Nvidia's $500bn Wall Street pact turns AI chips into assets

Nvidia confirmed it had signed memoranda of understanding with six of the largest names in global finance (Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs and KKR) to mobilise more than $500 billion in third-party capital for AI infrastructure.

The announcement followed a Financial Times report that leaked the talks earlier in the day and briefly knocked Nvidia’s share price. By the close of trading, the rumour had become a formal pact, and one of the more consequential financing structures the data centre sector has seen.

Rather than customers paying upfront for the GPUs, power and facilities that go into an AI data centre, the new platforms will let Nvidia’s hyperscaler, cloud and AI lab customers borrow against that hardware, with the six financial institutions supplying “dedicated pools of capital at significant scale at attractive rates”.

Nvidia has said it may backstop up to $125 billion of the total, or roughly a quarter of the facility, effectively underwriting a slice of the credit risk itself.

Speaking in a joint CNBC interview with executives from all six partner firms, Nvidia founder and chief executive Jensen Huang framed the move as a categorical shift for the company. “We began by building chips; today, we are helping create a new class of productive, investable infrastructure: AI factories,” he said.

It is a line that recasts Nvidia not merely as a hardware vendor, but as an architect of a new asset class, one in which GPUs and the data centres that house them behave less like rapidly depreciating IT kit and more like toll roads or commercial property: assets with a long, financeable revenue tail.

That framing will be familiar to anyone who has followed the sector’s financing story this year. It echoes, in scale if not structure, Apollo and Blackstone’s $35bn chip-backed special purpose vehicle for Anthropic, which Capacity described in June as a template rather than an outlier. It also sits close behind reports that Nvidia was separately in talks to guarantee financing for OpenAI’s roughly $500bn Ohio data centre project with SoftBank. Whether that arrangement now folds into the new platform, or runs alongside it, is one of the more pressing questions the announcement leaves open.

What the deal actually changes

Goldman Sachs chief executive David Solomon called the tie-up “a pivotal moment of a historic AI investment cycle”, while Blackstone president Jon Gray pointed to the demand side of the equation, noting that AI usage among the firm’s portfolio companies has grown sevenfold this year and that demand is outpacing supply. Apollo president Jim Zelter, meanwhile, has framed modern computing as a genuinely scarce resource, one that institutional capital is now structuring around rather than simply reacting to.

Morgan Stanley has estimated that tech companies and related parties could need to raise up to $800 billion in private credit through asset-specific structures by 2028. JLL’s most recent Global Data Centre Outlook puts total data centre investment at closer to $3 trillion over five years. Set against those figures, Nvidia’s $500 billion is not the ceiling; it is closer to a sixth of the sector’s likely capital requirement, arriving through a single, coordinated channel rather than the deal-by-deal patchwork that has characterised the market until now.

That coordination is itself the story. Capacity has tracked the individual pieces of this financing wave as they appeared: Google’s $150bn web of chip and infrastructure contracts underpinning Anthropic, the Anthropic SPV that first proved chip-backed private credit could be priced at investment-grade rates, and BlackRock, Microsoft and MGX’s AI Infrastructure Partnership, which Nvidia itself joined as an investor. What changes this week is that Nvidia is no longer a participant lending its name and its guarantee to other people’s structures. It is the counterparty of record, setting the terms on which its own customers access capital.

Where coverage needs to go next

The public statements answer the “what”. They say almost nothing about the “how”, and that is where the next wave of reporting has an opening. Nvidia has not disclosed individual commitments from each of the six firms, the interest rates on offer, or a deployment timetable, leaving several threads worth pulling.

The first is collateral quality. GPU-backed lending depends on the assumption that Nvidia hardware holds resale value across multiple customers and generations. Rapid architecture cycles complicate that assumption, and any lender underwriting a 10-year facility against three-year-old silicon is making a bet on obsolescence timelines that nobody in the industry can fully model yet.

The second is concentration risk. As Axios has noted, a financing structure in which the dominant supplier also backstops its customers’ borrowing revives long-standing concerns about circularity in AI capital flows, echoing the vendor-financing schemes that Lucent and Nortel ran for telecoms equipment in the late 1990s, with painful results when customers folded. Capacity’s own coverage of Meta’s repriced $12bn El Paso bond already shows lenders beginning to differentiate between AI infrastructure deals rather than treating the sector as a uniform, hyperscaler-backed risk. This platform will be an early test of whether that differentiation holds at scale.

There is also a geographic and regulatory gap. Every firm named in the pact is US-based, and the announcement is silent on how the facility will interact with sovereign and regional financing vehicles already active in Europe, the Gulf and Asia. Given how much of the sector’s near-term capacity growth sits outside the US, that is a natural line for further reporting, alongside the practical question of how smaller developers and colocation operators, rather than hyperscalers and frontier labs, might access these capital pools at all.

What it signals for the next wave of deals

If the Anthropic SPV proved that chip-backed private credit could be priced at investment-grade rates, this deal suggests that model is about to be industrialised. Rather than negotiating bespoke structures deal by deal, as Google, Meta and OpenAI have each done in different ways this year, Nvidia has built standing infrastructure that its customers can, in principle, draw on repeatedly. That is a meaningful shift in how quickly capital can move from commitment to construction, and it lowers the barrier for AI labs and cloud providers without an investment-grade balance sheet of their own to secure debt on comparable terms to the hyperscalers.

It does not resolve the questions Capacity has been raising all year about counterparty concentration, debt maturity mismatches, or what happens to collateral values if AI demand growth slows before the hardware depreciates. What it does is make Nvidia, rather than any single customer or lender, the fulcrum on which those questions now turn. For an industry already watching capital costs as closely as power availability, that is the detail worth following.

Data centres could use 2.7% of Texas’s water by 2030: Where they land matters