Nvidia trims OpenAI guarantee to $120 billion as AI financing shifts

Nvidia has cut its guarantee for OpenAI's Ohio data center to under $120 billion from $250 billion, while unveiling a $500 billion compute-financing platform that shifts AI risk off its balance sheet.

"In AI, compute is revenue," Jensen Huang, chief executive officer of Nvidia, said.

The Ohio reduction covers only the first phase of the 10-gigawatt campus developed by SoftBank's SB Energy, with a deal possible as early as this weekend. Nvidia's 13F filing shows a $63.4 billion portfolio as of June 30, up from $18.4 billion, including roughly $30 billion in Intel and $21 billion in SpaceX.

The moves mark Nvidia's transition from chip vendor to financial architect of the AI boom, tying its balance sheet to the infrastructure that consumes its processors. The market's verdict comes Aug. 26, when Nvidia reports quarterly earnings expected near $91.9 billion.

A Guarantee Trimmed, Not a Retreat

The Wall Street Journal reported Thursday that Nvidia scaled back its investment obligation for the first phase of the Ohio project — roughly 5 gigawatts of a campus ultimately planned for 10 gigawatts and more than $500 billion in total cost. Reuters indicated the revised deal could be signed as early as this weekend. The change came after investors raised concerns about Nvidia's exposure to large financing commitments; the original $250 billion disclosure knocked 5 percent off the stock in a single day.

The reduction looks less like an exit than a renegotiation of who bears what risk. Nvidia remains embedded in the SB Energy-developed infrastructure and committed in September 2025 to supply at least 10 gigawatts of its own systems, with potential participation of up to $100 billion. The company has also been in discussions to invest up to $3 billion in SB Energy, SoftBank's renewable-energy subsidiary, and is weighing a similar $3 billion investment in Lancium, the developer behind the Stargate data center in Texas, which could go public in 2027.

That interpretation gains weight alongside the financing platform Nvidia announced Aug. 10 with Apollo Global Management, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — an alliance designed to mobilize more than $500 billion for AI infrastructure. Huang has said Nvidia would backstop up to 25 percent of that sum, roughly $125 billion. Goldman Sachs is reportedly in talks with additional investors to join. The structure contrasts with rivals such as AMD and Intel, which still lean on conventional processor sales rather than underwriting their own demand.

The Software Track Runs in Parallel

While capital flows into concrete and turbines, Nvidia continues advancing its software ambitions. Reuters reported this week that the company is developing Nemotron 4, a new family of open AI models, with the largest version expected to exceed one trillion parameters and a possible launch targeted for late autumn. The company has also introduced Nemotron 3.5 Lightning, designed for tasks including code review, tool use, and monitoring security alerts.

The partnership channel is active as well. IBM and Together AI announced a multi-year, $240 million agreement to build a large AI cluster on IBM Cloud using Nvidia HGX B300 systems and Spectrum-X networking technology, initially deploying around 2,000 Nvidia Blackwell 300 chips. Nvidia also closed the acquisition of SchedMD, the company behind Slurm, an open-source workload-management system widely used in high-performance computing centers, pledging to keep the software open-source and vendor-neutral.

The stock's reaction to this flurry of activity has been measured rather than euphoric. Shares closed Friday at €194.74, down 0.4 percent on the day, though the 30-day picture shows a 5.1 percent gain and the year-to-date return stands at 22 percent. The stock sits 3.8 percent below its 52-week high of €202.50 reached in May — and 39 percent above its yearly low of €139.78.

Skeptics have their case. Investor Steve Eisman warns that the entire AI boom rests on just two customers — OpenAI and Anthropic — which together account for roughly 70 percent of AI-related revenue at Microsoft, Amazon, Alphabet, and Oracle. Michael Burry has positioned himself publicly against Nvidia and semiconductor stocks. Goldman Sachs has pegged unstarted hyperscaler leasing commitments at around $1 trillion. Insider selling at Nvidia totaling $767.2 million over 90 days is hard to ignore entirely.

The market will render its next verdict on Aug. 26, when Nvidia reports quarterly earnings. Analysts expect revenue around $91.9 billion, while management has guided to a midpoint of $91.0 billion. Bank of America projects revenue between $94 billion and $95 billion and anticipates a raise in the next-quarter outlook to $107 billion to $108 billion.

What's becoming clear is that Nvidia has crossed a threshold. It no longer merely sells chips; it finances the power plants, data centers, and credit structures that make those chips useful. Whether that makes the company a more resilient player in the AI economy or simply a more exposed one is the question the coming quarters will answer. Nvidia shares, with a market capitalization of roughly €4,687 billion, trade near record highs — the market is treating the reordering of its financial architecture with equanimity, not alarm.

This article is for informational purposes only and does not constitute investment advice.