Nvidia’s announcement earlier this week that it’s investing $100 billion into OpenAI to assist fund its large knowledge heart construct out has added to a rising sense of unease amongst buyers that there’s a harmful monetary bubble round AI, and that the revenues and earnings math underpinning the valuations of each private and non-private firms within the sector simply doesn’t add up.
Whereas Nvidia’s newest announcement is by far the biggest instance, the AI chipmaker has engaged in a collection of “circular” offers by which it invests in, or lends cash to, its personal clients. Vendor financing exists to a point in lots of industries, however on this case, round transactions might give buyers an inflated notion of the true demand for AI.
In previous expertise bubbles, income “roundtripping” and tech firms financing their very own clients have exacerbated the harm when these bubbles finally popped. Whereas the share of Nvidia’s revenues which might be at present being pushed by such financing seems to be comparatively small, the corporate’s dominance because the world’s most precious publicly-traded firm implies that its inventory is “priced for perfection” and that even minor missteps may have outsized influence on its valuation—and on monetary markets and even perhaps the broader economic system.
The extent to which your complete AI growth is backstopped by Nvidia’s money isn’t simple to reply exactly, which can be one of many unsettling issues about it. The corporate has struck a lot of funding and financing offers, a lot of that are too small individually for the corporate to think about “material” and report in its monetary filings, though collectively they could be important.
As well as, there are such a lot of interlocking rings of circularity—the place Nvidia has invested in an organization, reminiscent of OpenAI, that in flip purchases providers from a cloud service supplier that Nvidia has additionally invested in, which then additionally buys or leases GPUs from Nvidia—that disentangling what cash is flowing the place is much from simple.
Tangled webs of funding
Two of probably the most outstanding examples of Nvidia’s net of circuitous investments are OpenAI and Coreweave. Along with the most recent funding in OpenAI, Nvidia had beforehand participated in a $6.6 billion funding spherical within the fast-growing AI firm in October 2024. Nvidia additionally has invested in CoreWeave, which provides knowledge heart capability to OpenAI and can be an Nvidia buyer. As of the top of June, Nvidia owned about 7% of Coreweave, a stake value about $3 billion at present.
The advantages that firms get from a Nvidia funding lengthen past the money itself. Nvidia’s fairness stakes in firms reminiscent of OpenAI and Coreweave allow these firms to entry debt financing for knowledge heart tasks at probably considerably decrease rates of interest than they’d be capable to entry with out such backing. Jay Goldberg, an analyst with Seaport International Securities, compares such offers to somebody asking their mother and father to be a co-signer on their mortgage. It offers lenders some assurance that they could really get their a refund.
Startups financing knowledge facilities have usually needed to borrow cash at charges as excessive as 15%, in comparison with 6% to 9% that a big, established company reminiscent of Microsoft may need to pay. With Nvidia’s backing, OpenAI and Coreweave have been in a position to borrow at charges nearer to what Microsoft or Google may pay.
Nvidia has additionally signed a $6.3 billion deal to buy any cloud capability that CoreWeave can’t promote to others. The chipmaker had beforehand agreed to spend $1.3 billion over 4 years on cloud computing with CoreWeave. Coreweave, in the meantime, has bought at the least 250,000 Nvidia GPUs thus far—nearly all of which it says are H100 Hopper fashions, which price about $30,000 every—which implies Coreweave has spent about $7.5 billion shopping for these chips from Nvidia. So in essence, all the cash Nvidia has invested in Coreweave has come again to it within the type of income.
Nvidia has struck related cloud computing offers with different so-called “neo-cloud” firms. In accordance with a narrative in The Info, Nvidia agreed this summer season to spend $1.3 billion over 4 years renting some 10,000 of its personal AI chips from Lambda, which like Coreweave runs knowledge facilities, in addition to a separate $200 million deal to hire some 8,000 extra over an unspecified time interval.
For individuals who imagine there’s an AI bubble, the Lambda deal is evident proof of froth. These Nvidia chips Lambda is renting time on again to Nvidia? It purchased them with borrowed cash collateralized by the worth of the GPUs themselves.
Apart from its giant investments in OpenAI and Coreweave, AI chipmaker additionally holds multi-million greenback stakes in a number of different publicly-traded firms that both buy its GPUs or work on associated chip expertise. These embrace chip design agency Arm, high-performance computing firm Utilized Digital, cloud providers firm Nebius Group, and biotech firm Recursion Prescription drugs. (Nvidia additionally just lately bought a 4% stake in Intel for $5 billion. Like Arm, Intel makes chips that in some circumstances are options to Nvidia’s GPUs, however which for probably the most half are complementary to them.)
Earlier this month, Nvidia additionally pledged to take a position £2 billion ($2.7 billion) in U.Ok. AI startups, together with at the least £500 million in Nscale, a U.Ok.-based knowledge heart operator that may, presumably, be utilizing a few of that cash to buy Nvidia GPUs to provision the information facilities it’s constructing. Nvidia additionally mentioned it could put money into a lot of British startups, each straight and thru native enterprise capital corporations, and a few of that cash too, will doubtless come again to OpenAI within the type of computing purchases, both straight, or by means of cloud service suppliers, who in flip might want to purchase Nvidia GPUs.
In 2024, Nvidia invested about $1 billion in AI startups globally both straight or by means of its company enterprise capital arm NVentures, in keeping with knowledge from Dealroom and The Monetary Instances. This quantity was up considerably from what Nvidia invested in 2022, the yr the generative AI growth kicked off with OpenAI’s debut of ChatGPT.
How a lot of this cash winds up coming proper again to Nvidia within the type of gross sales is once more, tough to find out. Wall Avenue analysis agency NewStreet Analysis has estimated that for each $10 billion Nvidia invests in OpenAI, it’ll see $35 billion value of GPU purchases or GPU lease funds, an quantity equal to about 27% of its annual revenues final fiscal yr.
Echoes of the dotcom period
That sort of return would definitely make this type of buyer financing worthwhile. Nevertheless it does increase issues amongst analysts a few bubble in AI valuations. These sorts of round offers have been an indicator of earlier expertise bubbles and have usually come again to hang-out buyers.
On this case, the lease preparations that Nvidia is getting into into with OpenAI as a part of its newest funding may show problematic. By leasing GPUs to OpenAI, quite than requiring them to purchase the chips outright, Nvidia is sparing OpenAI from having to take an accounting cost for the excessive depreciation charges on the chips, which can in the end assist OpenAI’s backside line. Nevertheless it implies that as a substitute Nvidia must bear this depreciation prices. What’s extra, Nvidia may even tackle the danger of being caught with a listing of GPUs nobody desires if demand for AI workloads don’t match Nvidia CEO Jensen Huang’s rosy predictions.
To some market watchers, Nvidia’s newest offers really feel all-too-similar to the excesses of previous expertise booms. Through the dot com bubble on the flip of the twenty first Century, telecom tools makers reminiscent of Nortel, Lucent, and Cisco lent cash to startups and telecom firms to buy their tools. Simply earlier than the bubble burst in 2001, the quantity of financing Cisco and Nortel had prolonged to their clients exceeded 10% of annual revenues, and the quantity of financing the highest 5 telecom tools makers had offered to clients exceeded 123% of their mixed earnings.
Finally, the quantity of fiber-optic cabling and switching tools put in far exceeded demand, and when the bubble burst and lots of of these clients went bust, the telecom tools makers had been left holding the dangerous debt on their stability sheets. This contributed to a better lack of worth when the bubble burst than would have in any other case been the case, with networking tools companies dropping greater than 90% of their worth over the following decade.
Worse but had been firms reminiscent of fiber-optic large International Crossing that engaged in direct “revenue roundtripping.” These firms minimize offers—usually on the finish of 1 / 4 in an effort to hit topline forecasts—by which they paid cash to a different firm for providers, after which that firm agreed to buy tools of precisely equal worth. When the bubble burst, International Crossing went bankrupt, and its executives in the end paid giant authorized settlements associated to income roundtripping.
It’s reminiscences of those sorts of transactions which have prompted analysts to at the least increase an eyebrow at a few of Nvidia’s round investments. Goldberg, the Seaport International analyst, mentioned the offers had a whiff of round financing and had been emblematic of “bubble-like behavior.”
“The action will clearly fuel ‘circular’ concerns,” Stacy Rasgon, an analyst with Bernstein Analysis, wrote in an investor observe following Nvidia’s announcement of its blockbuster funding in OpenAI. It’s a great distance from a priority to a disaster, after all, however as AI firm valuations get larger, that distance is beginning to shut.
