Nvidia Shelved Its $36 Billion Program After Internal Criticism

Nvidia was prepared to lend money to customers buying the company's own GPUs, dictate the minimum price those customers could charge for renting the capacity out, and act as tenant of last resort if their data centers sat empty. On top of that, the company would take a share of its customers' revenues for six years.

That was the plan. According to the Wall Street Journal, as reported by heise online, preparations for the program have now been temporarily halted — and according to several people involved in the matter, it was parts of Nvidia's own staff who objected that the terms went too far. Exactly why the program has been put on hold is unknown.

So a company whose chips are in such demand that customers queue up felt it needed to finance its own customers — and when the terms were drawn up, parts of the company turned against the idea. That is the most telling detail in the story: the market power was so total that it frightened even the party wielding it.

Beyond Supply and Demand: A New Business Model

The program was reportedly elaborate in its mechanics. Nvidia had set aside $36 billion, CFO Colette Kress told shareholders, according to WSJ reporting relayed by heise. The partnerships were normally meant to run for six years. The customers — smaller AI infrastructure providers that lacked the billions required to outfit data centers with GPUs quickly — would receive loans from Nvidia itself to buy the chips.

The partners would then build data centers, rent the capacity to their own customers, and give Nvidia a share of the revenues over the term of the agreement. And this is where the model becomes more than vendor financing: according to the WSJ, citing several people with knowledge of the matter, the partners would be required to rent out capacity at a minimum of a few dollars per hour per GPU — a price negotiated with Nvidia in advance, sized to cover depreciation of the chips, the data center's fixed costs, and staffing. Nvidia, finally, would rent the data centers itself if no one else would.

Taken together, that is no longer just a loan. It is vendor financing fused with price control over the customer's own business, with Nvidia as safety net and revenue partner in the same contract.

Why Now

The timing is no accident. Nvidia's latest earnings lifted the markets and revived AI optimism, according to market coverage from Yahoo Finance. At the same time, Chinese labs are pushing hard: improving training efficiency and narrowing the gap with the leading models, as Business Times reports.

And costs are rising for Nvidia's customers. A reported 15 percent price increase, relayed by MSN, is working its way through the AI economy and hitting hardest those who already have to scrape together billions in equity just to get in the door.

These smaller players are precisely whom the program was built for. They represent Nvidia's growth frontier — and they are the only ones who cannot pay cash. That is why a supplier with a near-monopoly needed financing at all: demand for the chips is enormous, but the upfront cost per customer has grown larger than their balance sheets can bear.

According to figures from International Data Corporation, cited by Business Insider, global spending on AI infrastructure is forecast to reach roughly $487 billion in 2026 and pass $1 trillion by 2029. At that scale, the chips are no longer the bottleneck. The financing is — and who owns the risk.

Where the Line Falls

Vendor financing has historically been a late-cycle signal in capital-intensive industries: when the supplier must lend customers money to keep sales moving, demand has begun to be financed rather than real. In Nvidia's case, the reading is more nuanced — the chips are overwhelmingly in demand, as heise explicitly notes — but the mechanism that would have taken shape is the same: Nvidia would have become creditor, price-setter, and landlord to the same customers, booking its own demand through its own lending.

It is a circularity that creates three concrete problems. Legally: a supplier that dictates its customers' prices moves into territory competition authorities routinely scrutinize. Financially: if AI investments falter, Nvidia is left holding both unpaid loans and vacant data centers it rents itself. Reputationally: circular arrangements in which one party sits on both sides of the ledger are already a contested subject in the AI boom.

According to the WSJ's source material, relayed through heise, this is what some of Nvidia's own employees reacted to: the question of whether the company was meddling too deeply in its customers' businesses and exploiting its market power too aggressively. With the causes unconfirmed, it remains remarkable that the resistance came from inside — not from regulators or customers.

What the Pause Means

For the smaller AI cloud providers the program was tailored to, the pause changes the math noticeably. Without Nvidia loans, they must finance GPU purchases on their own — even as the reported 15 percent price increase makes entry more expensive. The options are consolidation (fewer, larger players able to carry the capital), costlier capital from other sources, or deferral. For Nvidia's growth story, that means the front line — the small buyers — is temporarily on hold.

There is a counter-explanation that deserves to be taken seriously: the pause may be tactical, not principled. The exact reasons are unknown, and Nvidia has several other new business areas in motion. Perhaps the program simply is not needed — if demand holds, the company can sell everything for cash and leave the financing risk with someone else. In that case, the pause is not a sign of self-awareness but of surplus time.

But it does not change what has already happened. Someone at Nvidia read the contract as it was actually written — with six-year revenue sharing, minimum prices set by the supplier, and the company itself as tenant in reserve — and concluded that the supplier had gone too far into its customers' business. The question that remains is not whether the program returns. It is how much of the AI economy's plumbing one supplier can own before the line is drawn — and who draws it: customers, regulators, or, as this time, the company's own employees.