Nvidia said Monday it has signed memorandums of understanding with six of the largest capital allocators on earth — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — to establish what it calls compute financing platforms, structures built to mobilise more than $500 billion of third-party capital into AI infrastructure over time.
The Financial Times reported the talks Monday morning, citing five people briefed on them. Nvidia confirmed the arrangement in its own announcement, describing dedicated pools of capital — at scale, at attractive rates — for its customers: frontier AI labs, enterprises and the AI clouds. The money is meant to buy the chips, the power generation and the data centres the buildout runs on.
“We began by building chips; today, we are helping create a new class of productive, investable infrastructure: AI factories,” founder and chief executive Jensen Huang said in the release. He told CNBC his chips now amount to an investable asset.
Nvidia shares closed down about 2.9%. The S&P 500 finished essentially flat.
Why a funding win reads like a warning
Read generously, Monday’s announcement answers the loudest criticism of the AI trade. For a year the complaint has been circularity: Nvidia putting capital into the labs and clouds that turn around and spend it on Nvidia silicon, with the chipmaker’s own balance sheet quietly underwriting its order book. Bringing six outside institutions in spreads that load across investors who have no stake in selling GPUs.
Read plainly, it also confirms what the criticism was pointing at. If this buildout could be paid for out of operating cash flow and ordinary corporate credit, it would not require the world’s biggest asset managers to assemble a new asset class to fund it. That is what a compute financing platform is: an admission that AI capacity has outrun the balance sheets of the companies that want it, and an attempt to turn compute into something leaseable, financeable and rateable instead.
Our take: The number that matters here is not $500 billion, it is the depreciation schedule. Financing works while utilisation stays high and each chip generation earns its keep for as long as the paper behind it takes to amortise. Nvidia has moved this risk off its customers’ balance sheets. It has not moved it out of the system — it has moved it into funds, insurers and private-credit vehicles, which is exactly where the last two credit cycles started.
Private capital was already the marginal lender to AI compute before Monday. Apollo and Blackstone led an initial $35 billion capital solution for Broadcom’s AI XPV platform in June. Nvidia itself has been discussed as a guarantor on roughly $250 billion of OpenAI’s data-centre rent. Moody’s has already flagged where Big Tech’s AI bill shows up in credit rather than in earnings. And CoreWeave’s $99 billion order book is the cleanest live test of whether a compute lender gets repaid on schedule.
What to watch
- The final agreements. These are memorandums of understanding, and Nvidia says the partnerships remain subject to execution of final documents. Announced intent is not committed capital.
- The structure. Lease, loan or special-purpose vehicle — and whether GPUs themselves are pledged as collateral. Someone has to put a residual value on a three-year-old accelerator.
- Guarantees. Whether Nvidia backstops any tranche. That single term is the difference between distributing risk and relabelling it.
- Who draws first. If frontier labs are the earliest borrowers, it says their own credit is stretched. If enterprises are, the demand base is genuinely broadening.
- Ratings treatment. How agencies score compute-backed paper will set the cost of capital for every data centre built after it — including the ones Nvidia is counting on.
Nvidia spent three years selling the picks. On Monday it started helping arrange the mortgages.