Nvidia's Reported $500 Billion AI Financing Talks Leave the Crucial Terms Unknown

A separate account said a $100 billion Nvidia investment agreement with OpenAI announced in late 2025 was ultimately reduced to $30 billion. That earlier episode does not determine the outcome of these talks, but it reinforces the need to distinguish an announced target from final commitments and funding.

Private capital is already financing AI compute

The group named in the talks matters because it combines firms that can structure long-lived infrastructure finance with Nvidia, the supplier at the centre of an expansion in computing capacity. Chips make up the bulk of the cost of new capacity, while the facilities also need data-centre construction and power production, the account said.

BlackRock's Global Infrastructure Partners is an infrastructure-investing platform, not an outside vendor to BlackRock: BlackRock completed its acquisition of GIP in October 2024 and retained it as GIP, a part of BlackRock. At that point, BlackRock said the platform managed about $170 billion, with a 600-person team and more than 300 active investments in over 100 countries, its announcement said. That provides a reason GIP belongs in a discussion of assets such as power and data centres; it is not evidence that it has committed money to this proposal.

Apollo, another named participant, is a global alternative asset manager that said it had about $1.03 trillion in assets under management as of March 31, 2026. It has recently financed a different route into AI infrastructure: in June, Apollo-managed funds and affiliates said they were leading an initial $35 billion capital solution for Broadcom's AI XPV Platform with Blackstone and global banks. Apollo said the platform was designed to enable more than 20 gigawatts of compute capacity for frontier AI labs through 2028 and to support Anthropic's expansion of more than one gigawatt of training and inference capacity from mid-2026, the company announcement said.

That $35 billion arrangement is not a like-for-like benchmark for an as-yet-unstructured Nvidia package. It does, however, show that private capital has an existing role financing a compute platform built around a different silicon provider, and that capacity figures and capital figures describe different things.

The headline figure does not settle the economics

Spending pressures help explain the talks without validating the $500 billion number. Large technology companies' combined AI outlays were set to exceed $730 billion in 2026, one account said. Separately, Morgan Stanley projected that hyperscalers would spend $3.5 trillion from 2026 through 2028, the earlier report said. The first is a one-year combined-outlay estimate; the second is a three-year forecast for a defined group, so they should not be added or treated as the same measure.

The reporting also raises a financing-governance issue, rather than proving one. The initial account said Nvidia often provides financial backing that helps AI partners raise debt, which in turn supports purchases of Nvidia equipment and its revenue; it said the circular nature of those transactions has prompted concern about concentrated risk. Separately, a person familiar with the matter said Nvidia was discussing a guarantee for a 10-gigawatt Ohio data-centre project leased to OpenAI, the report said.

Neither claim establishes that the reported package is circular financing. They identify the terms that would decide the issue: the independence of customer demand, the nature of any Nvidia guarantee and the party ultimately carrying utilisation risk.

What an announcement would need to disclose

The next useful evidence is not another round number. A credible disclosure would specify the investors' actual commitments and draw schedule; the split between debt, equity, guarantees and project finance; the data-centre and power assets being funded; and the customers or contracts expected to support repayment.

It would also need to show whether Nvidia is only the equipment supplier or has a financial obligation alongside the users of its chips. Until those details emerge, the reported $500 billion is best understood as the scale of an ambition to finance AI infrastructure—not a measure of capital already deployed.