Nvidia Pitches $500B AI Compute Financing as New Asset Class

Nvidia has unveiled plans with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise more than $500 billion in capital for AI data centers—what CEO Jensen Huang calls AI "factories." The initiative aims to establish AI compute infrastructure as a distinct investable asset class, similar to real estate or energy projects.

The announcement consists of memorandums of understanding rather than binding contracts, with no timeline, capital allocation breakdown, or named first projects. The $500 billion represents a fundraising target over time, not immediate revenue for Nvidia. Potential borrowers include AI labs, large enterprises, and cloud providers that lease computing capacity.

The core financing challenge

The viability of AI compute as an asset class hinges on a fundamental tension: can rapidly evolving hardware be financed using the long-term structures typical of infrastructure investments?

Traditional infrastructure lending relies on two pillars—contracted revenue streams and residual asset value. For AI data centers, the first depends on customer commitments to use the computing capacity. The second requires that the hardware retains meaningful value if those customers exit before loans mature.

Nvidia has indicated it may provide "residual-value support for up to 25% of an opportunity" on a case-by-case basis, though specifics remain undisclosed. This partial backstop likely makes outside capital more accessible, but the structure of loss absorption in default scenarios remains unclear until contracts are finalized.

Why it matters

This financing push arrives as AI infrastructure spending accelerates but faces growing scrutiny over returns. If successful, it would unlock massive capital for AI buildout without requiring chip buyers to fund purchases entirely from operating budgets or balance sheets. For investors, it creates exposure to AI infrastructure returns without direct technology risk—assuming the hardware economics hold. The model's success or failure will shape how the next wave of AI capacity gets funded and whether compute joins the ranks of established alternative assets.

The depreciation warning

Amazon's recent accounting change signals caution. Effective January 1, 2025, the company shortened the estimated useful life of certain servers and networking equipment from six years to five, citing "the increased pace of technology development, particularly in the area of artificial intelligence and machine learning." The adjustment added approximately $1.4 billion to 2025 depreciation expense and reduced net income by roughly $1 billion, primarily at AWS.

While Amazon did not specifically identify Nvidia chips, the move demonstrates that a major AI operator now expects some infrastructure to age faster than previously assumed. Investor Michael Burry estimated in November 2025 that large cloud providers were understating AI depreciation by about $176 billion from 2026 through 2028, though this remains his projection rather than reported figures.

Nvidia counters that its A100 chips from 2020 continue to attract multi-year commitments, potentially extending useful life toward a decade. The company also argues that its CUDA software platform enables ongoing performance improvements on installed hardware, supporting both cash flow and redeployment potential.

Rental rates versus resale value

Huang cited rising rental prices as evidence of durable demand, noting that one-year H100 rental rates climbed from approximately $1.70 per hour in October 2025 to $2.35 in March 2026. However, rental income and residual value answer different questions. Strong lease rates demonstrate current earning power but do not guarantee robust resale prices, particularly if newer chip generations arrive before loans are repaid.

The ultimate test will come in the contract terms: which customers have committed to capacity, whether loan maturities align with hardware refresh cycles, who can redeploy the equipment, and who absorbs the first loss if resale value disappoints. As the first deals move from memorandum to signed agreement, those details will determine whether AI compute truly functions as an asset class or simply repackages technology risk as infrastructure debt.

These details were first reported by Robert J. Szczerba in Forbes.