Nvidia Puts a Number on It: AI Labs It Finances Will Be ~25% of Next Year's Business

For most of the past two years, the debate over Nvidia’s growing role as banker to the AI industry has run on anecdote and aggregation: a $100B investment here, a multi-billion credit facility there, stitched together by analysts into a narrative of “circular financing” reminiscent of the telecom vendor-finance era. This week, Nvidia itself finally put hard numbers on the arrangement — and they are bigger than most critics assumed.

According to DIGITIMES, reporting on August 27 after the company’s blockbuster fiscal Q2 2027 earnings, Nvidia says the AI labs it is helping to finance will represent about a quarter of its business next fiscal year. The disclosure landed alongside earnings materials showing the company has now signed $366 billion in total future commitments — spanning supply, capacity, cloud agreements, leases, equity investments and capex — plus maximum gross guarantee exposure of $108.5 billion tied to large AI infrastructure buildouts, heavily concentrated in the next three fiscal years. Roughly a fifth of that guarantee exposure, some $105 billion, relates to a single project: SB Energy’s PORTS-Pike campus in Ohio, where OpenAI is expected to be the anchor tenant.

What Huang actually said

The numbers give new weight to CEO Jensen Huang’s unusually direct defense of the strategy, delivered on CNBC’s Mad Money the evening the earnings dropped. Nvidia’s investments in frontier AI labs are a “once-in-a-generation” opportunity, Huang argued, and the risk they pose to Nvidia itself is minimal.

His reasoning: “This is the first generation of startups that needed tens of billions of dollars to get funded.” Frontier-model development is so capital-intensive that the labs structurally cannot finance their compute through conventional channels — many lack the financial history or investment-grade credit profiles to raise cheap debt on their own. Nvidia steps in with equity and credit, the labs get their clusters, and Nvidia’s ecosystem deepens. “The money we’ve invested is going to generate tremendous returns,” Huang said. “I think the risk is low.”

The core of the defense is a redeployability argument: unlike a dot-com-era vendor loan that evaporates when the borrower fails, Nvidia’s exposure is backed by physical GPU infrastructure. If a financed lab stumbles, the chips sitting in a data center can be reassigned to other customers and other workloads. The asset doesn’t disappear; it finds a new buyer — at least while demand outruns supply, which, by Huang’s own telling, it still does.

The 70% year nobody is talking about

The financing disclosure almost got buried under the headline numbers, which were themselves remarkable. Q2 FY27 revenue hit $96.2 billion, up 106% year over year, with data center revenue up 117% to $89 billion. CFO Colette Kress then guided to something more extraordinary: approximately 70% revenue growth for fiscal 2028 — nearly double the ~45% consensus analysts had modeled — with quarterly guidance of $108 billion ±2% and gross margins holding around 74% (before a planned reset toward 72% as new financing structures and China dynamics flow through).

On the earnings call, Huang added a qualifier that deserves more attention than it got: even 70% understates reality, because the constraint is supply, not demand. “Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%.”

That framing does double duty. It reassures investors that growth is durable. But it also quietly reframes what the financed-lab revenue means: if Nvidia is supply-constrained, then chips allocated to labs it finances are chips not sold to someone else — the demand signal becomes harder to read as independent.

Why a quarter of revenue from financed labs matters

A ~25% figure for fiscal 2028 puts concrete scale on what was previously a fuzzy accusation. It means that within roughly a year, one dollar in four of Nvidia’s revenue would flow from customers whose capital, in part, originated with Nvidia itself — through equity stakes in OpenAI and Anthropic, credit lines to neoclouds, and guarantee-backed data center projects like the Ohio campus.

For bulls, this is a feature: Nvidia is converting its balance sheet into a demand moat, extending its technology into every layer of the stack, and capturing equity upside in its own customers. The $500B financing consortium with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — announced earlier in August — spreads the risk across Wall Street while Nvidia keeps the ecosystem lock-in.

For skeptics, the concentration cuts the other way. The redeployability argument is strong in a single-customer-failure scenario and weak in a broad-demand-downturn scenario — and those are precisely the scenarios investors can’t distinguish in advance. If several heavily financed labs retrench simultaneously while hyperscaler capex cools, the “reassign the GPUs” safety valve assumes there is someone else in line to take them. Kress’s own disclosure that Nvidia shipped less than 1% of data center revenue into China in Q2, and is assuming zero China compute revenue going forward, removes one potential absorber of freed capacity from the equation.

There is also an accounting-adjacent subtlety. Revenue recognized against financed customers is still revenue — the chips ship, the power draws, the models train. But growth that is partly a function of the seller’s own capital allocation is a different informational object from growth driven by unrelated third parties writing checks. Analysts who treat the two as interchangeable are reading Nvidia’s income statement as a cleaner signal of external AI demand than it now is.

The honest bull case, stated honestly

What makes this moment unusual is that both readings can be true at once. The capital needs of frontier AI are genuinely unprecedented — Huang is right that no previous startup generation required tens of billions in pre-revenue funding, and someone has to intermediate that capital. Nvidia deploying its $56.6B cash pile (per the 8-K) to lock in its ecosystem is rational corporate strategy, not fraud.

Equally, a company whose revenue growth is increasingly self-referential carries a structural fragility that a 106%-growth quarter doesn’t dispel — it obscures. The dot-com comparison is inexact (GPUs redeploy, dark fiber didn’t), but the inexactness is a matter of degree, not kind. Vendor financing became legendary not because the financed purchases were fake, but because they masked the point at which organic demand stopped clearing the market.

Nvidia’s fiscal 2028 will be, in part, a live test of the redeployability thesis. With ~25% of that year’s business tied to labs Nvidia itself bankrolls, $366B in commitments on the books, and $108.5B of guarantees standing behind concrete and power contracts, the company has moved from selling picks and shovels to also underwriting the mine. Huang says the risk is low. For the first time, we can now put numbers on what it would mean if he’s wrong.