Nvidia backs $500B data center plan with GPU guarantees
Nvidia has partnered with major financial institutions on a $500 billion data center initiative, offering to guarantee the value of its own GPUs to secure the massive funding.
Nvidia has secured commitments of up to $500 billion from major financial firms, including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, to fund new AI data centers. To convince these institutional investors, Nvidia is taking the unusual step of guaranteeing the residual value of its chips when used as collateral. Specifically, if a borrower defaults and the liquidated graphics processing units fail to fetch their projected value, Nvidia has promised to cover up to 25 percent of the financial shortfall.
This arrangement introduces what financial experts call wrong-way risk, meaning Nvidia's financial liabilities will expand if market demand for its hardware declines. Critics have compared the strategy to the downfall of Lucent Technologies, which collapsed after lending money to its own customers during the dot-com bubble. Nvidia chief executive officer Jensen Huang has defended the initiative, arguing that bringing in independent, long-term institutional capital prevents the circular financing trap. However, Bloomberg calculated that Nvidia was already working on an additional $750 billion in circular deals earlier this summer.
Beyond immediate sales, the strategy aims to establish a robust secondary market for aging hardware. Nvidia has already backed several prominent buyers of its chips, including frontier artificial intelligence labs OpenAI and Anthropic, alongside cloud providers like CoreWeave, Nebius, Firmus, and Lambda. By guaranteeing the value of older chips, Huang hopes to transform AI servers into long-term investable infrastructure, which he compares to factories or railroads rather than rapidly depreciating personal computers.
For startups and enterprises, a thriving secondary market for used GPUs could democratize access to AI compute. As hyperscalers like Oracle, Google, and Meta face rising debt and massive capital expenditures, alternative clouds and smaller developers may soon find a steady supply of affordable, older Nvidia hardware. This ecosystem could allow researchers to match their specific workloads with cheaper, previous-generation chips, mirroring the way developers currently choose budget-friendly open-weight models over expensive frontier systems.
This is our own summary of reporting by TechCrunch AI
