Technology

Nvidia's Finance Chief Pressed on Credit Guarantees for AI Chip Buyers

Martin HollowayPublished 5m ago4 min readBased on 2 sources
Reading level
Nvidia's Finance Chief Pressed on Credit Guarantees for AI Chip Buyers
Photo by Prime Minister's Office / GODL-India

Former Nvidia CEO Jensen Huang described Colette Kress as "a deer in the headlights" during a presentation in summer 2026, according to The Wall Street Journal. The remark, which the Journal reported, was made during a meeting with Nvidia's finance team.

The meeting concerned plans to provide guarantees worth billions of dollars to data center operators that could not afford to buy chips upfront. As described, the guarantees would support those operators' purchases.

Kress is Nvidia's chief financial officer. She replaced Huang as interim CEO after his resignation. Huang had held the top job at Nvidia for more than 33 years, and the company is searching for a permanent CEO.

According to TechCrunch, which cited the Journal's reporting, Huang also said Kress could not help him steer strategy or decision-making. The same report says Kress questioned the program-focused, credit-based deals involving Nico Caprez, Huang's son-in-law. In this context, "credit-based" means the buyer pays over time rather than settling in cash at the point of sale. Her questions, as described, were practical. She asked how to decide which of the deals were beneficial, how to value the guarantees, and how to determine the returns.

In my view, the personnel detail is the less consequential part of this story. The exchange concerns the economics of the buildout. A guarantee that lets a data center operator take delivery of accelerators, the GPUs that train and run AI models, before it can pay for them moves part of the counterparty risk, meaning the risk that the other side fails to pay, from the operator's balance sheet to the guarantor's. Whether that exposure is priced correctly depends on the valuation and return assumptions Kress was asking about. Those are standard questions for any finance function, but they carry more weight when the deals are large, tied to one supplier's hardware, and structured around credit rather than immediate cash sales.

The broader context here is a supplier moving from selling silicon to underwriting the customers who buy it. Vendor financing, where a seller extends credit to its buyers, is an old tool in capital equipment, and it has helped markets expand when buyers lacked capital. It also concentrates risk on the vendor if operators struggle to meet their obligations. The CFO, now running the company in an interim capacity, was asking the questions that determine whether such deals create durable capacity or a receivable, meaning money owed to the company, that is difficult to collect.

For readers tracking the AI infrastructure build, the story is a reminder that compute demand now runs through financing structures as much as through roadmaps. Credit-backed capacity can bring new data centers online sooner than buyers could fund them alone. Whether those facilities earn back their cost will depend on the discipline of the deal terms, which is the discipline Kress appeared to be insisting on.