Goldman Sachs Creates High-Yield Bond Baskets for AI Capex: Blocks of Up to $250 Million per Transaction

Three Baskets, A New Market for AI Debt

Goldman Sachs made available on July 23 three high-yield bond baskets designed for asset managers and hedge funds to trade exposure to the AI capex cycle in block transactions. The first portfolio includes 18 U.S. issuers in equal weights, including CoreWeave, Applied Digital Corp., and Cipher Digital. The second focuses on 15 AI infrastructure companies. The third groups 16 issuers from the semiconductor and hardware sector, with Nvidia among them.

This product allows transactions between $50 million and $250 million, with a price agreed upon for the entire basket in a single transaction. Investors can buy the bonds directly or trade total return swaps on the baskets, which expands hedging and directional positioning possibilities. Wall Street had lacked a standardized instrument to manage the specific sector risk of AI bonds: before these baskets, assembling or dismantling a representative position required individual trades with each issuer.

What Spreads Reveal About AI Risk

The bonds in the baskets pay an average yield of 7.45% with a spread of 319 basis points over Treasuries, compared to a 7.3% yield and 267 basis points for the broader U.S. high yield index. The 52 basis point difference is not marginal: it indicates that the market demands a measurable premium for the credit risk associated with AI infrastructure, even as the capex cycle is in full expansion.

David Solomon, CEO of Goldman Sachs, framed the context during the second-quarter earnings conference call on July 15: "We're in the midst of a supercycle of AI capex, with demand for financing across every instrument, in every region of the world and across all sectors of the economy." In the same quarter, the bank registered revenue of $20.3 billion, a 39% increase year over year, and a net profit of $6.63 billion, a 78% jump. The most profitable bank of the cycle is also the one quantifying the risk premium that issuers in the cycle must pay.

The interpretation is not unanimous. Jeremy Barnum, CFO of JPMorgan Chase, noted in the bank's earnings call: "You really don't need the cutting-edge, incredibly expensive model to summarize an analyst report." This statement points to cost efficiency as an alternative response to the same cycle, which implies that some of the debt from AI infrastructure companies may be issued to support demand that is still being rationalized by the very buyers.

Global Reading: What Changes in Credit Markets Outside the U.S.

The creation of standardized baskets by Goldman signals that AI-related credit has reached sufficient scale to structure an organized secondary market in the United States. For U.S. managers, the immediate effect is a compression of transaction costs for operations that previously required fragmented bond-by-bond executions in a still thinly liquid market.

Barclays and BNP Paribas dominate the primary high yield market in the UK and France, but they have yet to launch equivalent instruments for AI infrastructure bonds. The project pipeline for data centers and on-premises computing in Europe, largely financed by local issuances in euros and pounds, lacks the same hedging mechanisms that the U.S. market is beginning to have. European credit analysts point out that this liquidity asymmetry may make the cost of financing AI infrastructure in Europe structurally higher in the short term than for equivalent projects issued in the U.S. market.

In Japan, MUFG has announced plans to invest over 70 billion yen in AI and has a growing interest in exposure to computing company credit. Goldman’s baskets provide the bank with a more liquid alternative to access this risk without building proprietary positions in individual bonds of American issuers.

Implications for Technology Decision-Makers

For CIOs and consultants advising clients on AI infrastructure projects, the 52 basis point premium has a direct implication on the financial model: projects financed through the capital markets will incur a structurally higher cost of debt compared to equivalent investments in other sectors. The difference is not marginal for projects involving hundreds of millions of dollars in GPUs, data centers, and energy contracts. Designing the total cost of ownership for a large-scale AI initiative now requires a separate line of credit for the infrastructure component, with an explicit risk premium.