPC (pharma sim, fintech modeling, oil and gas), media streaming and transcoding, VFX/Hollywood, all use similar CPU/GPU ratios.
Moreover, the specialized HBM + ASIC infrastructure many companies are now turning to, is driving a ton of research and innovation in the space. These technologies aren't new, and continue to be widely used in many other applications other than "AI," which the author would believe is the sole utility of these "AI investments."
I put together a new desktop earlier in the year... it would literally cost me over $1000 more today between ram and storage. And I was upset about paying $200 over list for my GPU. With today's prices, I'd have stuck to my old computer.
While the article is correct about the basic fact about an AI bubble driving up memory costs, which in turn makes it harder to purchase laptops and other consumer technology, there's a lot of dubious economics shoehorned in. I'm not sure why this article is here when so many other articles have been written about the memory shortage. (That being said, I don't think it's flag-worthy, just mediocre.)
The tools very commonly don't work as advertised tho, but people still buy them because of the incredible promises of increased efficiency. And since it's usually not easy to measure, they keep paying, sometimes companies even force their employees to use it.
It's an even bigger grift from the POV of companies building AI into their products - investors pressure companies to add AI features, managers demand thing that aren't achievable, developers bodge things together because they know it's worthless, marketing sells it like it's perfect, customers don't actually want it so sales bundles it into every plan and increases the price. Number go up, everyone gets a promotion. The customer has no alternatives because everyone else is doing the same.
> The tools work as advertised and are currently priced way cheaper than they cost to create?
Well, that's the grift problem right there - some people use public funds to subsidize a product, undermine the competition, front-run and scalp the hardware market, create inflation for everything, misallocate capital and deprive other assets of investment, all the while attempting a vendor lock-in at somebody else's expense.
Why would anybody see any of these as good is beyond me.
Retail memory price history: https://whereismyram.com/us
I think all in all, a short term disruption in RAM price is acceptable considering most people don't upgrade their computers that often. In fact, it's a common sentiment on HN that since the introduction of Apple silicon, a number of people haven't found a need for new hardware yet.
RAM prices will eventually come down within 2-3 years.
My local Microcenter has tons of RAM at inflated prices, with even more inflated “list prices” to make it appear like it is on sale. Checking just now - they’ve lowered prices on some of it since I last looked 2 weeks ago.
So either sales have collapsed or the shortage doesn’t exist (yet?)
> Capital that once sustained the hardware and software ecosystem of the digital economy is being siphoned into subsidized “AI factories,” chasing artificial general intelligence instead of cheaper, more efficient investments in narrow AI.
The author seems to believe, wrongly, that these "AI factories" don't run on the same hardware or software as the digital economy.
But they...don't. Look at the CPU to GPU ratio in any business vs an "AI business". Outside of a few specific verticals, nobody had GPUs in servers.
> But they...don't. Look at the CPU to GPU ratio in any business vs an "AI business".
First off, the language of the article implies that the hardware itself, regardless of CPU/GPU ratio (which isn't mentioned at all), is inherently incapable of anything but serving AI. Its very assumption is a false dichotomy between "AI hardware" and "general purpose technology", an assumption that the hardware that runs "AI" can't serve YouTube videos, store cloud backups, or serve "narrow AI," whatever the author's definition of that is.
Second, it's incorrect to claim "nobody had GPUs in servers." Virtual desktop infrastructure, HPC (pharma sim, fintech modeling, oil and gas), media streaming and transcoding, VFX/Hollywood, all use similar CPU/GPU ratios.
Moreover, the specialized HBM + ASIC infrastructure many companies are now turning to, is driving a ton of research and innovation in the space. These technologies aren't new, and continue to be widely used in many other applications other than "AI," which the author would believe is the sole utility of these "AI investments."
https://archive.is/SxxXj
I put together a new desktop earlier in the year... it would literally cost me over $1000 more today between ram and storage. And I was upset about paying $200 over list for my GPU. With today's prices, I'd have stuck to my old computer.
While the article is correct about the basic fact about an AI bubble driving up memory costs, which in turn makes it harder to purchase laptops and other consumer technology, there's a lot of dubious economics shoehorned in. I'm not sure why this article is here when so many other articles have been written about the memory shortage. (That being said, I don't think it's flag-worthy, just mediocre.)
What part of the AI experiment is the grift exactly?
The tools work as advertised and are currently priced way cheaper than they cost to create?
Putting "Intelligence" in the name of a hallucinatory text generator?
The tools very commonly don't work as advertised tho, but people still buy them because of the incredible promises of increased efficiency. And since it's usually not easy to measure, they keep paying, sometimes companies even force their employees to use it.
It's an even bigger grift from the POV of companies building AI into their products - investors pressure companies to add AI features, managers demand thing that aren't achievable, developers bodge things together because they know it's worthless, marketing sells it like it's perfect, customers don't actually want it so sales bundles it into every plan and increases the price. Number go up, everyone gets a promotion. The customer has no alternatives because everyone else is doing the same.
> The tools work as advertised and are currently priced way cheaper than they cost to create?
Well, that's the grift problem right there - some people use public funds to subsidize a product, undermine the competition, front-run and scalp the hardware market, create inflation for everything, misallocate capital and deprive other assets of investment, all the while attempting a vendor lock-in at somebody else's expense.
Why would anybody see any of these as good is beyond me.