Key Highlights

• Nvidia's (NASDAQ:NVDA) potential financial exposure to customers through guarantees and purchase commitments totals roughly USD 300 billion, though these obligations only come due in a downturn.

• The company has made about 150 Equity investments in AI startups and customers over the past two years, nearly double the pace from two years earlier.

• Hyperscalers, which account for roughly half of Nvidia's revenue, are increasingly designing their own chips, threatening Nvidia's future sales to its biggest customers.

• Critics including investor Michael Burry have questioned whether Nvidia's financing arrangements resemble the vendor financing that led to losses at Cisco and Lucent during the dotcom era.

Why Nvidia Is Financing Its Own Customers

Nvidia's financial engineering partly reflects the fact that its biggest customers are becoming its rivals. Hyperscalers such as Amazon (NASDAQ:AMZN), Google (NASDAQ:GOOGL), Meta (NASDAQ:META) and Microsoft (NASDAQ:MSFT) account for roughly half of Nvidia's revenue, but most are now designing their own AI chips, which cost between a fifth and a third as much as Nvidia's and can be tailored to each company's own software. One data provider predicts custom chips will account for about 50% of the AI processor market by the end of the decade, up from roughly 40% today.

Because hyperscalers carry investment-grade Credit ratings, their own borrowing costs remain low. Upstart neoclouds have similar Capital needs but far less revenue, making their Debt considerably more expensive. Alphabet sold 50-year bonds in November at an annual rate of 5.7%, while CoreWeave (NASDAQ:CRWV), the largest neocloud, borrowed at nearly double that rate in July. Nvidia's financing arrangements are, in effect, an attempt to narrow that gap.

A Web of Equity Stakes and Backstops

Nvidia pursues this through several mechanisms. It takes equity stakes in startups that will become customers or otherwise fuel Demand for its chips, making roughly 150 such investments over the past two years, nearly double the pace of two years earlier. Some of these bets specifically aim to propagate open-weight AI models. In August Nvidia agreed to pay Poolside, an AI coding startup, USD 6 billion to license its software and a further USD 1 billion for an equity stake, and separately agreed to acquire Hugging Face, a platform hosting open AI models, for roughly USD 12.9 billion.

More directly, Nvidia has begun promising to top up neoclouds' income from new data centers to an agreed floor for periods often lasting six years, making future Revenue more predictable and thereby lowering the cost of the debt neoclouds take on. Sharon AI, an Australian neocloud, expects to draw on a backstop of this sort worth about USD 4.9 billion. Nvidia's cumulative exposure across all such commitments, including a USD 105 billion guarantee behind OpenAI's Ohio Data Center and roughly USD 125 billion through its Wall Street partnership, totals close to USD 300 billion, though these liabilities only materialize in a genuine downturn and do not currently appear on Nvidia's balance sheet.

Whether the Chips Will Hold Their Value

Underlying all these arrangements are two assumptions: that Nvidia's chips retain their value over time and that demand for computing capacity keeps growing. Neither is guaranteed. Huang argues the chips are durable, bankable Assets whose usefulness is extended through software upgrades, evidenced by continued demand for older chips like the H100, whose rental cost has fallen only modestly since its 2023 launch. But that durability may partly reflect current Scarcity of Supply rather than intrinsic value, since companies have little choice but to keep older chips running when newer ones remain constrained.

Conclusion

Nvidia's own finances currently appear strong enough to absorb its liabilities even in a downturn, given a cash and securities stockpile of roughly USD 99 billion and a Business expected to generate around USD 200 billion in cash this year. Whether that cushion remains sufficient as its guarantees keep expanding, with one analyst estimating exposure could reach USD 175 billion from its neocloud backstop program alone by 2028, will likely determine how much of any future AI industry downturn Nvidia ends up absorbing itself.