At an August 2026 meeting of Nvidia’s senior finance team, chief financial officer Colette Kress reportedly reduced the company’s signed guarantee cap for OpenAI compute to roughly 4.25 gigawatts, or up to $125 billion, after initial discussions contemplated roughly 8 gigawatts valued at up to $250 billion, Wccftech reported on 13 November. The reported decision limited Nvidia’s commitment to support a buyer of its GPUs.
Key points
- Kress reportedly sought to reduce Nvidia’s financing commitments to smaller cloud providers and AI companies.
- The reported OpenAI guarantee cap was roughly 4.25 gigawatts, or up to $125 billion.
- The account links the financing push to excess chip orders and cloud companies’ shift towards custom components.
- An analyst separately estimated that at least $200 billion of GPUs bought by cloud companies were idle.
Kress sought to limit Nvidia’s guarantees
Kress’s reported intervention concerned what Wccftech described as circular financing: Nvidia committing money to AI companies and cloud providers that would, in turn, purchase its GPUs. The publication said she wanted to rein in commitments involving smaller cloud operators and larger AI firms, including OpenAI and Anthropic. Her objection centred on the financial exposure Nvidia was taking on through those arrangements.
The OpenAI figures describe two stages of that exposure. Roughly 8 gigawatts and up to $250 billion were under discussion initially, according to the account. The roughly 4.25-gigawatt, up-to-$125-billion figure was described as a signed guarantee cap. Both amounts concern potential underwriting of compute rather than a reported payment for GPUs already delivered.
Jensen Huang, Nvidia’s former chief executive, reportedly challenged Kress over her effort to scale back the deals. TechPowerUp reported that he called her a “deer in the headlights” during the August confrontation. Kress was identified as acting chief executive following Huang’s departure.
TechPowerUp also reported Kress’s concern that some smaller cloud providers receiving Nvidia investment could not afford to buy its GPUs. That concern bears directly on the financing arrangement: if a buyer depends on the chipmaker’s funding to make a purchase, Nvidia carries exposure to the buyer alongside its commitment to the chips.
TSMC orders drove the reported financing push
Nvidia reportedly faced a large volume of silicon from TSMC and memory for which it had no immediate buyers. Wccftech reported that Huang privately acknowledged a shift among major cloud companies away from standard components and towards chips they designed themselves. Its account described Nvidia’s prepaid wafer and memory commitments as a source of pressure to find other GPU customers.
Under the reported arrangements, Nvidia would fund AI companies and data-centre construction while those companies bought Nvidia hardware. TechPowerUp characterised the alleged effect as maintaining the appearance of strong demand during a GPU and memory shortage. That account of scarcity is an allegation about the effect of Nvidia’s financing, not a reported measure of how much supply was available to other buyers.
Wccftech argued that funding purchases of surplus chips could prevent extra supply from pushing down GPU prices. Its account ties that proposed price effect to the financing of buyers; it gives the August meeting and the OpenAI guarantee figures as the financial terms of the reported dispute.
Zitron estimated $200 billion in idle GPUs
Industry analyst Ed Zitron separately estimated that at least $200 billion worth of Nvidia AI GPUs bought by large cloud companies were sitting idle in warehouses, according to Wccftech. The publication said those purchasers lacked sufficient electrical capacity or immediate customer demand to operate them. His estimate concerns chips already bought by cloud companies, rather than the TSMC and memory orders Nvidia reportedly faced.
The two accounts place the financial risk at different points in the chain. Kress’s reported concern involved commitments to finance prospective GPU buyers. Zitron’s estimate concerned the use of hardware after a purchase had been made. Nvidia reportedly contacted TSMC about reducing its forthcoming chip orders.
