Hacker News Reader

Comment by Silfen | Hacker News Reader

Reading this thread makes me realize that seemingly no one has heard the basic, mundane case for why Nvidia would want to do so much "circular financing" as a matter of totally normal business strategy. This may not be the whole story, but it's worth understanding.

Because of bottlenecks in the supply chain, Nvidia knows it is going to sell 100% of its inventory. They could raise prices quite a bit more, but then their only customers would be hyperscalers (building competing chips) and possibly OpenAI and Anthropic. Nvidia is willing to forgo present day profits for the strategic value of keeping its customer base diverse and promoting the network effects of e.g. CUDA. Instead of quoting a lower price per GPU to neoclouds and startups, they finance purchases of chips, but the two are roughly equivalent (financing means Nvidia takes on more risk, but has more upside and locks in future demand). It is in their commercial interest to have many competing buyers of their chips.

Obviously this creates fun mechanisms for fraud and wacky accounting (and I'm agnostic as to whether that's happening here), but it's worth knowing about the boring/obvious story before donning your tinfoil hat.

View on HN

>In Berlin during a conference, Huang, accompanied by his top sales executives, broadcast a new message to a half-dozen CEOs of companies that are major buyers of Nvidia chips who were seated around a circular table. He opened with a story that sounded like a fable, according to people who attended. Huang had once found a soup vendor in Japan, he said, with a long line of customers. He joined the line, and after a long wait, enjoyed one of the best bowls of soup he’d ever had. When he asked for another serving, the soup-seller told him no—he had too many customers waiting and not enough soup for all of them.

“He’s a better businessperson than all of you,” Huang told the cloud computing CEOs. The soup-man doesn’t sell out all his inventory to one or two customers, but instead keeps prices high and his customer base diverse, the CEO said.

“You are not establishing the value of my products,” he said.

When you ration you keep msrp prices low, and a secondary market emerges. The soup guy is keeping prices low, below what market would clear

Soup man wants to make soup, why do you assume otherwise?

very difficult to trade soup in secondary markets.

Yea, but in 20 years the secondary soup market will explode due to AI.

no soup for you!

You mean to say Jensen Huang is the Soup Nazi? - https://youtu.be/RqlQYBcsq54

And is on his way to Argentina it seems.

Hehehe my pet theory is that Jensen was using exactly the opposite strategy to secure that “avalanche of chips and wafers” from TSMC: Nvidia’s monopoly cash flows give it the volume and money to secure its massive supply-chain advantage by buying priority amongst TSMC’s customers.

That is, Jensen is going to the TSMC soup seller and saying “I’ll buy 80% of all your soup for twice the price because people are willing to pay 10x that with my ramen noodles in it.” And TSMC being the real bottleneck in this industry, knows there’s really nowhere else for its other customers to go. It of course, wants to keep all customers happy… but some will get much more soup than others.

This is probably a huge factor in keeping their advantage over their customers and competitors (but I repeat myself.) That is, TPUs and Trainium etc simply cannot acquire the necessary volume of chips at a competitive enough price-point to put a dent in Nvidia’s grip on the market. But the hyperscalers need all that compute NOW NOW NOW, so they have no choice but to turn around and begrudgingly pay Nvidia for that supply. Which puts more cash flow in Nvidia’s pocket, which buys them more soup, and on this loop continues.

There are a couple of places like this in LA, where I used to live. Lines would extend around the block and people would join the line just to see what the fuss was about. As a business move it is very smart, but as a customer it has never been worth it. The food is very often average and I could never work out why it was so popular.

> but as a customer it has never been worth it. The food is very often average and I could never work out why it was so popular.

..did you mean, as a customer it has never been worth it to you, or in general, for anyone?

> The food is very often average and I could never work out why it was so popular.

IMO:

FOMO to pull them in combined with the fact that for many people it is really easy to convince themselves it was all worth it after they committed to waiting so long.

Otherwise they'd be fools for having done it and most people prefer to not notice their own foolishness.

London has (or at least had some years ago) something like this, the fried-chicken equivalent of the Soup Nazi. A long line and... OK chicken. Never saw what the fuss was about, maybe people went more for the experience.

"No soup for you!" SAInfeld just got weirder.

> Nvidia knows it is going to sell 100% of its inventory.

The WSJ article is basically saying the opposite. The hyperscalers are making their own chips and they are buying less from Nvidia.

https://archive.is/20261113091227/https://www.wsj.com/tech/j...

This is my comment in another part of this thread

https://news.ycombinator.com/item?id=50493251#50496483

> Then he went on to describe the situation the company was in. Major cloud providers were running away from Nvidia, making plans to use alternative chips and lessen their reliance on the firm. Even worse, Huang said, the vast supply of chip wafers and memory Nvidia had committed to buying were coming at him like an “avalanche.”

Jensen's son-in-law was making these vendor financed deals and Kress the CFO now interim CEO was against them.

Large customers like Nvidia and Apple want billions of chips. TSMC can't just ramp up and down instantly. It takes 3+ years and $30 billion to build a modern fab. So TSMC has contracts where you are commiting to buy the wafers for years in advance.

I've worked for multiple fabless chip companies that use TSMC to manufacture. Amazon, Google, Microsoft, and the other big hyperscalers are making their own AI chips to move away from Nvidia. The startups like OpenAI are doing the same so the real demand for Nvidia chips is dropping. But Nvidia can't easily get out of the contracts with TSMC to manufacture and buy the chips. They would probably have multi billion penalties that would show up on their financial statements.

Some people think Nvidia is acting as a VC fund for AI startups but my view is that Jensen freaked out about the dropping demand for Nvidia chips. So he is funding these startups with circular finance deals to buy the Nvidia chips that he can't sell to the hyperscalers anymore.

Do you have the link to the WSJ article. I would love to read it.

It's linked at various places in the thread but here it is

https://archive.is/20261113091227/https://www.wsj.com/tech/j...

The Hyperscalers are often working with a company like Broadcom (the elephant in the room on this topic) to design said chips. They don't usually have the amount of ASIC and Networking and Optics hardware expertise in house that would be required. See, for instance, Google's TPUs. Broadcom's hardware people are world class and some of the best paid in the business. It's the most elite hardware company in a lot of ways in terms of total comp for top engineers.

Google hires some hardware people to come up with general requirements for specialized stuff like ASICs, then sends those requirements to Broadcom to do the hard parts and negotiates big multi-billion dollar contracts with them. OpenAI did the same and contracted Broadcom after liking what they saw with the TPUs. There are several other Broadcom customers that I'm probably not at liberty to disclose.

But no, as far as GPU replacements are concerned, the Hyperscalers generally are not experienced enough in the specific types of hardware required to do this by themselves. It's very specialized and boutique high price work and more difficult than say, designing an ARM chip (something a company like Amazon or Google could probably handle in house).

In general though, yes, I agree. Hyperscalers are trying to reduce dependence on NVIDIA. Many are quite far along in their ASIC projects (Google especially, they are on their 8th generation TPU! Way past just being a small scale experiment by now.) and have large volumes of them running in place of NVIDIA chips and much more efficiently already.

Broadcom is fabless, and uses TSMC for these sorts of chips. They are mostly a boutique design "mercenary" firm for exactly this sort of specialized hardware for Hyperscaler customers and other clients that run significant amounts of infrastructure in house (so, think JP Morgan for instance). That's the core of their business, along with increasingly Infrastructure Software in the form of virtualization and mainframes and private and hybrid cloud (so some natural sales overlap there, just generally, "big companies that need specialized big data center hardware and software").

I've been in the chip design industry for 30 years now and know tons of people doing this. The hyperscalers all have their own chip design teams but they have multiple chips (server CPUs, AI accelerators, complex multi chip module packaging, and more) so they can't do all of them in house. As you said they partner with multiple companies to help develop them.

I see you added to your comment. I have a bunch of former and current coworkers from Broadcom. I'm in digital physical design taking the RTL through synthesis and place and route

Yep, I'm ex-Broadcom. I removed it cause we get a lot of bad press whenever we fire smaller clients, but I enjoyed my time there and thought they were very well run. I doubt I'll work at another hardware company that operated like them again, it was really kinda special. Very flat hierarchy in a good kind of way. Perhaps if I look for work in the startup world.

And yeah I agree, it would be impossible to not outsource some of this stuff given the amount of hardware the hyperscalers need and how specialized some of it is. Even though the hyperscalers certainly have competent hardware teams now, there are some specialized things that they are just going to have to work with external partners on. Hardware can get very specialized depending on what you are trying to do. And so boutique consultancy and chip design "mercenary" firms are always going to need to exist even if the Hyperscalers have good hardware teams in house (they need them to be able to work well with those of us outside of the house).

Also, I love your HN user name !

and there's some increasing specialization in terms of chips for inference vs chips for training, and if you only need the big costly nvda chips for training, then nvda needs there to be a huge long tail of investment into models, not 1 or 2 dominant frontier model companies

You reminded me of the interview of Jensen with Dwarkesh. Dwarkesh was pushing the point of efficiency, and all that Jensen could say was "machine learning is like a cake". I was puzzled how poorly he defended himself.

So could the smuggling chips to China (and based upon a latest GamersNexus video chips are showing up in recovered Russian drones), be the thing that’s holding the whole thing up?

Is this Jensens “Delorean cocaine” ?

When you go into a Ford dealership there is a good chance that you will buy your truck with a loan from Ford credit. That probably helps keep the price of Ford trucks high and Ford makes a lot from loans. And there are problems today with vehicles being repossessed too much because of the long loan terms.

‘Circular financing’ sounds sinister but customer financing and vertical investments aren’t strange or uncommon. It could be bad for sure. But it is not like they invented a new form of wickedness.

Its not that circular financing is sinister or illegal, its that its very risky. If the startups that are purchasing the product are also getting a loan from the company making the product and that loan is also backed by that same company then if that startup is unprofitable and has massive debt at interest rates like 15% or has near junk bond ratings then the company that built, financed and backed the products is on the hook for it all.

Sure they have customers pounding at their door for more product. At some point the hype machine stops and new neocloud startups can't get funding because its too expensive or banks don't trust them. Then what happens? The circular financing starts to unwind itself and NVidia is the "bag holder". If you look up Lucent or Nortel this happened to them during the dot com boom. There is a positive difference here with NVidia, they have so much demand and usage this can probably go on for a while. Lucent and Nortel were bit by dark fiber and the usage/demand just wasn't there.

If Jensen at the helm of NVidia is bullying/convincing companies into purchasing more than they need, financing it, etc. and creating the hype machine himself and NewCEO ain't going to play that game, then it could be an end to the circular financing.

I haven't checked what companies exactly are they funding, but IMO that's the easiest way for Nvidia to have a high stake at a real future competitor for OAI or Anthropic, in whatever form, new proprietary or open sourced models. Besides the training data, compute is the biggest barrier to entry to this market. So bet on a few talented teams, give the tools, and see what happens..

> I haven't checked what companies exactly are they funding

Parent said "startups".

It is easy, but closer to "going to the gas station and buying lottery tickets"-easy.

NVidia was financing up to about 25% of purchases -- less than its margin. It's true this might be more risky than other sales strategies, but not enough to create a moral hazard of selling oversubsidized deals.

I agree with you completely and this is the real problem. Somewhere people were accusing them of accounting fraud, or this is because of the perils of effective altruism, or .. pick you poison. Capitalism is a great system but it carries within it the seeds of its own destruction.

It's not uncommon for dealer financing to be a loss leader. I've never used dealer financing that wasn't below 2% APR

>dealer financing to be a loss leader.

"dealer financing" is an umbrella term for the dealer's finance office acting as a "loan broker" and shopping around the customer's credit profile to find any bank willing to lend the money. (When the customer fills out the credit application at the dealership, the dealer then submits the info simultaneously to a list of potential banks.) The eventual underwriter for the loan may be the auto manufacturer's finance arm (e.g. General Motors Financial subsidiary or Toyota Motor Credit, etc) ... or ... it may be a non-automaker bank such as Chase or Capital One. Whichever one the customer ends up with depends on the particular credit score.

None of those financial entities are making car loans as a "loss leader". It's the reverse situation where both the dealer and the underwriting bank make money and the interest rates are high enough such that the banks pay kickback commissions to the dealer for funneling the loan to them.

> I've never used dealer financing that wasn't below 2% APR

Interest rates (such as "0% APR") that are artificially lower than prevailing market rates means the missing extra interest is embedded in the sales amount in the form of a higher price. TNSTAAFL.

> Interest rates (such as "0% APR") that are artificially lower than prevailing market rates means the missing extra interest is embedded in the sales amount in the form of a higher price. TNSTAAFL.

A more accurate way of saying this is that the dealer is using incentive money to buy points (as is available in the mortgage market). You can ask dealer to apply that to the price of the vehicle instead. Many dealers already offer that kind of choice in their advertisements already (cash bank or no interest financing). Which way works out better for you depends on the terms of the loan etc. It's all the same to the dealer though.

With below-market interest rates (0%, 1.99%, etc.) it’s almost always the manufacturer that is buying the points (if you want to look at it that way), not the dealer. Therefore, the dealer has no incentive to give you a cash discount in lieu of the lower rate, unless the manufacturer itself offers that trade off (which as pointed out, they sometimes do, but often do not).

This is a good sub thread for a lot of people buying a car, who haven't before.

Another trick in the industry, is the "price the dealer pays". This is mischievous, as the dealer does indeed pay that price for the car. However, they receive incentives quite often, in terms of kickbacks for sales made. So they dealer can gleefully say "This is what we paid for the car", whilst in parallel not indicate "Well, we received $5k as a kickback from the manufacturer later".

The manufacturer often also has monthly blitzes, discounts of the month, which the dealer will not really inform you of. After all, that's extra profit if they can trick you into paying full price.

In the early days of the internet, there used to be excellent services to discover all the discounts the dealers had, so you could a proper assessment of their actual cost. Thus, you'd be able to haggle with both eyes open.

For Canadians, The Automobile Protection Association (APA), https://apa.ca/ is legit, and offers such a service for a fee. Not sure what Americans might have. Or who to trust there.

Knowing what the car actually costs the dealer, is immense power when negotiating.

> This is a good sub thread for a lot of people buying a car, who haven't before.

Isn't the most economical option buying a used car? Less than 2-3 years old, initial kinks under warranty have been sorted out and you get a discounted rate beyond the retail price with barely any milage.

It can be, depending upon tax and depreciation laws in your area. EG, I run a small business and can deduct depreciation for business use. I have a full warranty too.

But one point too, if no one buys new, no used cars exist. And if too many buy used, their price goes up.

But really, my comment was just focusing on new, and how to get the best deal therein.

Yes the incentives I mentioned come from the manufacturer. My experience has been it's cash back (to the buyer via the incentive) or the special finance rate. For my last loan it worked out better to take the cash back so that's what I did.

Sure, but that's actually not necessarily the norm. Or even when a cash option is offered, its much less than the value of the below-market financing (like $1K-$2K vs. a 3% below market interest rate on a $50K car which is worth closer to $4K).

Dealer margins on initial sales aren't huge, usually, and there are times when 0% financing is simply a way to get things off of lots even at a loss so a car with better margins can replace it, but either way dealers don't really care because the margins on service are massive.

It's very uncommon for car and truck loans in the USA to be a loss leader.

Ford is lending money to people with jobs and credit history--startups don't have the business equivalent of either. One can also argue that while the collateral (cars) does depreciate, they'd don't depreciate as quickly as GPUs.

Gpus stopped depreciating

The 5090 I bought almost 2 years ago for 2k sold this fall for 9k. stopped depreciating indeed.

Very true. I wonder if that's the case for data center cards running 24 / 7.

Except, Ford is probably just a middleman, those loans are probably resold immediately to a financial institution, so it's not Ford's money. Then there is the lower portfolio risk. Thousands of customer owing 50 thousand bucks is not the same a 5 customers owing a billion bucks

While your explanation of in-house financing for cars is correct, that's not what the article and original title of this HN post is describing.

In your example, random customers are shopping at Ford.

In the article, Nvidia is creating the customers, and those customers are then engaging in the scenario you described with the automobile financing.

Nvidia is doing what you describe...plus they're artificially creating the customers who are buying from them, to make other customers want to buy from them.

So it's everything you describe plus a whole additional abstraction of AI...artificially in-demand

Sounds a bit analogous to advertising [creating customers]

Sure, if running an advertising campaign resulted in new human beings (who previously did not exist) with new bank accounts and new credit cards to buy the advertised product.

Literal bot has entered the room. I have a fast food chain. I give an applicant a loan to start their franchise, which wouldn't exist if I didn't offer that financing. The franchise is a customer of the chain and wouldn't exist without the loan. New franchises make other potential franchise owners consider working with that chain. That also happens and it is neither illegal nor immoral.

Not saying its good! Just OK under US law and sometimes help companies grow.

I think what makes this tough to be objective about is there is so much about this that is tied to things people hate: AI, tech, obscene wealth, poor regulation. I 100% agree but I do doubt the demand for AI is artificial.

Ford Credit is transparent about having financed the sale, nVidia books it all as sales (i.e. as if the customer had the money to buy the product) and is very quiet about how their 'investments' in the customer are actually financing it. It all looks good on paper until the music for this little dance stops.

That's a fair point and also a concern for SPVs at public companies. I would vote for tougher accounting regulation of things that are effectively on your balance sheet.

Working with a (comparatively small fry) customer who has a $5M grant to buy HGX B300s, current lead time is 37 weeks. This is for a chip which has already been superseded by the R100. NVIDIA has customers lined up around the block, including smaller shops that don't get special treatment.

I don't deal with their big customers, but the idea that NVIDIA doesn't have customers ready to spend $$$$ to get some GPU in their datacenter is proveably false. It's likely easier for them to handle big transactions for big customers, but there's also a loooong tail of customers in line for a smaller piece of the pie.

37 weeks is an entirely normal lead time in the semiconductors world. There are many sub-$10 components with similar lead time.

It makes me grateful that given that kind of supply chain, I can as a consumer more or less go buy anything I want on demand at a personal scale without a 37 week commitment / lead time.

Yes the prices are eye popping at times, but with money in hand newegg claims delivery by about dec 8.

LLM models are proving math theorems no humans could in ways no humans could. These GPUs can literally run that intelligence level and as humans we have an endless set of problems we need to work on.

Many people are deploying and using this new technology for very productive ways. That’s why Anthropic revenue is so impressive.

There is a very large cohort of reasonably well educated and articulate humans who are having problems processing this situation in some deep way that has them in denial and constantly insisting there is a massive bubble or something fraudulent or broken, of course there probably will be to a certain extent, some overshooting. and they will claim “see we were right” but even if that happens there is incredible rapid progress that is real, tangible, backed by real money and spending, and real productivity.

Anyway, going back to my desk now and steering LLMs on a code base created in 36 hours that is equivalent to 10-20 man years of work just 4 years ago, but sure “it’s all a bubble”, and like you said tell that to everyone in-line around the block who knows what’s really happening and wants more of this hardware which can do this.

Yes, AI and LLMs are useful tech.

Yes, there is massive fraud and circular financing happening around it, and the above is just one of three ways it’s happening.

Is it "circular financing" when american Honda finance corporation gives you a loan on your new civic?

Probably not but if they granted buyers with no income 0% interest for 20 years then yeah? Sales would grow through the roof. iow it can be essentially fraudulent, depending on the quality of the loans.

Which ai provider has no income?

If they’re doing it at the scale of subsidizing or offering favorable credit to businesses, like rental car fleets?

Sure, that’s a comparison.

Right, the snag is not that OpenAI and friends haven’t delivered the moon: it’s that they need to deliver at least another ten moons or so to in order to fulfil all of their promises and commitments.

Right! The bear case should be that in the long run, matmul gets commoditized and nobody will pay a premium for Nvidia.

To argue that demand for their chips & systems now is fraudulent is coo coo Cocoa Puffs.

I am concerned about the long term impacts on the economy when more and more people lose their jobs from Ai and automation. What happens when the middle class disappears completely and no one has money to buy anything? Who are all these companies spending all this money on Ai compute going to sell to? Demand dries up, then demand for Ai dries up, and then pop.

Also, how do we know at what point we'll have too many gpus online from all of these datacenter build outs? They are just assuming demand will continue to grow but what happens when supply becomes higher demand?

And at what point is a model good enough for most work? If someone releases an opus 5.5 level model as open source, does the whole equation flip and does everyone go to just running a local model?

Why wouldn’t the middle class have any money to buy anything?

World War? Investments wiped out by speculative financing on data centers and inference deals? Global financial restructuring?

They lose their ok paying jobs and are forced into minimum wage jobs because Ai automated away their job. There is already massive inflation and regular people are struggling. This is why some people spout on about minimum basic income but the reality is that our politicians are too captured by business and special interests to implement the needed taxation to implement a program like that.

Interesting that the main thing you can say about your codebase is that it took exactly 36 hours. Not a few days, like most people would say (because they would be focused on the product and not the exact hours). And nothing about what is does or whether there is even one customer that is using it.

I’ve started thinking naturally more in hours of autonomous work in implementation situations. You’re correct in that it’s not a full measure because it took a lot of iterations in design work before kicking off a specific goal pursuit for implementation only succeeded because the LLMs had clear instructions they could work autonomously on incrementally. The end result is still a work product that was produced in 36 hours, so less than a couple of days, and is of high quality and while hard to be precise I’m confident 4 years ago probably would have taken a team of 10 average developers at least a year of development time.

+1 for "steering LLMs on a code base"

The fact that the tech is useful doesn’t mean there’s no bubble. The web was and is useful too but there was nevertheless definitely a bubble around that at the turn of the century.

For companies like Anthropic a big risk is commoditization—the risk that open weight models get good enough that a lot of people don’t need an expensive subscription.

> There is a very large cohort of reasonably well educated and articulate humans who are having problems processing this situation in some deep way that has them in denial and constantly insisting there is a massive bubble or something fraudulent or broken, of course there probably will be to a certain extent, some overshooting. and they will claim “see we were right” but even if that happens there is incredible rapid progress that is real, tangible, backed by real money and spending, and real productivity.

While you are right, I think the elephant in the room is that they are rightly concerned about the ethical issues, namely the ecological and economic impacts.

I don't think anybody is concerned about the ecologic impact. I think theyre worried for their own livelihood, and ecology is something that they believe is a persuasive argument (it's not).

The collapse of the dot com bubble was not because the internet was actually useless. And despite the internet now being critical to the modern economy, Worldcom committed elaborate fraud that began during the peak in 2000 to mislead investors about their profitability to keep their stock price up.

Other than tulips, some of the most famous historical bubbles were similarly about hugely transformative tech that legitimately changed the world. That same feeling of certainty about future impact you get using LLM tools was why those technologies inevitably became a bubble.

Recently, I encountered a vendor with a near monopoly. I learned that they’re priced reasonably for new market entrants out of fear of monopsony.

monopsony - a market structure in which a single buyer substantially controls the market as the major purchaser of goods and services offered by many would-be sellers

Also, it's basically putting your profit into an investment (the loan that will have to be paid back) while keeping the inventory going and prices rising

>they finance purchases of chips, but the two are roughly equivalent (financing means Nvidia takes on more risk, but has more upside and locks in future demand).

The other upside is having that equity be counted as capital appreciation in their balance sheet. Even if it's essentially paper loss due to volatility.

Could we also include national security? The people of the US really want to avoid losing to our competition because they are forced to develop their own cutting edge chips. We need the world on our petrodollar and the world’s best AI researchers on our chips. Jackson Polloc is famous largely because the American government created false demand by driving his auction prices up for national security reasons. Lutnick has even been honest, about that. In this case, NVDA and US have aligned interests in NVDA’s dominance, and it seems like being given more leeway to engage in creative if not entirely kosher business practices might be part of a coordinated plan with the current administration.

Or Nvidia was trying to push up the marginal price of the GPUs paid by hyper-scalars/frontier labs and now has to deal with those teams moving off the platform due to cost.

It’s hard to talk about AI economics while looking at NVidia’s margins.

What the WSJ reported is that Nvidia does not know it is going to sell 100% of its inventory, and these deals were seen as a way to ensure strong demand despite cloud providers seeking to reduce their reliance and an avalanche of Nvidia's own purchase commitments. Perhaps that's wrong! But if it's right then I'm not sure there's a good way to square it with the boring/obvious story.

Hmmmm, New Allegations, very likely Sherman Antitrust Laws which involves Price Fixing & Collusion

The Sherman Antitrust Act (Section 1). Did Nvidia allegedly engineered scarcity by funding AI startups to buy its GPUs ?

What is the point of the opening part of your first sentence, "Reading this thread makes me realize that seemingly no one has heard..."? Why do you need to take a snipe at the commenters so far before making a perfectly valid point? You could have just made your point.

And it's totally fine if someone hasn't heard it before. That's why we read comments in places like HN. To learn.

> the strategic value of keeping its customer base diverse

That's one way to say, "maintain a monopoly where everyone needs you and you depend on no one."

Well funded hyperscalers are trying to reduce dependence on Nvidia. And Nvidia is conversely trying to push into cloud with these credit backstop/revenue sharing deals.

It does make them even more leveraged to the bubble though. If a neo cloud collapses, they're lose customers and gain debt collectors. It's a sound business strategy until it isn't. It's not good risk management, but then you don't become the world's largest company by making safe bets.

> It's not good risk management

Making a large number of very risky bets on small startups is how venture capital works. Most will fail, hopefully a few will make up for that.

Yes but normally you would want these to be uncorrelated bets.

Ding ding ding. One bet might work, but if it's debt and not equity the upside of a one success is just getting paid back.

Nvidia isn’t taking on any material debt to finance all these deals. They have the cash on hand to cover pretty much a complete collapse of every neocloud they’re currently backstopping.

So yeah if it goes tits up, they loose customers, but they’re customers that wouldn’t exist otherwise, and they can easily pay to make the debt collectors go away. But if it’s anything except a complete collapse, then they’ve grown the customer base for their products long term, and hedged against hyper scalers replacing them with custom silicon.

At the moment it looks like the downside risk is limited, and in the worse case, tolerable, whereas the upside is unbounded. Those are really good bets to be placing, the only trade-off is opportunity cost. But even there, what other opportunities could Nvidia be deploying their cash into, beyond share buybacks?

Amazon, Microsoft, Google, they all have a broad range of products and services. Google is probably the least diverse, relying too much on search, which makes their business riskier, and it's (probably) why their p/e multiple has always been the lowest of the tech megacaps. Nvidia's multiple is on the low side right now, so perhaps the market agrees that their risk is too concentrated.

> they can easily pay to make the debt collectors go away.

Sure, if it didn't come at the same time as all their other customers pulling back, and lots of other debt collectors simultaneously knocking on the door, and a spike in the cost of capital (read: stock starts crashing).

Keep in mind that NVidia has 75% gross margins, so any loan they backstop has an enormous amount of profit margin embedded in it. Even if they only recovered 50 cents on the dollar, they still come out ahead.

It's a good hedge on diversifying customers, but a bad hedge on exposure to fraud. I'm going to go out on a limb and say that Jensen exposed himself to too many frauds here for me to find much of a silver lining in any of it.

> Nvidia knows it is going to sell 100% of its inventory

Isn’t the entire problem here that they don’t know, and they haven’t?

Does that 100% include the gear that gets stolen in transit?

'Stolen', meaning 'delivered to wrong address'?

I mean, the wind can start blowing the other direction. And they can go overboard on the deals. But in the past few years, yes Nvidia has predictably sold out.

It's obviously impossible to tell without more insight, but the large AI labs are very much out compute and could use more of it, or they wouldn't be renting DCs from their competitors.

The neoclouds ensure there will be consistent demand of Nvidia GPUs even after everyone big moves to their own custom ASICs

Out of compute or out of chips? I swear I heard somebody saying Amazon had a whole datacenter worth of GPUs not even pluugged in sitting in storage because they were pressured into to purchasing them early by Nvidia and datacenter contruction was significantly delayed.

But that also just suggests* that the demand for nVidia’s products exceeds supply. The reason that nVidia was able to pressure people with take-it-or-leave-it deals (if that’s what did happen) is precisely because everyone involved expected (and correctly, so far, as it turns out) that there would be plenty of other customers ready to take all the places in the queue.

*EDITED: ‘shows’ -> ‘suggests’

Only if there was a convenient online market place where they could list them for sale, and some fulfillment centers that could deliver them to customers.

Running out of compute isn’t the same as “can buy more GPUs and use them”, else AWS and MSFT wouldn’t have billions of dollars of GPUs in warehouses that they have no datacenters in which to rack them.

GPU supply isn’t the constraint; datacenter space is. And that means Nvidia might not normally be able to sell all that they’d like even with seeming high demand.

I have customers begging, pleading for more GPUs. There is still strong demand.

Yeah now that they have engineered artificial scarcity.

How can we even talk about this without talking about price?

For AI data center chips, they pretty much are constrained by how much they can produce (TSMC is fully booked and RAM is scarce / expensive).

They are constrained by the chips they can power, at least in US

Interesting how much energy is expended here finding excuses for this unusual and borderline fraudulent strategy from Nvidia.

Is it really a conspiracy theory to think this level of circular financing is suspect? Prominent journalists and economists seem to think it is suspect.

Thank You. I stated something similar in my other reply But at least this gets the top vote which is most important. And I hope more people read it.

And I am just happy this being top comment shows there are silent majority upvoting and sanity prevails.

And it pisses me off the media are all trying to spin it as fraud.

People believe what they want to hear, and the media sells the stories the market demands. I believed this whole AI thing was unsustainable and a giant bubble for a long time. Then I ran numbers, and I actually tried to buy these GPUs. It's impossible. I've had unfulfilled orders sitting for months. The demand is obscene, and for good reason.

Not sure that'll work though.

Google and Amazon at least are making their own TPU silicon: Google TPUs, Amazon Trainium I think (goofy name), etc. They're big enough to make their own chips.

Apple is putting their own in-house AI accelerators into their laptops, desktops, and phones for edge inference and the local inference market.

Meanwhile AMD and Intel have GPU offerings. AMD's chips are probably as good as Nvidia, but their software is not... but software is exactly the moat that is vanishing due to AI. Intel's GPUs are mid but are priced aggressively. I actually picked up an Arc with 16G RAM for a very reasonable price to run local experiments and it works great... and I was able to fix the software rough edges with the Intel Arc with Claude Sonnet 5.5. "Hey Claude, take a look at this code... it's not performing well on this system's GPU. Please make a plan to fix..."

I don't think Nvidia's lead is as big as it might think it is.

You're only thinking about hardware. Most GPU applications are written using CUDA, which only runs (well) on Nvidia hardware.

AI can write great code, but the competition's hardware won't succeed until someone comes up with a CUDA replacement.

This was true as recently as one to two years ago. I’ve already used things like Opus 5 to code things harder than porting CUDA code.

Things like JAX and PyTorch already have good support for AMD, Intel, Google and Amazon TPUs, Apple Metal, etc.

Software compatibility moats like that are dead, dead, dead. Anything can be ported unless you don’t have the source. Pretty soon that won’t matter. Some of the progress porting binary games suggests maybe it already doesn’t.

People are still out of touch with how powerful this stuff is.

> Nvidia is willing to forgo present day profits for the strategic value of keeping its customer base diverse

That is precisely the opposite conclusion to the linked article though! They aren't "forgoing profits" if supply is constrained. Increasing demand increases prices and thus increases[1] profits. Which of course is exactly what we've seen in the market.

Your point would only work in an efficient market where they had competitors who would take the sale otherwise.

Basically: trying to sell this as NVIDIA being a Good Market Citizen and making sure everyone gets a chip is... well, it's kinda naive. No, they're really obviously making an extra buck on the market they already cornered.

[1] Superlinearly even, since the companies they fund can drive further investment based on the whale they landed.

Oligospony avoidance seems quite strategic.

Care to elaborate?

Monopsony is the state of affairs when there is only one buyer in a market, so oligopsony would be when there are only a few, big buyers.

Honestly wether the AI ecosystem created market bottlenecks by accident or intentional is irrelevant. The question is wether they exploited the situation or tried to mitigate. Observation mightily supports the former.

You should call this belief "Effective Altruism: Founders Edition"

What you're describing is the belief that they'll always sell their GPUs for a profit, so it doesn't matter what rules of accounting they violate _today_ because they'll be effectively more valuable _tomorrow_.

The same way an effective Altruist will poison a town's water supply if it grants them billions of dollars in the future and allows them to buy the town bottled water at some interdeterminate point in the future.

It boils down to the belief over reality, and feelings over facts; essentially, you're saying "as long as their intentions were good (gpus for everyone), it doesn't matter how much they hurt everyone"

No one is saying they broke a rule of accounting. Or, I haven’t seen anyone say how they violated GAAP rules. I think EA is dumb to but I don’t think that is the issue here.

I think this conflates the idea that - Sometimes conspiracies are boring. "I did a thing because of the boring, mundane case that it benefited me" is - while not sufficient for fraud - it is one of the building blocks.

And clearly one of the building blocks of cartels, monopolies, etc.

the point being "Boring and mundane" isn't a sufficient defense.

I think this is reasonable, but it doesn't necessarily make the endgame look any better.

A very similar thing happened to Sun during the dot com boom. All of these start ups were scrambling for Sun servers, and when the bubble burst, not only was the market flooded with lightly-used secondhand servers, but cheaper alternatives grew market share at the expense of Sun's expensive products.

Even if you believe in all the super AI and magic that is coming, I still don't see how you avoid a big shakeout (which has occurred in basically every historical boom: railroads, cars, PCs, dot com, etc.) that drastically reduces the number of NVidia customers.

Reply on news.ycombinator.com