NVIDIA has signed memorandums of understanding with six of the largest pools of private capital in the world — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — to create independent compute financing platforms the company says will mobilize more than $500 billion of third-party capital for AI infrastructure over time, according to the company’s August 10, 2026 announcement.
The platforms are designed to create dedicated pools of capital at scale and at what NVIDIA calls attractive rates for its customers, spanning frontier AI labs, enterprises and AI clouds. NVIDIA frames its compute as an investable asset, citing what it describes as the lowest token cost, longest useful life and a deep ecosystem of offtakers built on its CUDA software platform. The partnerships remain subject to execution of final agreements.
“In AI, compute is revenue,” Jensen Huang, NVIDIA’s founder and CEO, said in the announcement. “That is why we are bringing the world’s leading long-term capital providers together to independently underwrite AI infrastructure.”
Who is putting up the capital
The six firms signed onto the MOUs together manage several trillion dollars, by their own account:
- Apollo: approximately $1.05 trillion in assets under management as of June 30, 2026
- Blackstone: over $1.3 trillion in assets under management
- Brookfield: more than $1 trillion in assets under management
- Goldman Sachs, BlackRock and KKR: participating in investment, distribution and long-duration capital roles described in the release
Several of the firms positioned the deal as an extension of existing NVIDIA relationships. BlackRock chairman and CEO Larry Fink pointed to the AI Infrastructure Partnership, the data-center investment vehicle BlackRock launched with Global Infrastructure Partners, Microsoft and MGX that later added NVIDIA and xAI. KKR’s co-CEOs Joe Bae and Scott Nuttall noted NVIDIA is a founding investor in its Helix Digital Infrastructure platform. BlackRock has also been assembling direct exposure to AI campuses, including taking a majority stake in Meta’s El Paso data center and bankrolling that project through a $12 billion debt sale.
Goldman Sachs chairman and CEO David Solomon described his firm’s role as creating “a market for credit backed by NVIDIA compute,” language that signals the platforms may securitize or distribute compute-linked debt rather than only holding it.
Why NVIDIA wants its chips financed like infrastructure
The economic logic in the announcement is the part worth dwelling on. NVIDIA’s argument to capital providers rests on three claims: that its GPUs generate revenue for their operators (token sales), that CUDA software updates extend their productive life, and that the hardware is fungible across customers and operators if one tenant’s demand fades. Those are precisely the characteristics a lender needs to treat a rack of GPUs the way it treats a power plant or a toll road — as an asset with a long, underwritable cash-flow stream, rather than electronics depreciating toward zero.
The financing push also lands on top of NVIDIA’s DSX platform, the AI-factory design and operations playbook it unveiled at GTC Taipei on May 31, 2026, which Huang name-checked in the announcement as the factories the new capital will build. DSX standardizes reference designs, simulation and operations software for AI factories, and NVIDIA says cloud partners including CoreWeave, Crusoe, Lambda and Nebius are already deploying its components. Standardized factories are easier to underwrite than bespoke ones; a financing platform and a design platform solve the same problem from opposite ends.
The model follows a string of debt-funded buildouts across the sector, from Global AI’s first debt raise for sovereign AI data centers to Firebird’s 2-gigawatt AI factory pipeline.
What happens next
The near-term milestone is contractual, not concrete: the MOUs announced August 10, 2026 convert into final agreements, after which each platform’s structure, initial capital commitments and first financed projects become visible. The $500 billion figure is a mobilization target over time, not a committed pool, and NVIDIA’s release is explicit that the partnerships remain subject to execution of those agreements.
For NVIDIA’s customers, the practical effect to watch is the cost of capital on AI factory projects. Dedicated pools underwritten by six large institutions should price compute-backed debt more cheaply than project-by-project financing, which flows directly into the dollar-per-training-run and cents-per-million-token economics that decide who can afford to build.
