Nvidia moves from chips to capital with $500B financing plan and $10.5B OpenAI data center backing

The chipmaker said its financial support for the Ohio site, located at the PORTS-Pike Technology Campus in Pike County, covers about 4 gigawatts of development for portions of lease and power, as well as a specified residual-value commitment, with data centers opening between 2028 and 2030, according to CNBC. The agreement includes a $1.5 billion investment in SB Energy, a SoftBank affiliate that is building and managing the facility under a 20-year lease to OpenAI. Nvidia had previously invested $30 billion in OpenAI in February.

Under last week's memorandum of understanding with Apollo Global Management, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, Nvidia plans to help establish independent computing-financing platforms that could eventually mobilize over $500 billion in third-party capital, Nvidia said. The firms are expected to act as capital organizers and risk underwriters, pooling money from third-party investors to fund GPU-backed projects. Nvidia will have the option to backstop up to 25% of each loan, a structure that CEO Jensen Huang said turns GPUs into a new asset class. "These are revenue-generating assets now," Huang told CNBC. "They're productive, they're long-lived, they're fungible, they're flexible." Cantor analysts wrote in a client note that the agreement is a "clear signal that the current AI investment cycle will be elongated and durable," adding that it "facilitates the coming AI buildout" while creating competitive moats.

Nvidia's cash generation has soared during the AI boom, with quarterly free cash flow reaching $48.5 billion in the latest period, up 18-fold in three years. The company this year lifted its quarterly dividend from a penny to 25 cents a share and announced a new $80 billion stock buyback plan. It held $30.2 billion in marketable equity securities as of the most recent quarter, up from $12.9 billion a year earlier.

The financing push addresses a gap in the AI credit market, according to a detailed analysis by Chinese tech outlet Leiphone. SemiAnalysis estimates AI-related debt balances will exceed $7 trillion by 2029, while annual AI capital expenditures will top $2 trillion by 2028. But lenders have mostly been willing to fund only long-term compute contracts of five years or more backed by hyperscale cloud providers, leaving AI startups and inference providers that need shorter, flexible commitments struggling to secure loans. Leiphone reported that CoreWeave, a GPU cloud provider, paid a floating-rate spread of 2.25 percentage points on an $8.5 billion loan backed by cloud contracts in March, but two months later was charged 4.5 percentage points on a $3.1 billion facility backed by two weaker AI model companies. To close the gap, Nvidia in July began offering take-or-pay commitments to some neocloud projects, guaranteeing minimum revenue for up to six years in exchange for a share of profits above a threshold.

Leiphone also detailed the structure of a separate Google-Anthropic deal, which it says shows how chip financing can work. Google sold TPU systems to Broadcom, which then used a special-purpose vehicle with debt arranged by Apollo's Atlas SP Partners and Blackstone to buy roughly 1 million TPUs for $35 billion and lease them to Anthropic. Broadcom provided residual-value support of about $30 billion to protect senior debt holders, while $4.4 billion in subordinated debt carried no backstop.

These structures carry risk. GPU values may fall faster than debt matures, as Nvidia and AMD have accelerated product release cycles to about one year, and some analysts estimate GPU usable lives of just two to three years, according to Quinn Emanuel, a law firm cited by Leiphone. Constellation Research questioned whether AI compute demand will translate into sustained high-end GPU spending, noting that some enterprises, like Uber, have reduced unit costs by switching to cheaper models. Huang has acknowledged Nvidia may provide residual-value support of up to 25% for some financing opportunities, but it cannot fully remove the risk of demand and residual-value shortfalls.

In a separate development, Reach Capital announced the close of a $265 million Fund V to back AI founders building applications that "expand human potential," the San Francisco firm told TechCrunch. The fund will write checks from $1 million to $10 million across pre-seed through Series A, targeting around 50 companies in learning, health and work. Limited partners include Capricorn Investment Group, the Los Angeles Fire and Police Pensions, the LEGO Foundation and College Board. The firm's previous exits include GPTZero, an AI-detection startup acquired by Superhuman in June, which had grown to over 19 million registered users and $30 million in annual recurring revenue.