Nvidia's OpenAI pledge turns the chip supplier into the financier

Nvidia plans to invest up to $100 billion in OpenAI as 10 gigawatts of its systems come online. Demand for AI chips is now partly funded by the seller.
Nvidia intends to invest up to $100 billion in OpenAI, released progressively as each gigawatt of Nvidia systems comes online. The letter of intent, signed on 22 September, has OpenAI deploying at least 10 gigawatts, which Jensen Huang put at 4 million to 5 million GPUs, with the first gigawatt targeted for the second half of next year on the Vera Rubin platform and the details to be finalized in the coming weeks.
The chip supplier has become the financier of its own demand. Nvidia invests in OpenAI, OpenAI spends on Nvidia systems, and part of the chip maker's future revenue is paid for from its own balance sheet, the same circle that formed when clouds bought lab equity. Vendor financing is old. It has never run at this scale, or for a customer that is not yet profitable, and that turns a supply deal into a concentrated bet.
The scale shows in the week's other announcement. A day later OpenAI and its partners named five new sites for Stargate, the build-out unveiled in January, lifting planned capacity to nearly 7 gigawatts and investment to more than $400 billion over three years. The largest AI chip supplier, the largest AI lab and their partners now rest on one premise: demand for model usage compounds fast enough to pay for gigawatts of capacity. Valuations from chips to clouds to labs hang on fewer, larger and more interlinked commitments, so a slowdown at any one point would travel quickly through all of them.
Startups will not compete for that capacity and do not need to. A few buyers with the balance sheets, or the vendor backing, to sign gigawatt deals are reserving it years ahead. The opening for smaller companies sits where capacity turns into revenue: products that make heavy model usage worth paying for, and tools that cut the cost per task through routing, caching, smaller models and efficient serving. Providers that have committed to this much compute will push hard for usage growth to fill it, with generous pricing, credits and partnerships for companies that drive volume. That window is worth using, provided the product stays portable between providers.
A letter of intent still leaves room for the numbers to shrink. Whether the definitive agreement is signed in the coming weeks, and how the first tranche of investment is priced, will show how firm the $100 billion is. If other chip suppliers answer with similar financing for other labs, vendor financing becomes the norm for frontier compute rather than one company's choice. Power decides the rest, because the first gigawatt coming online on schedule is the physical constraint behind every number in this deal.
When suppliers fund part of the AI build-out, Fanaura keeps its exposure where revenue comes from end customers, the discipline its thesis rests on, rather than from partners inside the same circle. The companies worth owning profit from abundant compute without paying for it upfront.
Building an AI-native company? Apply in five minutes. →
More from Fanaura
Building tools that cut the cost per task of running AI models at scale? Send us the project.
Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.
Get the news by email
One short note each weekday. No marketing, and you can unsubscribe at any time.