Nvidia and Financial Giants Form $500 Billion Compute Financing Consortium

Nvidia has partnered with a consortium of major financial institutions, including BlackRock, Blackstone, Apollo, Goldman Sachs, and KKR, to establish a $500 billion financing facility. The initiative is designed to provide capital to neocloud providers for the purchase of graphics processing units, effectively reclassifying compute power as an investable asset class. Nvidia CEO Jensen Huang has framed these chips as long-lived, revenue-generating assets with a depreciation period of up to a decade, a shift from previous narratives regarding rapid obsolescence.

The move aims to reduce Nvidia’s reliance on its own capital for customer financing and mitigate accusations of circular financing. While the agreements are currently memorandums of understanding, Nvidia has indicated it may provide residual value support, potentially covering up to 25 percent of losses on some contracts. This development coincides with CME Group’s plans to introduce compute futures in October, pending regulatory approval.

Huang has publicly stated that Nvidia chips are revenue-generating assets that are productive, long-lived, and fungible. This assertion contrasts with comments made last year, where he suggested that older Hopper chips would be difficult to sell once the new Blackwell architecture shipped in volume. The company argues that software integration, specifically CUDA, allows installed bases to remain productive well beyond their initial depreciation periods, extending the economic life of chips such as the A100 towards a decade.

The financing structure mirrors similar arrangements in the broader market, such as a $35 billion package announced by Broadcom with Apollo and Blackstone. In those deals, private credit firms and banks provide funding secured by chip collateral. The new consortium includes one bank, Goldman Sachs, alongside several private credit firms, marking a shift in how the AI infrastructure build-out is funded. Hyperscalers and related companies have issued approximately $225 billion in bonds by mid-year, reflecting a broader pivot to debt financing for data centre expansion.

Analysts note that the arrangement could standardise how Nvidia chips are installed in data centres to satisfy lender requirements for uniformity. By providing outside capital, Nvidia seeks to make its products more accessible to neoclouds without cutting GPU prices directly. However, the deals remain non-binding memorandums of understanding, and the long-term viability of the asset class depends on sustained demand for inference and the profitability of the neocloud providers securing the loans.