Nvidia Would Back $250 Billion for OpenAI Data Center

What happened?

According to The Wall Street Journal (WSJ), Nvidia is in talks to offer a guarantee of approximately $250 billion that would allow OpenAI to lease a colossal 10-gigawatt data center campus that SoftBank's energy subsidiary, SB Energy, is building in southern Ohio (USA). The guarantee is not cash but a financial guarantee — a credit wrapper — that would reduce risk for lenders and lower the cost of debt needed to build the infrastructure. This type of operation is unprecedented in magnitude and the actors involved: a chip manufacturer backing the construction of a data center park for a client. The original source, WSJ, cites people familiar with the negotiations, who emphasize that the deal is not closed and could fall through.

Why is it important?

This move marks a milestone in the race for artificial intelligence infrastructure. Until now, large data center projects were typically financed by the hyperscalers themselves (Google, Microsoft, Amazon) or by investment funds. That Nvidia, a chip manufacturer, acts as a guarantor for an operation of this magnitude is unprecedented and reveals the extent to which AI computing demand is redefining strategic alliances. For context, the combined capital expenditure of hyperscalers in 2024 is estimated at around $200 billion, according to Synergy Research Group. A $250 billion guarantee for a single project doubles that annual figure, highlighting the unprecedented scale of AI computing demand.

For OpenAI, which is not yet profitable and lacks sufficient credit rating (it does not have investment grade), this guarantee is key to accessing the computing capacity it needs without diluting its capital. According to SemiAnalysis estimates, training models like GPT-5 could require clusters of up to 100,000 H100 GPUs, with an energy cost exceeding 100 MW. The Ohio campus, with 10 GW, would be enough to power about 100 clusters of that size, allowing OpenAI to massively scale its operations. For Nvidia, it means securing a top-tier customer that will consume its GPUs for years, consolidating its dominance in the AI accelerator market. Nvidia already controls approximately 80% of the AI chip market, according to Mercury Research, and this deal could strengthen its position against competitors like AMD and Intel.

Consequences and context

If the deal closes, it could set a precedent for other chip manufacturers (AMD, Intel) or tech companies to offer similar guarantees to AI startups. This would alter the infrastructure financing dynamic, where traditionally investment funds or the hyperscalers themselves assumed the risk. It would also increase pressure on regulators, who are already investigating Nvidia's control over the GPU market. The U.S. Department of Justice and the Federal Trade Commission have shown interest in Nvidia's practices, according to Reuters, and a deal of this magnitude could be scrutinized under antitrust laws.

This is not the first time a credit wrapper has been used in the sector: Google already backed Anthropic's data centers, partly to boost its own TPU chips. However, the scale of Nvidia's operation is ten times larger. Moreover, the Ohio project is not the only one: SoftBank is also investing in other data center campuses, such as the 5 GW one in Texas, according to Bloomberg. The involvement of SB Energy, a SoftBank subsidiary, adds an extra layer of complexity, as SoftBank is also a significant investor in OpenAI through its Vision Fund, with an estimated 15% stake according to PitchBook.

The historical context is relevant: during the dot-com bubble, companies like Cisco financed their customers to drive adoption of their networking equipment, but on a much smaller scale. At that time, Cisco offered loans of about $5 billion, a fraction of Nvidia's proposed guarantee. The difference is that Nvidia is not lending money but acting as a guarantor, which limits its initial exposure but obliges it to assume default risk. If OpenAI or SoftBank defaults, Nvidia might have to cover part of the debt, although its current balance sheet can support it: the company has more than $30 billion in cash and equivalents, according to its latest quarterly report.

"Nvidia not only sells the shovels but also finances the AI gold rush."

What should readers know?

• The deal is not closed: negotiations continue and could break down, according to WSJ. There is no guarantee it will materialize.
• It's not direct money: Nvidia is not lending $250 billion to OpenAI; it only acts as a guarantor so that SoftBank can obtain financing on better terms. The guarantee reduces the cost of debt, which could be around 5-6% annually, compared to 8-9% without a guarantee.
• Market implications: If it succeeds, other chip companies could imitate the model, altering the AI infrastructure financing dynamic. AMD and Intel are already exploring similar deals, according to industry sources. Additionally, it could accelerate data center construction, which currently faces bottlenecks in the supply chain for transformers and cooling systems.
• Risk for Nvidia: If OpenAI or SoftBank defaults, Nvidia might have to assume part of the debt. However, the risk is mitigated because the guarantee is tied to a tangible asset (the campus) and SoftBank has a solid balance sheet. Moreover, Nvidia could benefit from recurring revenue from GPU and software supply (like CUDA) over the data center's lifespan, estimated at 10-15 years.
• Impact on users: For end users of ChatGPT and other OpenAI services, this deal could translate into more powerful models and lower latency, but also higher energy consumption. A 10 GW data center is estimated to consume as much electricity as 8 million homes, posing sustainability challenges. OpenAI has committed to using renewable energy, but viability at that scale is uncertain.

In summary, Nvidia's potential guarantee to OpenAI represents a financial innovation in the AI sector, with implications ranging from chip competition to regulation and the environment. The coming months will be crucial to see whether this precedent solidifies or negotiations fail, leaving OpenAI to seek alternatives.