Listen, I work IT at a managed service provider, which means I see a dozen + small to medium business environments daily.
The number of users that I know of, out of the hundreds or thousands that I support, who are actively using “AI” continually to do their jobs… Well, I can likely count them on one hand
There’s a bunch that use whatever free tool they are allowed to access. Which kind of flies under the radar for me, but I still don’t think you’re dealing with double-digit percentages of people using “AI” that actually spend any money on it.
And these are business users, who work at businesses with rather deep pockets at times…
People are not buying the product. “AI” in the business world, isn’t making any money. So from what I’ve seen, this is less a matter of if, and more a matter of when and how bad.
As a tech person, I am not threatened by AI. It can’t do my job. Also, I’m the one who fixes it when things go wrong. And the only thing I’m looking forward to is being able to pick up cheap RAM, and storage when this all comes crashing down as AI companies cash out and liquidate their inventory. Maybe I can pick up a cheap AI server and repurpose it to be a 3D/gaming virtual desktop, so when my friends come over with their cheap laptops and on board graphics for a LAN party, I can just provision them a virtual gaming computer and we can go ham…
I’m going to save my sheckles until then…
I have limited experience with the market but I see it this way: anyone with AI holdings should liquidate at their earliest convenience. I don’t think they’ll be keeping up this charade for a lot longer.
Anyone who has tech holdings, either cash out, or be prepared to have that money held up for a while because it will take a long time for the prices to recover, and unless you want to cash out at a loss, you’re going to be holding for a long time. Everyone else, hold onto your butts. Probably everything will probably dip when the fecal matter hits the rotary air movement device. Non tech areas of the market should recover pretty quickly, so meh?
This is obvious to me that the whole thing has failed. That realization hasn’t hit execs yet.
some AI coprocessors, you’re right. they can’t be used.
Most of what nVidia is offering, while lacking display outputs, can still be used for gaming/VDI/vGPU use, and in general aren’t particularly bad at it, usually quite the opposite. might take a bit of work to get them going, and nVidia hasn’t made it easy with their whole licensing thing. Putting aside the challenges, it’s usually possible to adapt their SXM datacenter GPU modules to work with more common frameworks like PCIe, and then utilize those as virtual GPUs in a virtual desktop. This is not dissimilar to what they’re doing with the geforce now platform.
The big benefit of the chips is that they come with loads of memory, so cutting them up as a vGPU is actually a really optimal use-case. Most games are memory limited, which is to say they need more memory than most GPUs have, so the limit in performance is from the memory limitations, not the graphics processing limitations.
The AI processors that are GPU based, like almost everything that nvidia is using, can mostly be repurposed, but those that are ML-only, like AI coprocessors that show up in CPUs like the AI optimized CPUs from AMD and Intel, can’t really be re-used, but they’re usually not released as a standalone product. Those may still be able to perform physics processing offloads… so maybe not useless? I’m not sure.
Not strictly, there are usualy hurdles to overcome for home usage of datacentre tech, but it’s possible.
An easy example that comes to mind is that the datacentre GPU’s might have cooling configurations that rely on a large amount of airflow from the surrounding cabinet, which you’ll get in a datacentre, but not so much in a personal tower.
Also, noise, in a datacentre, noise is much less of an issue than in a personal computer, so they can eschew noise reduction in favour of performance.
Many people that live near data-centres would disagree, but i mean strictly from a usage point of view, not a broader community perspective.
Not strictly, there are usualy hurdles to overcome for home usage of datacentre tech, but it’s possible.
The hurdles are basically insurmountable with the hardware released after 2024.
The NVL72 for the Blackwell generation cost about $3 million and takes up a single server rack. The power consumption is about 130 kW, and most configurations require dedicated plumbing for the liquid cooling.
To put things in perspective, a residential electrical hookup is usually 50A or 100A for a house, with recommendations that anyone who is going to be charging electric cars should have 100A service. 100A at 240V is 24 kW.
So one server rack uses as much power as the maximum electrical capacity of 5 homes. You’ll never be able to pull that off in an actual residential environment.
Oh, and the newest 2026 generation, the Rubin NVL72s, use something like 230 kW of electrical power, almost twice as much as the previous 2024 generation.
Listen, I work IT at a managed service provider, which means I see a dozen + small to medium business environments daily.
The number of users that I know of, out of the hundreds or thousands that I support, who are actively using “AI” continually to do their jobs… Well, I can likely count them on one hand
There’s a bunch that use whatever free tool they are allowed to access. Which kind of flies under the radar for me, but I still don’t think you’re dealing with double-digit percentages of people using “AI” that actually spend any money on it.
And these are business users, who work at businesses with rather deep pockets at times…
People are not buying the product. “AI” in the business world, isn’t making any money. So from what I’ve seen, this is less a matter of if, and more a matter of when and how bad.
As a tech person, I am not threatened by AI. It can’t do my job. Also, I’m the one who fixes it when things go wrong. And the only thing I’m looking forward to is being able to pick up cheap RAM, and storage when this all comes crashing down as AI companies cash out and liquidate their inventory. Maybe I can pick up a cheap AI server and repurpose it to be a 3D/gaming virtual desktop, so when my friends come over with their cheap laptops and on board graphics for a LAN party, I can just provision them a virtual gaming computer and we can go ham…
I’m going to save my sheckles until then…
I have limited experience with the market but I see it this way: anyone with AI holdings should liquidate at their earliest convenience. I don’t think they’ll be keeping up this charade for a lot longer. Anyone who has tech holdings, either cash out, or be prepared to have that money held up for a while because it will take a long time for the prices to recover, and unless you want to cash out at a loss, you’re going to be holding for a long time. Everyone else, hold onto your butts. Probably everything will probably dip when the fecal matter hits the rotary air movement device. Non tech areas of the market should recover pretty quickly, so meh?
This is obvious to me that the whole thing has failed. That realization hasn’t hit execs yet.
AI chips arnt usable for personal use though, they seem to be limited to be used by the AI datancenters and not your computers
some AI coprocessors, you’re right. they can’t be used.
Most of what nVidia is offering, while lacking display outputs, can still be used for gaming/VDI/vGPU use, and in general aren’t particularly bad at it, usually quite the opposite. might take a bit of work to get them going, and nVidia hasn’t made it easy with their whole licensing thing. Putting aside the challenges, it’s usually possible to adapt their SXM datacenter GPU modules to work with more common frameworks like PCIe, and then utilize those as virtual GPUs in a virtual desktop. This is not dissimilar to what they’re doing with the geforce now platform.
The big benefit of the chips is that they come with loads of memory, so cutting them up as a vGPU is actually a really optimal use-case. Most games are memory limited, which is to say they need more memory than most GPUs have, so the limit in performance is from the memory limitations, not the graphics processing limitations.
The AI processors that are GPU based, like almost everything that nvidia is using, can mostly be repurposed, but those that are ML-only, like AI coprocessors that show up in CPUs like the AI optimized CPUs from AMD and Intel, can’t really be re-used, but they’re usually not released as a standalone product. Those may still be able to perform physics processing offloads… so maybe not useless? I’m not sure.
Not strictly, there are usualy hurdles to overcome for home usage of datacentre tech, but it’s possible.
An easy example that comes to mind is that the datacentre GPU’s might have cooling configurations that rely on a large amount of airflow from the surrounding cabinet, which you’ll get in a datacentre, but not so much in a personal tower.
Also, noise, in a datacentre, noise is much less of an issue than in a personal computer, so they can eschew noise reduction in favour of performance.
Many people that live near data-centres would disagree, but i mean strictly from a usage point of view, not a broader community perspective.
The hurdles are basically insurmountable with the hardware released after 2024.
The NVL72 for the Blackwell generation cost about $3 million and takes up a single server rack. The power consumption is about 130 kW, and most configurations require dedicated plumbing for the liquid cooling.
To put things in perspective, a residential electrical hookup is usually 50A or 100A for a house, with recommendations that anyone who is going to be charging electric cars should have 100A service. 100A at 240V is 24 kW.
So one server rack uses as much power as the maximum electrical capacity of 5 homes. You’ll never be able to pull that off in an actual residential environment.
Oh, and the newest 2026 generation, the Rubin NVL72s, use something like 230 kW of electrical power, almost twice as much as the previous 2024 generation.
i was talking more of the prosumer and close to prosumer units that could theoretically be retrofitted or housed in homelabs.
If datacentres went down en-masse i’d assume there’d be mix of hardware generations there.
Damn though, i didn’t know anything about the stuff you just mentioned, so my estimates of what could be available/usable are probably way off.
Now i have to go and do a deep dive, thanks.