My thought exactly. "May guarantee" is meaningless nonsense.
> Nvidia has revised its plans to support a proposed OpenAI data center project in Ohio and is now expected to initially guarantee less than $120 billion
There is no actual information in this article. "Plan", "proposed", "expected", "initially", "less than". It's just a report on the thoughts of some people.
I think they just mean "is expected to guarantee". Since the deal hasn't been signed, Reuters is hedging, but "may guarantee" is definitely awkward wording for something whose entire purpose is certainty...
If guarantee could be scaled down, it’s not a guarantee, is it? Expected guarantee, proposed guarantee, something along those lines. I think the point here is that the press has been presenting this as an iron-clad guarantee, while in reality, it’s not a guarantee at all.
Except once again Ed Zitron is right, and as he mentioned, there is no guarantee or contract signed, there is only a memorandum of understanding...
NVIDIA keeps showing the smarts acting like a bank, while having none of the liabilities and deferring them to Goldman...
"...Memorandums of understanding signed with six of the world’s premier financial institutions to create these partnerships aim to establish the first compute financing platforms of their kind at global scale to enable the AI infrastructure buildout across NVIDIA’s ecosystem..."
I mean, you can get quotes from four people convinced the earth is flat and that the earth is 6000 years old.
Ed Zitron is bearish on everything to do with AI.
I also don't believe the ROI is great investing in OpenAI and other similar corporations. But unlike Ed, I see the value in the technology. It's just the valuations that are wrong.
To be fair, and I haven't read Zitron in the last 6 months because I have a busy life, previously he also said AI was mostly useless. If he's come around on coding agents, I don't know.
Would that be the most expensive single thing ever built? The ISS cost around $150B and is commonly said to be the most expensive single item, but that does include running costs.
Yes, if this thing ever gets built, it will set a whole bunch of new records.
If it doesn't get built, it will still set a bunch of records; some of them probably quite infamous.
It is hard to describe the ridiculous scale they are trying to do there. For comparison, typical electrical demand is 17 GW and peaking to 25, with total generation capacity being 30 GW. That includes us-east-2, which is not a small data centre (consumes probably right around 2 GW, so represents about 10% of the state's power demand).
So they're talking about a project that would increase total power consumption in the state over 50%... in addition to building multiple nuclear power plants to fund it. Predicting 2,500 permanent jobs in a county of 27,000 total people, so that's a lot of people moving in.
The permanent jobs number really puts it in perspective. You're not just dropping a giant data center into an existing community at that point, you're potentially reshaping the county around it
That’s broadly known and at least they clearly account for it in their numbers, without all the “special financing vehicle” off balance sheet stuff that’s become common in the AI bubble era.
The loans will just take longer to repay. There is a market for Anthropic & OpenAI, it just likely doesn't have the 200B profit each year required for the maths to make sense.
If shit hits the fan, the companies collapse, then Nvidia gets their money from the investors anyways.
Buying the team and then actually making the models open is an interesting avenue for driving hardware demand (basically making openai models the new nemotron).
They don't need $200B profit for the math to make sense. Where are you getting that?
Anthropic is expected to IPO around $2T valuation.
That's around half Google's value, and Google's profit is around $130B.
So using the same P/E ratio as Google that implies $65B profit.
Of course Google is a mature company and Anthropic is growing revenue faster than any company in history so you'd expect Anthropic to have a higher P/E ratio than Google which means a lower profit to justify that valuation.
In any cay startups are valued on revenue rather than profit so that's the real number people will be looking at.
Nvidia is guaranteeing OpenAI's financing because, at a high level, people are more willing to lend/invest money with a profitable company than one losing money.
And Sam Altman needs a new source of financing because the US government is getting squirrelly about him continuing to raise capital from the Middle East in exchange for technology transfer.
I think a more likely scenario is that GPU time in the cloud will be very cheap. If LLM use/profitability doesn't follow the expected path, new deep learning applications could be heavily subsidised.
it'll be a rack based GPU without video ports. The only thing getting sick GPU's will be landfill when they all burn out and the data center rationalization happens.
More like rack-sized GPU that takes in half-megawatt of power as 800 volts DC, and requires 10 gallons per second of liquid cooling or it'll catch fire.
Nope. The scientific community is eagerly waiting for prices to drop.
Single 5090, single H100, a few 4090s, a few Blackwell PRO 6000... that's all there often is as dedicated devices for a faculty (apart from the oversubscribed bigger clusters with big stuff).
Apart from the chairs doing vision, most of them use them headless.
So, landfill will be highly unlikely, if things get sold fast enough.
Totally agree these won’t go to consumers but even after a crash I’d home they are able to be sold for parts and not just to landfills. If this all goes into the trash that would be even more tragic than it already is.
> Totally agree these won’t go to consumers but even after a crash I’d home they are able to be sold for parts and not just to landfills. If this all goes into the trash that would be even more tragic than it already is.
So I kinda wouldn't be surprised if a bunch of these GPUs go in the dump, given that it sounds like they'd be extremely power-hungry and difficult to use outside a hyperscale data center setting.
The proper server AI chips come on a board that isn't compatible with consumer pcie lanes and have no built-in fans. Completely different ball game in the ultra dense server world.
If Nvidia sells hardware for $100B with 75% cross margin, and provides $50 billion in backstop for that same hardware, it would be still be nicely profitable deal ($25B) if the backstop capacity would be a total write-off recovering $0. Reselling that capacity in some large discount below already low backstop price would increase the profits.
It's all those pension funds, sovereign wealth funds and Softbank getting into that $500 billion deal that will be hurt.
Pretty much, I have said it for a while now, Softbank and Oracle are the ones I would be worried about. Both of them have put their companies wealth behind this, if it goes down so will they.
Others have played it fairly smart in terms of insulating potential issues.
SoftBank has always managed to squeeze by after every mistake selling some early huge wins Alibaba, Nvidia, arm. Wonder if they still have any of those left in the back pocket.
Businesses aim to make the most profit possible with their resources. If they can make a 25% margin that is good, but if they can turn around sell thr same thing for a 50% margin, that is much better.
Basically what i am saying is maybe there is a better buyer than openai.
All economy is like infinite Hilbert Hotel. You create money out of thin air to get work done in real physical world to build products and services which will in future justify the past creation of money. It's like pulling yourself forward into the "desired" future with the help of newly minted money rope. Think about it this way, most of the money in the economy just sits there in bank accounts waiting to be deployed in future. So newly minted money in a way rearranges the physical world making the future world more suitable to justify past money supply increases. In case of hilbert hotel, we make room for the new guest by simply shifting everybody by n -> n+1. The settlement never arrives because the hotel in infinite, we call always do this.
Uh no, NVDIA helping startups get financing so they can buy NVDIA chips is inherently damaging because eventually the debtors will not help with the financing and startups will not be able to buy chips.
A Toyota dealership arranges a loan for you. Through a bank for a used vehicle, sometimes through Toyota itself for new cars.
A house builder will routinely take on part of the loan providing burden to get some of the interest.
Even someone selling you their thirty year old house will often provide seller financing.
You may have ideological opinions against this, which is fine. There are billions of people, for example that are fundamentally opposed to the idea of interest. But like it or not, this is how it has worked for the last ~500ish years.
No, I get this; it is not a problem. It becomes one when they cannot pay back this financing, in the event that they cannot build a sustainable business, which they cannot, because the capital cycle leads to overinvestment, meaning the financiers cannot meet their returns.
Taking a multi-decade perspective, I wonder if our general analysis is focused too much on the initial wave of LLM tech and current gen GPUs.
Owning a massive data center connected to water and power and network that can be targeted or converted to developing needs seems like a decent problem to have for the big cloud companies. We have compute hungry companies and media, in addition to cryptocurrencies etc, and we’ll have more of them in 2045.
I don’t know if I’m underestimating how purpose-built these datacenters are, or overestimating the accountants in the corporate vehicles building them, but the broader situation doesn’t seem as fragile as 1929 or 2008 (even if the businesses are overvalued and LLMs fall totally out of fashion).
H100, almost a five year old GPU, costs more to buy used now than it was to buy brand new at release.
Of course, everything has a lifespan.
Consider a simple arithmetic problem, and this mania will start making sense.
An H100 costs approximately 30k. You can run a decent latest open model on it at 1000 tokens per second batched. Cost on open router is 4$ per 1m tokens.
That's about 120k revenue per year if there's demand. So far, there's unlimited demand.
You, as one person, can likely not make the logistics of this work. But this really works with the economies of scale.
Now, because of that everyone wants to buy GPUs and we don't have enough.
The math works much better with a newer GPU that produces more tokens per second and consumes less energy to do so, even if it costs double. So why would anyone buy an old one? Because demand is orders of magnitude more than supply.
"A Toyota dealership arranges a loan for you. Through a bank for a used vehicle, sometimes through Toyota itself for new cars.
A house builder will routinely take on part of the loan providing burden to get some of the interest"
Wrong - when you borrow money the bank has instantaneously created money for you with the asset of your future promises of delivery of cash flows.
The bank is not using somebody elses money - it is literally creating it. Debt is akin to raw material for banks - the debt being the money it now owes you today.
Its interesting how many people get close to 90% of getting it, but the last 10% is actually 90% of the understanding.
It is strained analogy, heavily. I am not paid by Toyota. I am vetted for my ability to pay the loan. Me buying a car with borrowed money is not circular financing.
If most of Toyota earnings went from money they borrowed to me, it would be an issue. But, in fact, that is not how Toyota business works.
Car manufacturers, famously, have a pretty decent fraction of their revenue coming from their financing subsidiaries.
> I am vetted for my ability to pay the loan
Exactly!
Now see the article we are commenting on. Nvidia reduced the loan amount, presumably because they had doubts about OpenAI being able to pay it back.
The framework used to loan you money for buying a car and loaning a company billions of dollars to buy GPUs is largely the same. That's one of the accomplishments of modern economics.
Of course it can and does fail, but everything can go wrong.
This is probably a lot more related to the fact they want to make GPUs an asset class. Nvidia is banking on the fact there will be an entire market that will guarantee whatever anyone needs.
His numbers might be off or he might not have all of them.
But still I can not escape that he is most likely correct. All of this equipment needs to be paid. With interest and profit. With the usual overheads that the companies run. And if more is being bought each year. It doesn't seem like one and done deal. And then just asking where will all that money come from is very good one. And one we should be very honest about.
If you have someone walking on tightrope over the grand canyon in stormy weather, you don't need the exact wind speed to make educated guess that the winner will be gravity.
I think he is too emotional and overly sensational though.
What would happen to Nvidia, Anthropic, OpenAI, if tomorrow someone released an open weights model on HuggingFace that matched performance and accuracy of Opus 5 running locally on an RTX 5070? That won’t happen tomorrow, but it will likely happen someday… what’s the plan beyond “don’t be the one holding the bags?”
There’s no reason to assume frontier-level intelligence eventually collapses all the way onto a midrange consumer GPU. In fact, there are quite a few reasons not to assume that (information-theoretic constraints, etc).
I'm pointing out that there's no known information theoretic constraint about the impossibility of frontier AI models being improved to fit/run on a small GPU.
Please do not make up plausible sounding science facts.
Please do not assert I am making a claim I’m not making. Information-theoretic constraints exist. My comment does not require some specific, hard constraint to have been clearly defined, for my point to be valid.
If I were to say you could put a motorcycle in my car’s trunk, it would be perfect valid for me to say there are space constraints that make your idea unlikely. The same is true in this discussion, even though I have not computed the exact dimensions of the motorcycle and my car’s trunk.
Your claim was about frontier intelligence and midrange consumer GPUs. That's a pretty specific constraint.
Sure, there could be some point between a midrange consumer GPU and a pocket calculator where you can't fit enough 'intelligence'. But we really have no idea if the constraint is information theoretic or something completely different. Demonstrating that is the hard part, not finding the exact number of bits.
Talking about motorcycles in car trunks is just lazy false analogy here.
I will not claim a 5070, but there is already evidence in nature that you can get very good general intelligence with an order of magnitude less wattage.
There are constraints of course- training takes way longer.
I wonder if we'll eventually find that Darwin style evolution gets us close to the global optima of intelligence given constraints like size and energy.
We don't really have the tools to reason about this stuff yet. Exciting times.
But, it could happen for a coding-focused model, or an accounting-focused model, etc. most tasks only need a subset of the total model to be done effectively.
Could you not say the exact same of image gen models? For those that haven't kept up with that domain, you can now efficiently run high quality image gen models on any plain old video card, with phenomenal results.
If I could have shown up somewhere in 2022 with a Mac Studio M1 Max w/ 64GB of RAM running Qwen-3.6-27B or 35B-A3B, I would have pretty much been a demigod - to a degree far more impressive than being able to run Opus 5 locally today.
So yes, I think your scenario is likely to eventually happen, but there will be a much more powerful, capable frontier model then.
Those companies will be quick to copy the tech, inference cost would plummet and there is a greater chance that these companies could make it to solvency. At least in the short term. Long term it might not be so great as consume hardware catches up.
Nothing would really change IMO? 99% of users don't have anything like a RTX5070 (mobile especially).
Even if it did, it still doesn't make much economic sense running a model locally vs on a datacentre.
For example, I managed to just about squeeze a Q2 quant of Qwen 3.7 27b on my 9070XT. I get around 60tps decode (slightly faster prefill). _but_ it uses 300W of power to do so. At UK electricity rates of 30c/kWh this works out at something like 42c/MTok. I can get far far better models on openrouter cheaper than that, plus I'm not horrendously constrained on context length.
I dunno a lot of things said about AI economics sound like an IBM executive making reassuring statements about their terminal/mainframe business before the personal computer took off.
Like even if you run it in a datacenter in this scenario, you could do it on a cheap GPU instance in Azure, you still wouldnt need OpenAI or Anthropic specific clouds.
>uses 300W of power to do so.
There are plenty of people with phat electricity pipes in their on prem server rooms that have been vacated for cloud. Companies who want the benefits of AI but dont want the risk of sending their data to foreign API endpoints.
The analogy with IBM mainframe completely ignores Murphy's law which came up and lead to the small and fast chips we have today.
But Murphy's law is dead. No future chip will leapfrog easily current chips because we have reached hard phyical limits in chip density and downsizing. Huang's law by Jensen Huang focuses on something else and that is token performance per Watt at scale.
Blackwell needs double TDP than Hopper and Rubin again needs almost double TDP on a rack but in the end Rubin will be like 100x token performance per watt on a scaled data center. This means you have more energy need but you get multiples of token performance because you start scaling in the data center.
The local chip will never be able to keep up with the data center scaling economics. This is why everyone is so crazy about building data centers because they can see the economocs behind it.
What people don't seem to understand if tokens become more available and cheaper then not only more people can use them but a single person can use more as well. Why should you be limited to 1 AI agent? Why can't have you have multiple agents running on multiple devices daily for you?
This is why demand will grow exponentially with the growth of token economics. We have seen it for the last few years and much more is yet to come.
You're suggesting that if a very good and cheap AI model came out tomorrow everyone would rush out to rent Azure instances to run batch size 1 inference on their model?
Workloads will inflate just as they have been. Remember when llm assisted development used to be good only for a function, then a whole file, then a handful of files, then a code base, then a full stack, etc etc etc.
People will claim to have “enough” even though they already have the equivalent of last years capabilities locally.
It seems very very unlikely that an Opus 5 matching local model that runs on a 5070 will be released within the next 5 years (I don't want to say "ever").
If it does happen then NVidia will sell a lot of 5070s though!
"eventually" is actually a function of frontier model capabilities. You only get Qwen6-27B when you have Opus 7 producing extremely high quality tokens for them to train on. So the market for local models is always significantly behind the frontier, by definition.
probably not all that much... the market would dip, just like every time a new open weights model gets announced. but hundreds of millions of people aren't going to immediately self-hosting their own models.
the biggest winner in that scenario would be ai providers, who suddenly have a capable model that they can serve much more efficiently. and the incumbents have a whole lot of compute. wouldn't anthropic and openAI just start offering that open weights model at prices that nobody else could compete with?
They could but then their valuation is no longer justifiable, which breaks a lot of things downstream (loans being the biggie). They'd rather lose money than start making money in a non defensible way.
The article title is "Nvidia scales back funding guarantee for Ohio OpenAI data center, WSJ reports".
> Nvidia has revised its plans to support a proposed OpenAI data center project in Ohio and is now expected to initially guarantee less than $120 billion
There is no actual information in this article. "Plan", "proposed", "expected", "initially", "less than". It's just a report on the thoughts of some people.
"An insider told me NVidia spent 10 billions on hamburgers yesterday."
"An insider told me NVidia is thinking about maybe spending up to 10 billions on hamburgers or some foodstuffs by 2050".
Can you spot any differences?
"This article is about nothing". "Well of course it is, they say right on the page they talked to some people".
NVIDIA keeps showing the smarts acting like a bank, while having none of the liabilities and deferring them to Goldman...
"...Memorandums of understanding signed with six of the world’s premier financial institutions to create these partnerships aim to establish the first compute financing platforms of their kind at global scale to enable the AI infrastructure buildout across NVIDIA’s ecosystem..."
https://nvidianews.nvidia.com/news/nvidia-partners-with-apol...
A broken clock…
"Aswath Damodaran: Big Tech Has No Idea How AI Pays Off" - https://news.ycombinator.com/item?id=49229981
"Why Wall Street is Ignoring Big Tech's Debt" - https://youtu.be/NufJ7g63KSY
"Just how big is the hidden leverage of AI hyperscalers?" - https://archive.is/iLeYs
Lots a broken clocks would you say?
Ed Zitron is bearish on everything to do with AI.
I also don't believe the ROI is great investing in OpenAI and other similar corporations. But unlike Ed, I see the value in the technology. It's just the valuations that are wrong.
Now dont you go around quoting Ed Zitron :-) not fair...
"is bearish on everything to do with AI", "(not) see the value in the technology"
You totally got it wrong.
To be fair, and I haven't read Zitron in the last 6 months because I have a busy life, previously he also said AI was mostly useless. If he's come around on coding agents, I don't know.
There's a release from DoE about it: https://www.energy.gov/articles/fact-sheet-department-energy...
That's a horrible amount of gas energy generation.
https://www.datacenterdynamics.com/en/news/openai-in-talks-t... has more details. The whole campus build could be as much as $500B.
Would that be the most expensive single thing ever built? The ISS cost around $150B and is commonly said to be the most expensive single item, but that does include running costs.
If it doesn't get built, it will still set a bunch of records; some of them probably quite infamous.
It is hard to describe the ridiculous scale they are trying to do there. For comparison, typical electrical demand is 17 GW and peaking to 25, with total generation capacity being 30 GW. That includes us-east-2, which is not a small data centre (consumes probably right around 2 GW, so represents about 10% of the state's power demand).
So they're talking about a project that would increase total power consumption in the state over 50%... in addition to building multiple nuclear power plants to fund it. Predicting 2,500 permanent jobs in a county of 27,000 total people, so that's a lot of people moving in.
If shit hits the fan, the companies collapse, then Nvidia gets their money from the investors anyways.
Unlikely but quite interesting.
Anthropic is expected to IPO around $2T valuation.
That's around half Google's value, and Google's profit is around $130B.
So using the same P/E ratio as Google that implies $65B profit.
Of course Google is a mature company and Anthropic is growing revenue faster than any company in history so you'd expect Anthropic to have a higher P/E ratio than Google which means a lower profit to justify that valuation.
In any cay startups are valued on revenue rather than profit so that's the real number people will be looking at.
99.9% of you should stop doing valuation, especially since you don't know truly 'comparable firms' are.
Lazy slop. Worse than LLMs.
Is Nvidia guaranteeing the financing because OpenAI doesn’t have investor cash to pay the costs outright?
And Sam Altman needs a new source of financing because the US government is getting squirrelly about him continuing to raise capital from the Middle East in exchange for technology transfer.
Single 5090, single H100, a few 4090s, a few Blackwell PRO 6000... that's all there often is as dedicated devices for a faculty (apart from the oversubscribed bigger clusters with big stuff).
Apart from the chairs doing vision, most of them use them headless. So, landfill will be highly unlikely, if things get sold fast enough.
I'm not an accountant (so I could be wrong), but I'm vaguely under the impression it's sometimes financially beneficial to "write off" inventory, and to do that you have to destroy the items (e.g. https://en.wikipedia.org/wiki/Atari_video_game_burial, https://appleinsider.com/articles/23/05/30/apples-lisa-entom...).
So I kinda wouldn't be surprised if a bunch of these GPUs go in the dump, given that it sounds like they'd be extremely power-hungry and difficult to use outside a hyperscale data center setting.
Still fine for running APL with the dfns compiler, Futhark or other array languages directly hosted on the GPU itself!
If Nvidia sells hardware for $100B with 75% cross margin, and provides $50 billion in backstop for that same hardware, it would be still be nicely profitable deal ($25B) if the backstop capacity would be a total write-off recovering $0. Reselling that capacity in some large discount below already low backstop price would increase the profits.
It's all those pension funds, sovereign wealth funds and Softbank getting into that $500 billion deal that will be hurt.
Others have played it fairly smart in terms of insulating potential issues.
Goose value 71 here: https://group.softbank/media/Project/sbg/sbg/pdf/ir/investor...
Is Softbank making any money or just dissipating Alibaba gains?
Also, actual talk that goes with the presentation: https://www.youtube.com/watch?v=DtM0Cjb0dEU
What a fugly deck filled with nonsense.
Basically what i am saying is maybe there is a better buyer than openai.
It is not the first time, either; the capital cycle will prevail.
https://s-1.vercel.app/posts/the-capital-cycle-theory/
You pay Apple for a MacBook, Apple uses it to develop a better MacBook.
What goes wrong is leverage. We haven't seen much hint of the 10x leverage kind of deals that brought down the house in 2008.
But underestanding isn't entirely complete.
Money is created upon the issuance of debt.
Without money - trade would not be continuous.
A Toyota dealership arranges a loan for you. Through a bank for a used vehicle, sometimes through Toyota itself for new cars.
A house builder will routinely take on part of the loan providing burden to get some of the interest.
Even someone selling you their thirty year old house will often provide seller financing.
You may have ideological opinions against this, which is fine. There are billions of people, for example that are fundamentally opposed to the idea of interest. But like it or not, this is how it has worked for the last ~500ish years.
https://s-1.vercel.app/posts/the-capital-cycle-theory/
Neither of those look likely yet.
Owning a massive data center connected to water and power and network that can be targeted or converted to developing needs seems like a decent problem to have for the big cloud companies. We have compute hungry companies and media, in addition to cryptocurrencies etc, and we’ll have more of them in 2045.
I don’t know if I’m underestimating how purpose-built these datacenters are, or overestimating the accountants in the corporate vehicles building them, but the broader situation doesn’t seem as fragile as 1929 or 2008 (even if the businesses are overvalued and LLMs fall totally out of fashion).
This is more like a real estate cash grab.
I'd guess that it'll only get more and more difficult to obtain permits to do this in future. And expensive.
Similar to how if you want to buy land and build a house, it's pretty much impossible nowadays in the vicinity of a decently sized city.
But fifty year old houses are dime a dozen.
And if you look at the ARR of the companies “buying” them, I think we can see there’s some significant leverage going on.
Of course, everything has a lifespan.
Consider a simple arithmetic problem, and this mania will start making sense.
An H100 costs approximately 30k. You can run a decent latest open model on it at 1000 tokens per second batched. Cost on open router is 4$ per 1m tokens.
That's about 120k revenue per year if there's demand. So far, there's unlimited demand.
You, as one person, can likely not make the logistics of this work. But this really works with the economies of scale.
Now, because of that everyone wants to buy GPUs and we don't have enough.
The math works much better with a newer GPU that produces more tokens per second and consumes less energy to do so, even if it costs double. So why would anyone buy an old one? Because demand is orders of magnitude more than supply.
A house builder will routinely take on part of the loan providing burden to get some of the interest"
Wrong - when you borrow money the bank has instantaneously created money for you with the asset of your future promises of delivery of cash flows.
The bank is not using somebody elses money - it is literally creating it. Debt is akin to raw material for banks - the debt being the money it now owes you today.
Its interesting how many people get close to 90% of getting it, but the last 10% is actually 90% of the understanding.
If most of Toyota earnings went from money they borrowed to me, it would be an issue. But, in fact, that is not how Toyota business works.
> I am vetted for my ability to pay the loan
Exactly!
Now see the article we are commenting on. Nvidia reduced the loan amount, presumably because they had doubts about OpenAI being able to pay it back.
The framework used to loan you money for buying a car and loaning a company billions of dollars to buy GPUs is largely the same. That's one of the accomplishments of modern economics.
Of course it can and does fail, but everything can go wrong.
Which is not the same thing as circular financing we are talking about here.
Yes, if you abstract everything enough, everything is exactly the same as everything. But that does not mean it amounts to meaningful argument.
It could be wrong, sure. But extraordinary claims require extraordinary evidence.
For what it's worth, I agree with you on the leverage part, just not the scale part.
But still I can not escape that he is most likely correct. All of this equipment needs to be paid. With interest and profit. With the usual overheads that the companies run. And if more is being bought each year. It doesn't seem like one and done deal. And then just asking where will all that money come from is very good one. And one we should be very honest about.
I think he is too emotional and overly sensational though.
It's almost a given that whatever is frontier intelligence today will run on a potato in a few years.
I'm pointing out that there's no known information theoretic constraint about the impossibility of frontier AI models being improved to fit/run on a small GPU.
Please do not make up plausible sounding science facts.
If I were to say you could put a motorcycle in my car’s trunk, it would be perfect valid for me to say there are space constraints that make your idea unlikely. The same is true in this discussion, even though I have not computed the exact dimensions of the motorcycle and my car’s trunk.
Sure, there could be some point between a midrange consumer GPU and a pocket calculator where you can't fit enough 'intelligence'. But we really have no idea if the constraint is information theoretic or something completely different. Demonstrating that is the hard part, not finding the exact number of bits.
Talking about motorcycles in car trunks is just lazy false analogy here.
There are constraints of course- training takes way longer.
We don't really have the tools to reason about this stuff yet. Exciting times.
So yes, I think your scenario is likely to eventually happen, but there will be a much more powerful, capable frontier model then.
Even if it did, it still doesn't make much economic sense running a model locally vs on a datacentre.
For example, I managed to just about squeeze a Q2 quant of Qwen 3.7 27b on my 9070XT. I get around 60tps decode (slightly faster prefill). _but_ it uses 300W of power to do so. At UK electricity rates of 30c/kWh this works out at something like 42c/MTok. I can get far far better models on openrouter cheaper than that, plus I'm not horrendously constrained on context length.
Like even if you run it in a datacenter in this scenario, you could do it on a cheap GPU instance in Azure, you still wouldnt need OpenAI or Anthropic specific clouds.
>uses 300W of power to do so.
There are plenty of people with phat electricity pipes in their on prem server rooms that have been vacated for cloud. Companies who want the benefits of AI but dont want the risk of sending their data to foreign API endpoints.
But Murphy's law is dead. No future chip will leapfrog easily current chips because we have reached hard phyical limits in chip density and downsizing. Huang's law by Jensen Huang focuses on something else and that is token performance per Watt at scale.
Blackwell needs double TDP than Hopper and Rubin again needs almost double TDP on a rack but in the end Rubin will be like 100x token performance per watt on a scaled data center. This means you have more energy need but you get multiples of token performance because you start scaling in the data center.
The local chip will never be able to keep up with the data center scaling economics. This is why everyone is so crazy about building data centers because they can see the economocs behind it.
What people don't seem to understand if tokens become more available and cheaper then not only more people can use them but a single person can use more as well. Why should you be limited to 1 AI agent? Why can't have you have multiple agents running on multiple devices daily for you?
This is why demand will grow exponentially with the growth of token economics. We have seen it for the last few years and much more is yet to come.
People will claim to have “enough” even though they already have the equivalent of last years capabilities locally.
If it does happen then NVidia will sell a lot of 5070s though!
the biggest winner in that scenario would be ai providers, who suddenly have a capable model that they can serve much more efficiently. and the incumbents have a whole lot of compute. wouldn't anthropic and openAI just start offering that open weights model at prices that nobody else could compete with?
RTX 5070 prices go up ~N times. Nvidia makes more money because it's easier to make these things than it's to make a GB300.