Nvidia, CoreWeave และ Nebius: ภายในการจัดหาเงินทุนแบบวงกลมของ GPU Boom(io-fund.com)
io-fund.com
Nvidia, CoreWeave, and Nebius: Inside the Circular Financing of the GPU Boom
https://io-fund.com/ai-stocks/nvidia-coreweave-nebius-circular-financing-gpu-boom
91 comments
My understanding is that it's not about the money itself but the model:
- you fund a new company and sign long terms contracts with it - this new company uses the money you gave it and a lot of debt (backed by long term contracts) to build datacenters and buy a lot of GPU - your figures look great
What happens when they run out of debt or funds? If they reach some kind of profitability it's not a big deal, but if not ...
EDIT
Forget to mention the buyback of unused capacity problem: what happens to your figures when you have to buy back tons of unused GPUs?
- you fund a new company and sign long terms contracts with it - this new company uses the money you gave it and a lot of debt (backed by long term contracts) to build datacenters and buy a lot of GPU - your figures look great
What happens when they run out of debt or funds? If they reach some kind of profitability it's not a big deal, but if not ...
EDIT
Forget to mention the buyback of unused capacity problem: what happens to your figures when you have to buy back tons of unused GPUs?
Yes, circular financing is not by itself a problem.
It being that size, lasting for that long, and the total lack of viable products created by it are the problem. Financing only adds leverage, that makes every loss or profit larger.
It being that size, lasting for that long, and the total lack of viable products created by it are the problem. Financing only adds leverage, that makes every loss or profit larger.
It's not circular!
And if it is, it's not a problem!
And if it's a problem, it doesn't affect me!
And if it is, it's not a problem!
And if it's a problem, it doesn't affect me!
Those are 3 thresholds that a situations typically has to meet before people get upset about something. Arguably the 3rd one is not great, but the other two are just obvious and basic requirements. In this case even that last one is fine, the financial system is set up so that, in theory, other people losing money doing something stupid is a problem firewalled to just them.
This is not remotely new. When I worked at Intel ~20 years ago, Intel Capital invested in startups that would buy Intel hardware. Some of them succeeded, some did not.
But "invest in companies that may grow your own TAM" is an ancient strategy. Sometimes it works, sometimes it doesn't (like any strategy).
I'm not disagreeing with you, just saying it's business as usual.
But "invest in companies that may grow your own TAM" is an ancient strategy. Sometimes it works, sometimes it doesn't (like any strategy).
I'm not disagreeing with you, just saying it's business as usual.
For anyone else wondering, TAM seems to be ‘total addressable market’ if my searching is accurate.
I don't think its really the novelty of the situation that has people worried, its the scale of it and how that scale impacts the speed at which billions of dollars of market value could poof away when/if the music stops.
I don't think it is a Novelty to people in the sector but it is a novelty to people hearing about it from a YouTube video.
People have always had difficulty understanding large scales.
I don't feel that I have the expertise to analyse business structures like these accurately and impartially, yet I am under the impression that I have a better understanding than many who confidently talk about it and preach the end is nigh.
Even if the end is,in fact, nigh. It will not render their reasoning sound. They will have been right more by coincidence than judgement.
People have always had difficulty understanding large scales.
I don't feel that I have the expertise to analyse business structures like these accurately and impartially, yet I am under the impression that I have a better understanding than many who confidently talk about it and preach the end is nigh.
Even if the end is,in fact, nigh. It will not render their reasoning sound. They will have been right more by coincidence than judgement.
I think a whole lot of people understand quite well the difference in scale in this era compared to past eras of tech industry investment.
Webvan, Pets.com, eToys.com, Kozmo.com…all these dot com busts maxed out at less than 0.3 billion dollars in investment/IPO scale before they went under. A good amount of these share similarities with the AI bubble with a lot of them promising to be the e-commerce infrastructure of the future with “unlimited potential” as brick and mortar purchases were all predicted to move online. Webvan was going to be the automated warehouse of the future, for example.
Even the successful giant unicorns look minuscule in comparison. YouTube’s total investment was under $12 million before Google bought it for $1.65 billion, which looks like peanuts compared to these Hertz rent-a-server companies.
SoftBank dumping $8 billion into Uber looks positively quaint by comparison.
Webvan, Pets.com, eToys.com, Kozmo.com…all these dot com busts maxed out at less than 0.3 billion dollars in investment/IPO scale before they went under. A good amount of these share similarities with the AI bubble with a lot of them promising to be the e-commerce infrastructure of the future with “unlimited potential” as brick and mortar purchases were all predicted to move online. Webvan was going to be the automated warehouse of the future, for example.
Even the successful giant unicorns look minuscule in comparison. YouTube’s total investment was under $12 million before Google bought it for $1.65 billion, which looks like peanuts compared to these Hertz rent-a-server companies.
SoftBank dumping $8 billion into Uber looks positively quaint by comparison.
> If they reach some kind of profitability it's not a big deal, but if not ...
What is the end of this sentence?
What is the end of this sentence?
There are two types of people. Those who can extrapolate from incomplete data, and
... then it is a big deal.
Simple answers: because it makes for a good headline.
You're probably just responding to the headline but this person is an AI bull and isn't claiming it's a big deal, she's going into it and explaining it.
It's a bad headline because most of the article isn't about circular financing and it's only 5.7% of anyway.
People are looking for the AI bear case - so this headline gotta work better. Its not a bad idea haha. More people suspect there is some circular shenanigans but want confirmation -- so maybe this is the best way to lure them in. Come as the bear, stay for the bull.
With just these 2 comments, now I'm really gonna read that article.
With just these 2 comments, now I'm really gonna read that article.
Can someone even outline the AI bull case? I can’t fathom one at all.
https://isaiprofitable.com/
The only profitable company is the one running the scam.
https://isaiprofitable.com/
The only profitable company is the one running the scam.
Do you use it or is this speculation?
The AI tooling I used 12-24 months ago if frozen in time, monetized correctly is probably 100x the capability of what software could do before (And software was already eating the world long before AI). The bull case is that we just invented the 21st century equivalent of the printing press or electricity. And that website is the 19th century equivalent of someone criticizing electricity as a concept because it would be expensive to build power lines.
If anything it's a miracle that the US economy is so efficient that we can just skip all the small talk and bullshit and build out the infrastructure to support AI immediately.
The AI tooling I used 12-24 months ago if frozen in time, monetized correctly is probably 100x the capability of what software could do before (And software was already eating the world long before AI). The bull case is that we just invented the 21st century equivalent of the printing press or electricity. And that website is the 19th century equivalent of someone criticizing electricity as a concept because it would be expensive to build power lines.
If anything it's a miracle that the US economy is so efficient that we can just skip all the small talk and bullshit and build out the infrastructure to support AI immediately.
0. Pour money on the fire
1. AI somehow becomes AGI because money implies "emergence" I guess?
2. Profit somehow?One sentence summary of the bull case: Anything you would have paid a human to do, you will pay an AI to do instead.
If we're all out of work then who buys the things AI makes?
All the corporations buying inputs for their products and services and all the people on unemployment.
Just the look and feel and the subscribe fixed position in particular, made me bounce.
It sounds like Nvidia is not only supplying GPUs first to neoclouds, it is also supplying them for free if they cannot be resold:
"Furthermore, in the case of CoreWeave, Nvidia has also provided a significant financial backstop against unsold GPU capacity. Under the agreement with an initial value of $6.3 billion, “in instances where [CoreWeave’s] datacenter capacity is not fully utilized by its own customers, NVIDIA is obligated to purchase the residual unsold capacity through April 13, 2032.” In other words, Nvidia is committed to purchasing unsold GPU capacity if CoreWeave is unable to find another buyer. With an initial value of $6.3 billion, there is the potential that the arrangement could become larger over time."
I don't know how Nvidia is handling Coreweave GPU sales revenue in their accounting, but it sounds to me like it should have a pretty big asterisk attached to it. It's more like a consignment arrangement than an actual sale. And it obviously creates a huge incentive for Coreweave to over-order GPUs, since there's no risk (I doubt they're paying cash up front).
"Furthermore, in the case of CoreWeave, Nvidia has also provided a significant financial backstop against unsold GPU capacity. Under the agreement with an initial value of $6.3 billion, “in instances where [CoreWeave’s] datacenter capacity is not fully utilized by its own customers, NVIDIA is obligated to purchase the residual unsold capacity through April 13, 2032.” In other words, Nvidia is committed to purchasing unsold GPU capacity if CoreWeave is unable to find another buyer. With an initial value of $6.3 billion, there is the potential that the arrangement could become larger over time."
I don't know how Nvidia is handling Coreweave GPU sales revenue in their accounting, but it sounds to me like it should have a pretty big asterisk attached to it. It's more like a consignment arrangement than an actual sale. And it obviously creates a huge incentive for Coreweave to over-order GPUs, since there's no risk (I doubt they're paying cash up front).
From an accounting perspective, this absolutely isn't a consignment agreement.
The sale of the GPUs by Nvidia to CoreWeave is real. CoreWeave pays Nvidia cash and becomes the owner of the asset, so it's properly booked as a sale. If it can't sell capacity, the GPUs are not returned to Nvidia.
CoreWeave is using debt to make the purchases but the backstop provided by Nvidia ostensibly helps it get better loan terms. That doesn't change the accounting.
If Nvidia has to purchase unused capacity, it simply becomes an operating expense for Nvidia.
Nvidia's exposure is the $6.3 billion backstop obligation and the equity it holds in CoreWeave.
The sale of the GPUs by Nvidia to CoreWeave is real. CoreWeave pays Nvidia cash and becomes the owner of the asset, so it's properly booked as a sale. If it can't sell capacity, the GPUs are not returned to Nvidia.
CoreWeave is using debt to make the purchases but the backstop provided by Nvidia ostensibly helps it get better loan terms. That doesn't change the accounting.
If Nvidia has to purchase unused capacity, it simply becomes an operating expense for Nvidia.
Nvidia's exposure is the $6.3 billion backstop obligation and the equity it holds in CoreWeave.
>CoreWeave is using debt to make the purchases but the backstop provided by Nvidia ostensibly helps it get better loan terms.
According to the article, the $6.3B is a floor, not a ceiling. And it's not clear whether CoreWeave is actually paying cash or getting the GPUs on credit. If the full amount is getting booked, it's an accounting loophole that's being exploited. If GM sells Hertz a million cars, but says "Hey, we'll buy these back if you can't rent them," can GM book all those cars as actual revenue? What if Hertz only has to pay 10% up front and the rest in 5 years?
According to the article, the $6.3B is a floor, not a ceiling. And it's not clear whether CoreWeave is actually paying cash or getting the GPUs on credit. If the full amount is getting booked, it's an accounting loophole that's being exploited. If GM sells Hertz a million cars, but says "Hey, we'll buy these back if you can't rent them," can GM book all those cars as actual revenue? What if Hertz only has to pay 10% up front and the rest in 5 years?
Your GM/Hertz comparison is not applicable here. Under GAAP accounting rules, GM wouldn't be able to book those as sales because it was obligated (or likely) to buy back the asset. Under the rules, this means the transaction gets treated as an operating lease. The cars would stay on GM's balance sheet and the revenue would get recognized over the lease period.
The CoreWeave-Nvidia deal is not the same because Nvidia is not buying back the asset (the GPUs). CoreWeave has title to the chips and if they're worth nothing in 5 years, that's a problem for CoreWeave and its lenders.
What Nvidia obligated itself to was buying compute capacity, which Nvidia would be able to use for its own workloads.
In the GM/Hertz analogy, this is like GM selling Hertz the cars and saying "If you can't find renters for them, we'll rent them from you at market rates, up to $x." Under GAAP accounting rules, GM would book the car sales as revenue, the commitment to rent would be a purchase obligation, and if the rentals ever occurred, GM would incur the costs as an operating expense.
There is a question of whether the CoreWeave-Nvidia deal structure is sensible economically, and how much risk is being created. But there's no GAAP accounting question here. At all.
The CoreWeave-Nvidia deal is not the same because Nvidia is not buying back the asset (the GPUs). CoreWeave has title to the chips and if they're worth nothing in 5 years, that's a problem for CoreWeave and its lenders.
What Nvidia obligated itself to was buying compute capacity, which Nvidia would be able to use for its own workloads.
In the GM/Hertz analogy, this is like GM selling Hertz the cars and saying "If you can't find renters for them, we'll rent them from you at market rates, up to $x." Under GAAP accounting rules, GM would book the car sales as revenue, the commitment to rent would be a purchase obligation, and if the rentals ever occurred, GM would incur the costs as an operating expense.
There is a question of whether the CoreWeave-Nvidia deal structure is sensible economically, and how much risk is being created. But there's no GAAP accounting question here. At all.
Great explanation. Maybe another metaphor, it’s like a builder/developer buying land from someone. They own the land, they get the title, it’s theirs.
The land owner saying “hey if you can’t sell all the apartments we’ll buy what’s left” doesn’t in any way negate the sale or revenue accounting as per GAAP etc.
The land owner saying “hey if you can’t sell all the apartments we’ll buy what’s left” doesn’t in any way negate the sale or revenue accounting as per GAAP etc.
I mean, okay sure, but modify the counterexample they suggested slightly and then it's the same thing.
If GM promised to "rent out" (instead of buy back) the cars it sold to Hertz as a backstop (if not enough customers are renting), then the comparison is apt.
If GM promised to "rent out" (instead of buy back) the cars it sold to Hertz as a backstop (if not enough customers are renting), then the comparison is apt.
No, it isn't and this is simple GAAP accounting.
If GM sold cars to Hertz and then agreed to rent them from Hertz if Hertz was unable to rent them, it would not be consignment. It would be a sale and then purchase commitment, with the cost of the rentals taken as an operating expense.
Is the CoreWeave-Nvidia arrangement "good"? Time will tell. But there's no accounting issue here and even non-accountants can educate themselves on the subject because the least effective way to criticize these deals is to make accounting arguments that don't align to actual accounting principles.
If GM sold cars to Hertz and then agreed to rent them from Hertz if Hertz was unable to rent them, it would not be consignment. It would be a sale and then purchase commitment, with the cost of the rentals taken as an operating expense.
Is the CoreWeave-Nvidia arrangement "good"? Time will tell. But there's no accounting issue here and even non-accountants can educate themselves on the subject because the least effective way to criticize these deals is to make accounting arguments that don't align to actual accounting principles.
> there's no accounting issue here
This seems like a really narrow interpretation of what's going on. Is there any room to doubt/discuss whether GAAP rules could be improved? Or why the deal has been structured this way?
Why shouldn't we look through this arrangement? NVIDIA isn't in the business of purchasing outsourced GPU time. They could make better use of unused GPUs by repurchasing them for resale to another customer. If they're not doing that, it already seems likely that they specifically did this to guarantee that the revenue could be recognised.
Sure, NVIDIA's risk exposure could (legally) sit on their books without being recognised until it's already too late. That doesn't mean we shouldn't scrutinize them.
This seems like a really narrow interpretation of what's going on. Is there any room to doubt/discuss whether GAAP rules could be improved? Or why the deal has been structured this way?
Why shouldn't we look through this arrangement? NVIDIA isn't in the business of purchasing outsourced GPU time. They could make better use of unused GPUs by repurchasing them for resale to another customer. If they're not doing that, it already seems likely that they specifically did this to guarantee that the revenue could be recognised.
Sure, NVIDIA's risk exposure could (legally) sit on their books without being recognised until it's already too late. That doesn't mean we shouldn't scrutinize them.
> Is there any room to doubt/discuss whether GAAP rules could be improved?
Looking at the structure of the deal and analyzing the risks is perfectly valid. Screaming "accounting subterfuge!" when this is simple GAAP accounting is a different matter.
CoreWeave is buying chips from Nvidia, paying Nvidia full price, and taking title to them. Nvidia has no right to take them back. It instead has a potential obligation, subject to various conditions, to purchase a separate service (compute) from CoreWeave.
GAAP rules are updated on a regular basis. If you want different GAAP rules for this type of deal, you at least need enough knowledge about accounting to make a sensible suggestion.
> NVIDIA isn't in the business of purchasing outsourced GPU time.
This simply isn't true. Google "DGX Cloud". Nvidia has a real business selling cloud-based compute for training foundation models and running heavy AI workloads, and leasing compute from its hyperscaler chip customers instead of competing with them was a strategic decision Nvidia made.
So yes, these types of arrangements should be scrutinized. But to do so intelligently requires a basic grasp of accounting rules and the business models.
Looking at the structure of the deal and analyzing the risks is perfectly valid. Screaming "accounting subterfuge!" when this is simple GAAP accounting is a different matter.
CoreWeave is buying chips from Nvidia, paying Nvidia full price, and taking title to them. Nvidia has no right to take them back. It instead has a potential obligation, subject to various conditions, to purchase a separate service (compute) from CoreWeave.
GAAP rules are updated on a regular basis. If you want different GAAP rules for this type of deal, you at least need enough knowledge about accounting to make a sensible suggestion.
> NVIDIA isn't in the business of purchasing outsourced GPU time.
This simply isn't true. Google "DGX Cloud". Nvidia has a real business selling cloud-based compute for training foundation models and running heavy AI workloads, and leasing compute from its hyperscaler chip customers instead of competing with them was a strategic decision Nvidia made.
So yes, these types of arrangements should be scrutinized. But to do so intelligently requires a basic grasp of accounting rules and the business models.
I'm not the commenter claiming that this currently violates GAAP - that's someone else.
To summarise my opinion, subjectively it seems like a better distinction could be made in GAAP to look through this agreement and others like it. (Hypothetically if Hertz agreed to rent back rather than repurchase, like mentioned in a previous comment, that would also be suspect). But I'm not the one to propose what the preconditions would be.
> Looking at the structure of the deal and analyzing the risks is perfectly valid. Screaming "accounting subterfuge!" when this is simple GAAP accounting is a different matter.
It can be legal and still be subterfuge. Everyone involved in the deal has a clear incentive to ensure Coreweave gets to recognise revenue, and gets to show growth on paper. It's the same reason why SoftBank paying OpenAI $800mln for services in 2025 stinks a bit - they don't need the services but the deal goes ahead anyway.
> This simply isn't true. Google "DGX Cloud". Nvidia has a real business selling cloud-based compute for training foundation models and running heavy AI workloads, and leasing compute from its hyperscaler chip customers instead of competing with them was a strategic decision Nvidia made.
Sorry - you're entirely correct here. Though remember we're talking about a scenario where Coreweave aren't able to sell their capacity. If there's such a dramatic hole in demand, who are NVIDIA selling their compute to? This repo agreement won't give NVIDIA capacity that they need in the 90% of cases but will force them to purchase capacity they won't need in the 10%.
s/10%/some other probability/
To summarise my opinion, subjectively it seems like a better distinction could be made in GAAP to look through this agreement and others like it. (Hypothetically if Hertz agreed to rent back rather than repurchase, like mentioned in a previous comment, that would also be suspect). But I'm not the one to propose what the preconditions would be.
> Looking at the structure of the deal and analyzing the risks is perfectly valid. Screaming "accounting subterfuge!" when this is simple GAAP accounting is a different matter.
It can be legal and still be subterfuge. Everyone involved in the deal has a clear incentive to ensure Coreweave gets to recognise revenue, and gets to show growth on paper. It's the same reason why SoftBank paying OpenAI $800mln for services in 2025 stinks a bit - they don't need the services but the deal goes ahead anyway.
> This simply isn't true. Google "DGX Cloud". Nvidia has a real business selling cloud-based compute for training foundation models and running heavy AI workloads, and leasing compute from its hyperscaler chip customers instead of competing with them was a strategic decision Nvidia made.
Sorry - you're entirely correct here. Though remember we're talking about a scenario where Coreweave aren't able to sell their capacity. If there's such a dramatic hole in demand, who are NVIDIA selling their compute to? This repo agreement won't give NVIDIA capacity that they need in the 90% of cases but will force them to purchase capacity they won't need in the 10%.
s/10%/some other probability/
Billions of dollars sounds like a literal "big deal", but not necessarily "a problem". Worst case for nVidia is they lose 2 billion dollars, NBD.
Unless you believe that we are at the point where
- AI is providing economic value detached of humanity (which we are not)
- There exists AI companies generating a profit from non AI companies (not yet)
Then AI companies investing in AI companies means jack
- AI is providing economic value detached of humanity (which we are not)
- There exists AI companies generating a profit from non AI companies (not yet)
Then AI companies investing in AI companies means jack
anyone can isolate one number to fit their bias. if you look at the wider financing in the industry and the context of multiple AI deals in the billions without any cold hard cash flow or reasoning it kinda makes sense
> Why is it a big deal? Nvidia invested $2b into CoreWeave for 9% equity stake.
Depends if they actually got the $2b in real money. There's a difference.
It's a big deal if no money was involved. Nothing even entered the company directly. Some deals have structured with Special Purpose Vehicles where money goes to the SPV. The SPV buys GPUs with it (from Nvidia). GPUs is loaned back to the company involved. So this company is stuck with this GPU rental, which may or may not be what they want and not $2b.
This sounds like a bad deal? So Nvidia had to sweeten the deal and promise min utilization on those GPUs by renting it themselves even if they don't need it.
So what's income and what's expense here?
That's the problem. It's inflated and messed up.
Depends if they actually got the $2b in real money. There's a difference.
It's a big deal if no money was involved. Nothing even entered the company directly. Some deals have structured with Special Purpose Vehicles where money goes to the SPV. The SPV buys GPUs with it (from Nvidia). GPUs is loaned back to the company involved. So this company is stuck with this GPU rental, which may or may not be what they want and not $2b.
This sounds like a bad deal? So Nvidia had to sweeten the deal and promise min utilization on those GPUs by renting it themselves even if they don't need it.
So what's income and what's expense here?
That's the problem. It's inflated and messed up.
NVIDIA's $2b into CoreWeave was a stock purchase.
https://investors.coreweave.com/news/news-details/2026/NVIDI...
https://investors.coreweave.com/news/news-details/2026/NVIDI...
This isn't really related to the post, but I need to vent I suppose.
CoreWeave feels very YC-ish. I thought I had an in as a referral for a position there and got interviewed by someone who knew a lot of my peers where I worked. Dude seemed to ask very textbook style questions that you would only learn if you went to a school system for this particular position/subject. I guess I didn't answer to their satisfaction despite knowing more than them on almost everything else. I suppose I'm still bitter seeing as I interviewed with them three times for two different roles. Absolutely wild.
CoreWeave feels very YC-ish. I thought I had an in as a referral for a position there and got interviewed by someone who knew a lot of my peers where I worked. Dude seemed to ask very textbook style questions that you would only learn if you went to a school system for this particular position/subject. I guess I didn't answer to their satisfaction despite knowing more than them on almost everything else. I suppose I'm still bitter seeing as I interviewed with them three times for two different roles. Absolutely wild.
Circular financing is a dead horse - dont beat it. Instead, what is more interesting could be: Is there a path to these builds becoming economically profitable ? Towards this, some metrics to watch are: 1) ROI per token per dollar 2) Enterprise token budgets. And at what point there is an overbuild relative to the token roi. Alternatively, pressure on token costs due to the open weights models etc.
But how do you even measure the ROI of tokens? I don’t think it’s possible, tokens aren’t fungible. You can spend millions on tokens that don’t contribute one bit to the company revenue, then spend $10 that will actually result in useful things
These questions can't really be answered now because things are moving too fast. That may explain why people are latching on to things they can prove like circular financing even if those arguments are pretty weak.
if the money moves in circles the consequences of new money stopping when predicted profitability falls become a lot more dramatic
all money moves in circles, it's just a question of number of stops.
think about stuff like pork barrel funding for aerospace, which props up jobs, which generates funding for political campaigns that perpetuate pork barrel funding.
think about stuff like pork barrel funding for aerospace, which props up jobs, which generates funding for political campaigns that perpetuate pork barrel funding.
IMO, it can be profitable - but only at the business level. A business with many software devs can pay the steep price for access.
For almost everything else, the answer is no. No one else would pay the real costs to run them.
It'll require the whole industry to shrink down massively compared to what we're seeing now - down to a profitable (and much smaller) core.
For almost everything else, the answer is no. No one else would pay the real costs to run them.
It'll require the whole industry to shrink down massively compared to what we're seeing now - down to a profitable (and much smaller) core.
> For almost everything else, the answer is no. No one else would pay the real costs to run them. It'll require the whole industry to shrink down massively compared to what we're seeing now - down to a profitable (and much smaller) core.
If there is any data to support this, please share.
If there is any data to support this, please share.
Dumb question, but when the Nebius capacity dashboard says they have around 3 non-preemptible B200s available, does that mean _total_, or is it just how many I myself might be able to rent on demand?
One aspect of the profitability might be the utilization and the pricing a few years down the line for slightly older hardware. Already now it seems like the increased processing you get from newer devices versus the cost difference makes something like an H100 or even A100 significantly less desirable than newer more powerful ones. As an individual, I am happy to be able to get an H200 on demand, but the B200 or B300 can do so much more work with optimized software and models for only modestly more cost that if those become available then from a business perspective you really have to prefer that if you can keep it occupied.
Then with Vera Rubin being like 3 times more effective or whatever, that adds a new layer of gradual obsolescence. So the question is can they keep the pricing up on the older ones a few years down the line enough to fill out the end of those expected payback periods.
The real boogeyman for a neocloud that has heavily invested in expensive Nvidia hardware might be a variation of that beyond Nvidia with startups that have even more dramatic efficiency increases pushing the leading edge even further. For example, if companies like Mythic AI and d-Matrix could somehow rapidly rapidly scale, that would push prices down for all of Nvidia hardware that is significantly less efficient.
I guess so far it doesn't look like any startups with really big efficiency breakthroughs are even close to being able to scale like Nvidia though, especially with the manufacturing and power crunch. But I suspect some of that is because of favoritism and strong arming protecting investments rather than a free and fair ecosystem.
One aspect of the profitability might be the utilization and the pricing a few years down the line for slightly older hardware. Already now it seems like the increased processing you get from newer devices versus the cost difference makes something like an H100 or even A100 significantly less desirable than newer more powerful ones. As an individual, I am happy to be able to get an H200 on demand, but the B200 or B300 can do so much more work with optimized software and models for only modestly more cost that if those become available then from a business perspective you really have to prefer that if you can keep it occupied.
Then with Vera Rubin being like 3 times more effective or whatever, that adds a new layer of gradual obsolescence. So the question is can they keep the pricing up on the older ones a few years down the line enough to fill out the end of those expected payback periods.
The real boogeyman for a neocloud that has heavily invested in expensive Nvidia hardware might be a variation of that beyond Nvidia with startups that have even more dramatic efficiency increases pushing the leading edge even further. For example, if companies like Mythic AI and d-Matrix could somehow rapidly rapidly scale, that would push prices down for all of Nvidia hardware that is significantly less efficient.
I guess so far it doesn't look like any startups with really big efficiency breakthroughs are even close to being able to scale like Nvidia though, especially with the manufacturing and power crunch. But I suspect some of that is because of favoritism and strong arming protecting investments rather than a free and fair ecosystem.
> So the question is can they keep the pricing up on the older ones a few years down the line
They don't expect to keep the prices flat over time, and everyone involved will have planned for this. Prices are highest when they're the newest and greatest (part of why it's valuable for neoclouds to be first in line for new models), and drop year by year as newer GPU models can do equivalent work at lower cost.
You can see a pretty cool dataset of this at [1]; H100 prices where $3/hr in 2023, and dropped linear-ish to $1.75/hr by 2025. And also the notable exception that prices are up this year due to shortage.
[1] https://semianalysis.com/gpu-pricing-index/
They don't expect to keep the prices flat over time, and everyone involved will have planned for this. Prices are highest when they're the newest and greatest (part of why it's valuable for neoclouds to be first in line for new models), and drop year by year as newer GPU models can do equivalent work at lower cost.
You can see a pretty cool dataset of this at [1]; H100 prices where $3/hr in 2023, and dropped linear-ish to $1.75/hr by 2025. And also the notable exception that prices are up this year due to shortage.
[1] https://semianalysis.com/gpu-pricing-index/
I don’t know if the circular financing is a problem. NVIDIA is the best my name in town, any company has to spend on NVIDIA assets for their compute. Now that makes NVIDIA rich and so they don’t know what to do with their money. They are just propping up companies they find interesting
Might be a blessing in disguise that these companies can't roll out datacenters as quick as they want (due to financing, power issues, permit delays or whatever).
That puts a cap on surplus (potentially unused?) datacenter capacity that's around by the time the AI bubble pops.
That puts a cap on surplus (potentially unused?) datacenter capacity that's around by the time the AI bubble pops.
It’s all fine till it’s not. Then it’s a gigantic financial house of cards that comes crashing down.
This bubble bursting will make us all poor
I have the same feeling, but after reading this report I believe big hyperscalers will survive the bubble when it pops.
Now I've got the feeling they don't have huge amounts of GPUs sitting in their DCs, but rented for Opex. In case the bubble pops they might get it at discount as CapEx (like Amazon did with dark fiber after the dotcom bubble).
Now I've got the feeling they don't have huge amounts of GPUs sitting in their DCs, but rented for Opex. In case the bubble pops they might get it at discount as CapEx (like Amazon did with dark fiber after the dotcom bubble).
Would this author prefer that Nvidia buy equity using GPUs directly? I don't think it actually counts as circular.
> I don't think it actually counts as circular.
It is. The GPUs go on to be used to get loans to then get more GPUs.
It is. The GPUs go on to be used to get loans to then get more GPUs.
I've said it before, I will say it again: all that circular investment, all the IOUs, all the billions of dollars of money that are floating around in the entire AI web... it will seriously wreck the US economy, the volume is orders of magnitude worse than what caused the 2007ff global financial crisis. But if OpenAI and Anthropic both manage to enter the fray as well and automatically get made part of the NASDAQ and MSCI World like SpaceX already did... yeah, then it will fry the US pension system alive as well.
>> the volume is orders of magnitude worse than what caused the 2007ff global financial crisis.
Nobody lives in GPUs and what was the ratio of equity/debt for the toxic assets in 2007?
Nobody lives in GPUs and what was the ratio of equity/debt for the toxic assets in 2007?
It looks more similar to the 1929 crash to me, where "too big to fail" blue chip stocks were overinvested and overvalued, and the value adjustments rippled through the rest of the economy. If NVIDIA does get a meaningful value adjustment downwards, it'll probably survive, but it'll impact the S&P500. People will need to sell off other stocks to cover the losses, etc. etc.
nvidia's forward p/e is 24. walmart's is 39.
That is exactly the point. These circular deals artificially increase the earnings of company and as a result artificially decrease price–earnings ratio.
That seems to indicate the market is more confident in walmart's earnings than nvidia's.
It's the earnings that seem unsustainable. The current rate of GPU buying can't last forever. And eventually competitors with lower margins will spring up.
484994949595(2)
All financing is circular. This concern is beyond the pale contrived
Financing is circular because creating a liability for one party (debt) creates an asset for another (the bank) off of which more debt can be secured
A bank / financier sells trust and reassurance. They otherwise invent most money from thin air.
Financing is circular because creating a liability for one party (debt) creates an asset for another (the bank) off of which more debt can be secured
A bank / financier sells trust and reassurance. They otherwise invent most money from thin air.
That’s not the point. The issue is that loaning/investing to a client so they can buy from you conflates your investments with your revenue.
It may be fine, or not. It it has been a frequent type of manipulation to obfuscate the real accounting situation.
It may be fine, or not. It it has been a frequent type of manipulation to obfuscate the real accounting situation.
Yeah, it's basically creating the illusion of demand and revenue. Lots of fraud in the past relied on companies "investing" into companies which then bought from the investor. I'm not sure to what degree this is happening now, though, and to what degree this is benign.
I’m not sure anyone can really say now, the terms and details are too opaque. But, given the history the opacity is itself a red flag.
People are investing because if Nvidia are essentially buying shares with graphics cards then they're motivated to make this stuff work. If the invested in company's share price tanks, Nvidia loses out, and I imagine quite a few people are willing to win or lose alongside Nvidia.
In general for these deals, and the ones with SPVs even more we don’t know. It may be as straightforward as equity for GPUs, but not enough information has been released.
Yandex, not nebius. Surprised how the world gets on kgb again and again, and again
Didn't they move to escape that world?
Yandex’s parent holding company was Dutch, and Nebius is now the same. A lot of their employees were effectively transferred between the two. Nebius is basically still just Yandex, just rebranded and legally a different entity.
It’s also by many accounts a bit of a weird company to work for, but they can afford to pay above-market for many roles.
It’s also by many accounts a bit of a weird company to work for, but they can afford to pay above-market for many roles.
You can even check their stock ticker...
Nvidia invested $2b into CoreWeave for 9% equity stake. CoreWeave is spending $35b in CapEx in 2026. Therefore, Nvidia's investment is only 5.7% of CoreWeave's single year CapEx. The other $32b is coming from other sources that isn't Nvidia. This is hardly circular.
Nvidia invests in Neoclouds because it's a hedge against hyperscalers having too much power, ie designing and prioritizing their own chips, and not fully using Nvidia's rack design. Neoclouds give hyperscalers competition. Neoclouds accept Nvidia investments because it allows them to secure Nvidia chips first, which is a competitive advantage since new Nvidia chips have been as much as ~5-20x more efficient than old Nvidia chips.
Nvidia was planning to directly compete against hyperscalers through DGX Cloud. They cancelled public DGX Cloud access when they found that investing in Neoclouds would accomplish the same goals without having to compete against their biggest customers.
If you're Nvidia, it's smart because Neoclouds that you have a large stake in will deploy your full stack from GPUs to networking to storage racks. They will share valuable usage data back to you so you can design a better next generation. Hyperscalers are likely a lot less cooperative, prefer to use their own designs if possible, and will guard their usage data.