In the past few years of artificial-intelligence investment, we've had to get used to some very large numbers. Last week, I wrote about how five Big Tech companies have $2.7 trillion in future commitments, and this week comes news of $500 billion more in AI data-center funding from major firms, centered around Nvidia, plus a $20 billion equity sale from Intel.
Nvidia and Intel are trying to solve two very different problems, though both are tied to the AI data-center boom.
For Nvidia, there are doubts that the company can keep up its rocket-like growth and protect its 75% gross profit margin. Nvidia's biggest customers -- hyperscalers like Amazon.com, Microsoft, Alphabet, and Meta Platforms -- have reached the limits of their substantial operating cash flows and are dipping into debt and equity financing to fund 2026 and projected 2027 capital expenditures. Investors are wondering, how far can that go?
Moreover, the hyperscalers want to be less dependent on Nvidia and are seeking to diversify their supply chains, including by making their own AI chips. Nvidia is also seeing increased competition from the likes of Advanced Micro Devices and a host of smaller chip companies.
Wall Street analysts are still mostly bullish on Nvidia for the next couple of years, but other investors don't necessarily share that enthusiasm. The stock's forward price/earnings ratio has been below that of the PHLX Semiconductor Sector Index since late last year. Nvidia CEO Jensen Huang has tried many messages to break the stock out of its valuation rut but has been largely unsuccessful so far.
The new $500 billion arrangement is an attempt to unleash even more private capital to fund data centers filled with Nvidia equipment. Secondarily, it's meant to assuage fears that the data-center boom may run out of willing investors, and that the hyperscalers are the only important customers.
The details remain fuzzy, but in social-media posts Huang referred to it as a "platform" funded by Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs, and KKR that will provide capital to companies that have "demand for compute but do not yet have access to financing." Each deal will be evaluated by the financiers on a case-by-case basis.
"A conduit that brings $500 billion into the market can do nothing but drive down my cost of capital," CEO Mike Intrator of cloud provider CoreWeave told Barron's in a postearnings interview this week. "Whether I use that facility or I go to a different part of the market, I am very, very excited about the maturation of the financing market coming to bear on the infrastructure that we have to build." CoreWeave is a Nvidia customer and relies heavily on outside funding.
This isn't the "circular financing" we've heard so much about, but it could become that. With this platform, Huang is trying to make the case that Nvidia equipment is an investible asset class that can have a useful life "toward a decade." Indeed, Nvidia's 2020 A100 GPU chips are still in wide use and going strong six years in. This week, CoreWeave revealed that it signed a deal to rent out its A100s through 2029.
But should that useful life shrink in the future, Nvidia may put itself on the hook to make up some of the difference with a "residual-value support mechanism." This will be limited to 25% of a deal and entered into on a case-by-case basis. Huang says that it's "designed to complement -- not replace -- independent underwriting."
Nvidia is able to throw its weight around and even backstop some of the financing because of its mounting free cash flow, expected to hit $200 billion this year and keep rising. Intel, on the other hand, has had negative free cash flow since 2022, and analysts don't expect it to get above water until 2028.
But Intel suddenly has a catalyst that doesn't depend on its success in finding outside customers for its chip manufacturing. While the lion's share of AI chip investment will still go to Nvidia and its competitors, earlier this year customers had a growing realization that data centers will also need plenty of CPU servers to run agent software on top of AI models. After years of missteps, Intel isn't nearly as dominant as it once was in this segment -- now seeing serious competition from AMD, Arm Holdings, and Nvidia. But 2030 market projections have ballooned from about $50 billion at the end of last year to more than $200 billion.
Intel needs to start building up its manufacturing capacity for that now. At its last earning call in July, the company raised its 2026 capital expenditure guidance from $18 billion to $20 billion and forecast that 2027's spend would be "significantly above" that. Operating cash flow is rising, but not as quickly as capex in the next 18 months. Intel needed financing, and with the stock trading at 56 times forward earnings per share, and $50 billion in debt already on the balance sheet, a share sale was the logical choice.
In the past year, the company has made large share sales to the U.S. government and Nvidia. Intel's diluted share count had risen 17%, and the recent offering increased that to 22%. Another 71 million shares are in escrow, awaiting transfer to the U.S. government, and aren't included in that count. In all of the deals, buyers got a discount to the market share price.
The wild card continues to be whether Intel can find outside customers large enough to erase losses in its manufacturing segment, which topped $10 billion last year. To do that may require even more capex, and more financing.
Intel's prospects may be improving, but this week investors were reminded that it comes with a cost.
Comments