AI Chips Become Collateral: $869M ABS Has 15% GPU Loans

For AI teams, the bottleneck has often been compute, not code. Now Wall Street is turning GPUs into a financed asset class: Stonebriar's $869M ABS includes 15% GPU loans, Wingspire's $407M deal includes 20%, and Broadcom's lenders are arranging over $60B to help Anthropic and others buy chips. That could accelerate hardware access across the industry.

AI chip financing is breaking into structured credit. Stonebriar Commercial Finance is selling around US$869 million of asset-backed securities, with GPU loans representing about 15% of the collateral, according to people with knowledge of the matter. Wingspire Equipment Finance late last month raised more than US$407 million through an equipment loan and lease ABS in which GPU loans made up about 20% of the collateral. These two deals are among the first broadly syndicated equipment-financing ABS backed in part by AI chip loans, and they show how capital demand for artificial intelligence is moving from data centres to the semiconductors inside them.

“

Now Wall Street is turning GPUs into a financed asset class: Stonebriar's $869M ABS includes 15% GPU loans, Wingspire's $407M deal includes 20%, and Broadcom's lenders are arranging over $60B to help Anthropic and others buy chips.

For years, the ABS market has securitized equipment loans and leases covering aircraft, railcars, construction machinery and IT gear. But dedicated GPU exposure has historically been minimal or absent. That is changing as AI buildouts require enormous amounts of compute hardware. Broadcom Inc's Wall Street lenders are working on more than US$60 billion of financing that could help Anthropic PBC and other companies access AI chips. A group of banks is also providing a US$22 billion chip loan to Blackstone Inc and Alphabet Inc's new cloud venture. The Stonebriar and Wingspire deals are small relative to those figures, but they matter because they route GPU-backed loans into broadly syndicated bond markets, giving traditional fixed-income investors exposure to AI hardware economics.

From a structured-credit perspective, the collateral mix matters. Michael Nowakowski, head of structured products at Conning & Co., said equipment ABS have not typically had GPU financing as collateral in the past, though a few recent deals had relatively little exposure. He noted that as long as loans are amortising with lower loan-to-values and the percentage doesn't start to creep up over time, investors are not as concerned as they would be with 100% GPU finance deals because collateral pools are often very diverse. That framing captures the central investor debate. A 15% or 20% GPU share inside a diversified equipment pool is manageable; it adds residual-value risk tied to rapid semiconductor obsolescence, but the overall pool still includes many other asset types with more stable secondary markets.

The biggest risk is that GPU loans could become a larger share of future deals. GPUs lose value quickly as new architectures are released, and collateral performance depends on whether borrowers keep making payments and whether used chip values can support recovery if loans default. If GPU exposure creeps toward 50% or more, ABS investors would likely demand wider spreads, shorter maturities, higher credit enhancement or explicit residual-value stress tests. For now, the Stonebriar and Wingspire transactions provide early data points rather than a broad market shift. Their moderate shares suggest both issuers and underwriters are testing investor appetite without overloading pools with unproven collateral.