NVIDIA said on August 10, 2026, that it had signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish independent financing platforms for AI compute infrastructure. The announced target is to mobilize more than $500 billion of third-party capital over time.
That figure does not represent money already raised, committed or deployed. The parties have outlined a financing architecture, but the platforms remain subject to final agreements.
What the parties have agreed so far
The MOUs contemplate dedicated pools of capital intended to finance NVIDIA-based infrastructure for customers including frontier AI laboratories, enterprises and AI cloud providers. NVIDIA says the proposed platforms are meant to give those customers access to large amounts of compute without every infrastructure project relying solely on capital supplied directly by its operator.
The announcement does not identify the financing instruments the platforms would use. Debt, leases, equity participation or combinations of those structures may be possible, but none has been specified. NVIDIA’s description of capital being available at “attractive rates” is an intended outcome, not a disclosed or independently tested financing term.
How the compute-financing model is meant to work
In practical terms, the proposed platforms would connect institutional capital with projects that build or acquire infrastructure based on NVIDIA systems. The financial institutions would independently assess and underwrite opportunities, while NVIDIA would supply the compute platform around which the projects are constructed.
Reuters described the intended investments as longer-duration and linked to infrastructure usage. That framing treats compute capacity as an income-producing asset whose economics depend on customers continuing to pay for access. It is not yet a disclosed cash-flow model: the parties have not published utilization assumptions, pricing structures, collateral terms, residual-value estimates or the allocation of construction and technology risks.
Why NVIDIA calls compute an investable asset
NVIDIA’s commercial case is that accelerated-computing systems can support different models, workloads, customers and operators, while its CUDA software ecosystem can extend their useful role over time. On that basis, the company describes compute as productive infrastructure capable of generating revenue from usage rather than merely as equipment purchased for a data centre.
The participating capital providers similarly frame AI compute as a new infrastructure category suitable for long-duration investment. Those descriptions are the business thesis behind the MOUs, not evidence of guaranteed demand, utilization, returns or useful life.
What remains unresolved
Reuters reported that no financial terms, individual investment commitments or deployment timetable were disclosed. The announcement also does not identify initial projects, customers, geographies or the amount each platform might manage.
If final agreements are executed, the platforms could broaden the financing available for expensive AI infrastructure. They would also make underwriting depend on uncertain variables including compute demand, utilization, power availability, construction delivery and the pace at which hardware becomes obsolete.
The next material developments will be signed final agreements, named projects and customers, disclosed capital commitments, financing terms and deployment schedules. Until then, the $500 billion-plus figure should be read as a long-term mobilization target attached to MOUs—not as a completed fundraising round or an installed base of AI infrastructure.
