Nvidia projects $673B in sales as AI demand widens(forgeeks.net) |
Nvidia projects $673B in sales as AI demand widens(forgeeks.net) |
Can anyone who actually understands finance please explain a couple things to me?
1. Are those accusations are true in a significant way, and are actually a bad thing?
2. This claimed $673B in sales, how much of it comes from NVIDIA's own money, invested into their clients? Is there any way to know that?
Still that's not to say these companies aren't leveraging the Nvidia capital with others' in a way that magnifies or multiplies some of the effect.
But it looks like a second order contributor unless Nvidia's actions are acting like a backstop that causes way more risk and leverage to pile up in a way that could come tumbling down
NVIDIA says it has invested nearly $50B in frontier labs. According to NVIDIA, “the AI labs for which NVIDIA expects to leverage its balance sheet should account for roughly one-quarter of NVIDIA’s business next year.”
To be clear, this doesn’t mean 1/4 of $673B is NVIDIA money.
Also if you think about this for anymore than a few milliseconds you realize that if Nvidia was giving away 90B to get back 90B in revenue, then none of the capex spend being reported by the hyperscalers would make any sense.
We know that Google, Meta, SpaceX, Microsoft, Nebius, CoreWeave, Amazon, are all buying huge amounts of Nvidia chips, with their own money!
This is all public info. The amounts of money Nvidia has invested in companies are tiny in comparison with their own revenue.
The amounts of “circular financing” are a drop in the bucket compared to, surprise, actual companies buying their product.
I am not sure if the distinction makes a difference, but the latter sounds a lot more reasonable to me.
They walk a bit more and the first one says, "You know, I gave you $100 to eat shit, and you gave me the same $100 to eat shit. I can't help but think we both just ate shit for nothing." The second one responds, "That's not true at all! We increased the GDP by $200!"
Nvidia is investing - it is using its shareholder's cash under the assumption it will create value for them.
1. Small models are rapidly growing in capability, require less compute to train and serve
2. There are more suppliers now, both in China & the US (OpenAI even have their own inferencing hardware now)
3. Memory still constrains how much they can ship in the short term
For me the important question is where the economy will be in the next 5 years. Because if the economy is doing well, I have no doubt the AI demand will continue to sky rocket. I don't think it matters to Nvidia how uses their compute, closed or open models. The win either way.
Which is just silly on the face of it. Data centers need concrete, copper, DRAM, SSDs, and labour. That alone will cost more than 33%.
Surely the land is a double digit percentage of their budget? I know they build in the middle of nowhere, but even then.
Anyway, according to the latest articles[0], 1000B is the lower bound.
[0] https://sg.finance.yahoo.com/news/ai-infrastructure-investme...
It's obviously hard to understand to people bad at everything.
Same with ASMl who are the sole company behind their machines.
As could millions of other companies.
This is an unprecedented circular debt gamble. The market price is not based on affordability but distorted by the seller.
SK Hynix (together with Samsung and Nvidia) claims $700B investment [1], Samsung itself is investing $70B, Micron $25B, Sandisk announced $31B today [2]
[1] https://asia.nikkei.com/business/technology/artificial-intel... [2] https://www.wsj.com/tech/kioxia-sandisk-to-invest-more-than-...
Or from individuals and smaller companies?
It’s mutual dependency unfortunately, and will remain that way whilst there are shareholders and investors who own the publications and seek only ongoing returns
In a game where the odds are stacked against you, the only winning move is not to play.
This is a financial advice.
According to Jevons' paradox a reduction in resource requirements (improved resource efficiency for the same payoff) leads to an increase in demand. This stops working when demand for compute is completely exhausted, but we are very far from that. There's even some very silly predictions floating around (see the latest Dwarkesh Patel podcast) that say compute will soon be most of the economy, even dictating market interest rates. Now, that has to be wrong, but the directional outlook is closer to correct than "very small and efficient models mean there will be ~0 demand for HPC-like compute".
Right now there are a small number of very very resource intensive use cases that are being subsidized by OpenAI and Anthropic. There are people generating millions of lines of code because it’s basically free at the point of use, despite the code producing very little value. Anthropic and OpenAI frequently “reset” customer limits to allow them to use even more resources at no additional cost.
The majority of use cases across business are not generating millions of lines of code per employee. The majority of businesses need just a little bit of automation to radically improve the way they operate. A software engineer making endless projects because it’s free to do so might use hundreds of billions of tokens per year, but an entire manufacturing business could be revolutionized with a few million tokens per year.
I think 2 things can be true:
1. There is very little penetration of AI across the economy and huge room to grow in the number of businesses deriving economic value from AI
2. The compute usage today is vastly overrepresented by usage outliers who are not paying the cost of their usage and will stop when forced to pay the cost
We could see AI usage 10x while seeing compute decrease 10x if the type of usage shifts. Most businesses just need smarter macros.
The PC era, call it 1975-2005, was one of the greatest wealth creation events in history, was characterized by the cost of the underlying commodity dropping mercilessly for the whole time. Each time it did, the space of problem you could solve with a PC would increase, to the point that by the end, they were both replacing mainframes and powering users who do nothing but chat and post cat pictures.
Could there be a correction in the short run? Quite possibly. I think an underestimated last mile problem is just the massive weight of bureaucracy and human process inertia. But in the long run, cheap, efficient intelligence is a new engineering capability that we've just begun to even explore.
1. Depending on the data source you look at, about 50 - 60% of people use AI at work but only for 5 - 15% of work hours. That leaves about 2x (from users) times 7 - 20x (from work hours) for growth. Furthermore agentic usage is much more token-intensive than regular prompts, that's another unknown multiple that will get applied.
Small models will make a dent for sure, but even they need to run on hardware. It's not clear how much their lower resource requirements will cancel out the scope for growth, but I think it will take time for that dynamic to play out; people are only just starting to ease up on tokenmaxxing. Anthropic revenues would be the canary in the coalmine, and thankfully they'll be IPO'ing soon.
2. All the relevant fabs (mainly, TSMC) are extremely capacity-constrained, so who actually gets the chips depends on who has the best vendor relationships... and who can pay the most for them. Even Apple, famed for its supply chain mastery, is having trouble these days.
I would assume TSMC will try to keep all its customers happy but will prioritize supplying the customer that will pay it the most money, and these days that's Nvidia. Simply because that's where ~all the AI boom money is flowing. Heck, you could even imagine some form of revenue share to keep the spice errr chips flowing...
3. Memory constraints affect all vendors, they will just pass those costs on to customers, like Nvidia with its recent 15% price bump. Notably the bump was announced BEFORE the earnings; I wonder if the effects of that was reflected in these projections.
Nvidia is in the same position with acquiring chip supply that Google is with acquiring search traffic: monopoly profits shared with suppliers make it very hard for other companies to compete.
Must be very clear that China’s undercut strategy, which is a well-known and studied tactic that they’ve used for a long time, it is absolutely dominating this point.
Right now you can LLM, code, make songs, images, and esp video on gaming hardware in your PC that would’ve been absolutely datacenter shit last year.
So the question will be does the scaling continue to benefit efficiency or ability?
If ability (needs datacenter storage and performance), how much better can the code get? How much more realistic in the images videos get? There are definitely strides to be made everywhere, but man, just like the bottleneck wasn’t coding, I’m not sure the creation bottleneck is rendering.
And it's pretty unclear what is included in the "capital expenditure" numbers. E.g. does it include training costs? does it include research costs? etc.
But most of us would be better off not gambling on high risk bets like shorting NVDA.
Jevons paradox is a veridical paradox, which, as you said, means that it's a true statement that merely looks wrong because it is counterintuitive.
I know that some people think that the word "paradox" should be only used to refer to antinomy paradoxes which are often called "true paradoxes" (such as "this sentence is false") which lead to a contradiction without requiring a flaw in reasoning.
Surely this applies to fixed-price subscriptions, not per-token spend? Large enterprises (the "very very resource intensive" large-scale users) have to pay per token.
Facebook is reportedly the company that spent $500 million in a single month on tokens. There are individual non-enterprise users rotating multiple subscriptions incurring $10k+ in tokens per subscription. Facebook’s $500 million month… is equivalent to ~10k individual subscriptions which could be as little as a few thousand of the heaviest users. That’s $500 million when billed on usage, or ~$2 million on plans.
The reason resets are such a big deal (people have set up websites to track them, tweets announcing them get millions of impressions) is because there are huge numbers of users pushing their plan limits every single day. If there was huge demand from usage-based customers (the large enterprises) that OpenAI and Anthropic couldn’t meet, they wouldn’t be handing out resets like candy.
I think a realistic belief is that Anthropic and OpenAI have vastly overstated demand and are using resets as a way to keep usage artificially inflated at a substantial financial cost. I’d guess fixed price plan users make up at least 95% of usage.
Even within one country, this should be 99% of what happens, onless it's an oil producer or something like that.
These circular deals are two paired transactions:
1. Nvidia buys equity in an AI lab or cloud provider with cash.
2. The counterparty agrees to buy X number of GPUs from Nvidia and in exchange Nvidia guarantees to rent some Y fraction of the compute if the counterparty cannot find customers.
This structure goes south during a pullback because all this liquidity Nvidia is essentially providing vanishes and contracts rapidly if the counterparty cannot find customers.
The other circular deal type is via private equity and the Special Purpose Vehicle (SPV).
1. The private equity firm loans money to the SPV.
2. The SPV buys GPUs from Nvidia for a data center.
3. Nvidia guarantees to the private equity firm residual value of the GPU which lowers the risk for the lender.
This deal also breaks down if the demand for GPU compute never materializes because now Nvidia is on the hook to the private equity firm (the lender) for the residual value of the GPU, which again saps Nvidia's liquidity.
Basically these deals are extremely sharp double edged swords. As long as demand for compute outpaces the compute capacity Nvidia can provide, Nvidia's revenues grow exponentially. But if demand growth slows, stops, or goes negative, Nvidia is suddenly on the hook for their counterparties' losses. Suddenly Nvidia's cash flow goes extremely negative and the company's financial situation becomes dicey.
Another thing NVIDIA does: when hyperscalers are looking for debt to build more datacenter capacity, NVIDIA offers to be a backstop in case the compute isn’t actually used. If we take CoreWeave for example, NVIDIA has ownership in it, and also sell them GPUs, and also goes to the banks telling them they will for sure buy the unused capacity as a way to reduce the bank risks.
So you can add that to the whole circular thing
On paper it’s “capital to spend on whatever helps them grow.” In practice, their growth requires enormous amounts of compute, so it’s pretty close to “Here's more money so you can acquire more compute [silent part: much of it from us].”
Now imagine adding a complex liquid cooling system to my house, upgrading the wiring to support the extra power draw, and some full time staff to maintain the space and keep the GPUs running. Wouldn't be hard to switch the balance back away from GPUs toward the infrastructure around them.
And let's also imagine I include my massive GPU usage bill in the house column. And while we're at it let's also include my full time staff of researchers in the house column.
This is just hyperbolic nonsense.
There is a desire from a certain group of people of make-believe - doesn't mean the 'demand' is actually real given the economics.
Source(s)?
I think this misses the actual limits here.
The problem isn't demand it's, "how much people are willing to spend on it".
Cheap AI has to be served on cheap compute, and if inference gets cheap enough to unlock massive usage numbers, by definition it also doesn't require anywhere near as much infrastructure per unit of demand.
Take DeepSeek serving ~100T tokens/day, depending on workload and utilization, you're potentially talking about only a few thousand last-gen GPUs. With current-gen GPUs maybe closer to ~1,000, and with Rubin even fewer I will be damned if I could get my hands on one.
That's the part I think people are missing when they extrapolate token demand into enormous infrastructure or AI revenue.
Yes usage will explode. But if the cost per unit collapses, the revenue doesn't necessarily go up with it.
You can't simultaneously argue that intelligence becomes so cheap that everyone uses enormous amounts of it, while also assuming customers will somehow spend trillions of dollars a year consuming it.
There is no obvious $1T customer-facing AI revenue number at the end of this rainbow in the short/medium term.
The average person isn't going to spend anything remotely comparable to what they spend on a car every year for an AI service. Even businesses have budgets now, huge demand doesn't matter if the willingness to pay isn't there.
The only path I can see to numbers like that is AI consuming existing business domains, even then it's very thin.
Say SaaS + legal + consulting + BPO + various other service industries collectively represent something like $10-20T globally.
Even if AI eventually replaces an enormous portion of that, it's probably not doing so at the same price. Why would customers switch otherwise?
Either the AI product has to be dramatically better, which is difficult for mature workflows, or dramatically cheaper which is much more plausible.
If it replaces $10-20T of existing services at roughly 1/10th or 1/100th (more likely) the cost, then you're looking at maybe a ~$1T AI revenue opportunity after replacing an absurdly large fraction of the existing service economy.
Who are now unemployed and can't pay for shit.
And that's before competition.
I think it's crazy to assume AI companies won't compete aggressively on price. As capabilities diffuse, smaller models catch up, inference hits pareto frontier the open-source alternatives have already improved and caught up, margins on routine intelligence should compress "hard" (emphasis on "hard").
We've already seen how difficult adoption can be even when the technology looks impressive on paper. Cheap here means 100x cheaper for 10x more demand that's a net 10x loss before any software or hardware optimizations.
So yes, I completely agree that cheap intelligence can bring an enormous amount of new usage.
"I just don't think usage means revenue." (you can plaster it on a wall if you want to, "usage doesn't mean revenue", if you want to find that out I have foss software bridge to sell)
The PC analogy actually reinforces this if you really think about it. Compute became "vastly more useful" while the cost per unit of compute collapsed. Society captured enormous value, but all computer companies are literal failing giants without the AI hype. Value got caught by people who provided productionization.
Now if people expect AI to self productize itself I am happy to tell your try it. We all saw how OpenAI fell behind Anthropic because they thought that would work...
Google couldn't productize the search, instead they sold the eye balls and web-real-estate. Maybe that's the AI business model, but that's not $1T worth given you need to unglue people from other stuff.
Unless we get something approaching genuine ASI producing so much additional economic value that entirely new trillions, I don't see a path to $1-2T in direct AI revenue from customers.
The market simply can't absorb that level of spending.
Demand can be effectively infinite at the right price. But I think people are delusional on HN and SF if they think that number is in Trillions like the investments seem to suggest.
I am not saying Nvidia will fall tomorrow but someone will have to pull the breaks before this car goes to hell.
Lol its not even that - its what can I do with it? Which eventually has to show up somehow in the financials - from a macroeconomic stand point. Software production is microeconomic.
This is the right way to look at it, but a few of your estimates are a bit off. AI is being sold as an accelerator (or, if you're in a dystopian mood, total replacement) of knowledge workers. Currently knowledge worker salaries are $50 - 70 trillion a year globally, $10 - 11T in the US alone: https://gist.github.com/danielmiessler/2dc039762a202b083753b...
> Even if AI eventually replaces an enormous portion of that, it's probably not doing so at the same price. Why would customers switch otherwise?
AI is wayyyyyyy easier to wrangle than humans; no sick leaves, health insurance, perks, HR issues... heck they don't even sleep! If companies could replace us with robots, they would do so in a heartbeat. Capitalism!
So in a "what the market will bear" sense, we have an upper bound on the TAM. Indeed, I expect this is where Anthropic's ridiculous "$30 trillion" number is coming from... except now we see how they came to it.
If AI makes workers even 1% more efficient, that's a $500 - 700 billion value annually. In reality AI makes workers way more efficient (studies from the ancient era of 2024 showed about a 30% boost) so AI companies could realistically charge that much more. But then all the other factors you mentioned -- smaller models, competition, self-hosting, etc -- come into play, which put a downward pressure on revenues.
It's impossible to predict how these dynamics will play out, but the numbers involved are astronomical. This is why everyone from the frontier labs to Big Tech to VCs to nation states are scrambling to get in on it.
(it is sage advice, but reminded me of nuclear war)
In general this is true. Participation is not rigged with eg VTI/ITOT and VXUS/IXUS and a long enough time horizon.
There is no case where one cannot participate - doing nothing means inflation will eat away at assets.
I have a very brutish 50/50 international/US split (was 30% international before 2024 when Trump promised to destroy the US economy and started to act on that...). Each quarterly equity vest I put more in, and realize capital "losses" while buying the near equivalent security when there is an opportunity to do so.
Instead of blaming others, improve yourself and your station.
Life is too short to spend it putting up road blocks. Create a new, shorter circuit path instead.
No capitalist conspiracy theory is needed, save the conspiracy theories for the Trumpers.
The boring staid firms like Fidelity/Vanguard have vastly superior products not designed to incentivize gambling
All articles that I found on "ai datacenters cost breakdown" say that >60% is for the HW inside.
Yeah bro, just learn to code.