Nvidia Wants to Raise $500 Billion to Build AI Factories. Here's What's Happening.
Nvidia has announced a $500 billion plan to pay for AI data centers and chip factories, with backing from six major financial firms: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. The company unveiled the plan during the week of August 13, 2026. An official press release dated August 10 confirmed the six partners would set up financing platforms to raise over $500 billion from outside investors Nvidia News. The Guardian reported the money would cover both data centers and chip factories The Guardian.
Here is the unusual part. Nvidia agreed to guarantee that its chips, when used as collateral for these loans, would hold their value. If the chips turned out to be worth less than expected, Nvidia promised to cover up to 25% of the difference out of its own pocket TechCrunch. The idea of using Nvidia's chips as collateral for loans started with a company called CoreWeave, which built its business by borrowing money against the chips it owned.
Think of it like a car loan where the carmaker promises the car will still be worth a certain amount in a few years. If the car loses value faster than expected, the carmaker has to make up some of the difference. Now imagine the carmaker is also the biggest seller of that type of car in the world, and the loan is for billions of dollars.
After bond markets reacted negatively to the arrangement, CEO Jensen Huang went on X and business television to explain that Nvidia's risk would be limited. Financial experts have called this a "wrong way" risk. That means Nvidia's obligations would grow at exactly the wrong time, when demand for its chips is falling. If GPU prices drop, the gap Nvidia promised to cover gets bigger at the same moment its main source of revenue is under pressure.
Nvidia has also committed billions of dollars separately to buyers of its chips, including AI companies OpenAI and Anthropic, and cloud providers CoreWeave, Nebius, Firmus, and Lambda. These commitments, combined with the $500 billion financing plan, create a web of connected obligations. Nvidia is simultaneously selling the hardware, guaranteeing the hardware's future value, and in some cases investing directly in the companies buying that hardware. Bloomberg reported that Nvidia was working on another $750 billion worth of similar deals during summer 2026, meaning the $500 billion is likely just one part of a larger plan TechCrunch.
CNBC flagged China-related risk factors in the plan, adding a political dimension to the financial exposure CNBC.
The announcement builds on partnerships Nvidia has been putting together over the past year. In May 2026, Nvidia and IREN announced a partnership to build up to 5 gigawatts of AI infrastructure, with IREN giving Nvidia the right to buy up to 30 million of its shares at a set price Nvidia News. In July 2026, NAVER, Nvidia, and Brookfield announced plans to expand an AI factory in Korea from 55 megawatts to 200 megawatts Nvidia News. In September 2025, Nvidia partnered with the United Kingdom and a company called Nscale to build national AI infrastructure, planning to scale 300,000 Grace Blackwell GPUs worldwide, including up to 60,000 in the UK Nvidia News.
The financing structure Nvidia has chosen deserves a closer look. The company is effectively turning its chips into an investment that generates returns, backed by Nvidia's own financial strength. Tech companies have long helped customers buy their products through financing and leasing. What is different here is the scale and the specific approach: Nvidia is not just helping people buy its chips, it is promising that those chips will hold their resale value.
The "wrong way" risk label matters because it points to a problem ordinary financing does not create. In a normal deal where a company helps a customer buy its product, the company's risk is whether the customer can pay. Here, Nvidia's risk is tied to the market value of its own product. If demand for AI computing drops, chip prices fall, cloud providers make less money, and Nvidia's guarantee kicks in all at the same time. Nvidia is betting that demand for AI will stay strong enough that its 25% guarantee rarely gets used. If that bet pays off, the plan unlocks huge amounts of money for building infrastructure at low cost. If it does not, everything moves in the wrong direction at once.
There is a genuine upside to what this plan would make possible. Five hundred billion dollars, combined with the earlier IREN, NAVER, and UK partnerships, would fund the data centers and chip factories that AI requires to operate at scale. These facilities take years to build and cost enormous amounts up front. Money from large institutional investors like BlackRock and Brookfield is the kind of long-term funding that can cover those costs. The question is whether Nvidia's promise to guarantee chip values is a reasonable price to pay, or a hidden weakness that would only show up in a downturn.
The $750 billion in additional deals Bloomberg identified suggests Nvidia is not treating the $500 billion as a limit. Together, the total pipeline approaches $1.25 trillion in financing backed by GPU values. Whether investors will keep accepting Nvidia's chips as reliable collateral, and whether Nvidia can maintain its guarantee at that scale, are the two factors that will decide whether this whole approach holds together.
