NVIDIA Corp. is in talks to provide financing guarantees for OpenAI’s roughly $500 billion data center program, with the chip maker potentially backing as much as $250 billion of the project, according to people familiar with the matter. The two companies are close to an agreement, The New York Times reported, after The Wall Street Journal first disclosed the discussions.
The talks include related agreements to lock in memory supply from SK hynix, the people said, tying the world’s largest AI chip maker to the memory supplier that makes the high-bandwidth chips its accelerators depend on. A deal of this scale would mark the first time a chip company has guaranteed the financing of a customer’s infrastructure buildout.
The arrangement would recast NVIDIA’s role in the AI economy. The company has spent two years as the indispensable supplier of the industry, selling accelerators that data center operators queue up to buy. A guarantee of this size turns NVIDIA into something else: a financier of its own customers, willing to put its balance sheet behind the belief that the AI buildout will keep growing.
For OpenAI, the deal would open a new path to capital. The company has relied on a mix of equity, debt and cloud credit to fund its expansion, but a $500 billion program outruns what traditional lenders have been willing to provide on their own. NVIDIA’s backing, and the chip maker’s credibility with suppliers, would let OpenAI build infrastructure without depending on a single hyperscaler’s cloud credit or a crowded bond market.
The structure under discussion shows how deeply the AI industry has come to rely on its own supply chain for capital. Google has taken similar steps for Anthropic, guaranteeing leases and power contracts in exchange for equity in a Texas data center project. Broadcom has financed chip purchases for the same customer. NVIDIA’s proposed guarantee for OpenAI extends the pattern to the industry’s largest scale yet.
The SK hynix component is the most intricate part of the package. High-bandwidth memory is the tightest bottleneck in AI computing, and NVIDIA’s new accelerator designs depend on it. By linking its guarantee to memory supply agreements, NVIDIA would be securing its own input pipeline while helping its largest customer build, a circular arrangement that locks the three companies together for years.
The risks are equally large. A guarantee of up to $250 billion would be the largest single contingent liability in the technology industry, and it would be tied to one customer’s ability to keep its data centers full. If AI demand slows, or if OpenAI’s economics deteriorate, NVIDIA would face obligations that no chip company has ever carried. Analysts said the exposure explains why negotiations have taken months and why the final terms matter more than the headline number.
The deal also changes the competitive math. NVIDIA’s rivals, including AMD and the custom chip programs at Google and Amazon, are trying to dislodge it from the center of the AI industry. By becoming indispensable to OpenAI’s financing as well as its computing, NVIDIA would deepen the relationship beyond the point where switching suppliers makes sense, a moat that no competitor can match with price cuts alone.
The guarantee would put NVIDIA in the unusual position of vouching for a customer whose spending drives its own revenue. The logic, analysts said, is defensive as much as aggressive: if OpenAI’s buildout stalls, NVIDIA’s growth slows with it, so backing the project protects the order book. And unlike a bank lending against cash flow, NVIDIA’s collateral is the infrastructure itself, data centers full of its own chips that could be sold or leased to other customers if the original occupant cannot pay.
The talks come as asset managers have begun warning that AI-related debt is growing faster than bond markets can absorb. Pimco, the world’s largest bond manager, cautioned this month that the volume of AI infrastructure financing could strain credit markets, and other investors have questioned whether the industry’s combined spending plans, which now exceed $1 trillion annually, can be funded without a crisis. Guarantees from chip makers and hyperscalers are one answer; they shift risk from lenders to the companies that benefit most directly from the buildout.
For SK hynix, the arrangement would lock in demand for years at a moment when its factories are running at full capacity. The memory maker has been a quiet winner of the AI boom, with high-bandwidth memory prices rising for several consecutive quarters as NVIDIA’s accelerators consumed every available wafer. A supply agreement tied to OpenAI’s program would extend that visibility well into 2027 and beyond, and it would strengthen SK hynix’s position in the memory market it shares with Samsung and Micron.
The talks also show how the industry’s biggest players are reorganizing risk. Microsoft, which has committed tens of billions to OpenAI, has structured much of its support as cloud credit; other operators have sold equity stakes to funds to pay for campuses; now the chip maker itself is being asked to carry part of the load. Each layer of guarantees makes the next one easier to arrange, and the pattern is pulling the entire supply chain into the business of financing the buildout.
People familiar with the talks cautioned that agreements of this size can still fall apart, and that the guarantee structure, the memory agreements and the equity terms all remain under negotiation. But the direction is clear: the line between chip maker and lender is dissolving, and NVIDIA is the company drawing it.
