J.P. Morgan: AI financing spreads widen: public bond issuance slows as equity funding plays a greater role

What has changed in the financing market?

As of the report, the spread on JPMorgan’s North American high-yield high-performance computing bond index had risen to 503 basis points, widening by 244 basis points over the past month. It reached 544 basis points last week when concern about emerging cloud providers was at its height.
Weaker prices slowed issuance in the public market. The volume of related bonds tracked by the bank rose from 602 billion dollars in July to 629 billion dollars as of the report, but has barely grown since August.
AI high-performance computing has become significantly more influential in the high-yield bond market: 41 billion dollars had been issued from the start of 2026 through the report date, accounting for 30% of that year’s non-refinancing high-yield issuance. The sector’s weight in the high-yield bond index rose from 1.07% at the end of 2025 to 3.44%.

Spreads on project bonds relative to those of the relevant cloud giants widened further. Investment-grade projects carried an average premium of 135 basis points, compared with 99 basis points in July; high-yield projects carried an average premium of 238 basis points, compared with 208 basis points in July.
As financing costs approach project returns, pressure to adjust deal structures rises. The bank said the average yield on cost is around 12% for data center leases and around 15% to above 18% for GPU leases, depending on the deal structure. If the all-in cost of high-yield financing rises above 10%, more deals may be restructured to seek investment-grade ratings.
JPMorgan believes that fundamentals and construction performance remain strong for projects, cloud giants and the broader ecosystem, and that overall credit capacity remains available. It sees investor expectations about the scale of future financing supply and willingness to provide it as the main drivers of spreads.

GPU financing has been concentrated in several large private transactions with a combined final size exceeding 100 billion dollars. These transactions use varying degrees of subordinated structures or vendor support to help some financing obtain investment-grade ratings.
JPMorgan believes this should not simply be understood as private markets replacing public markets: GPU financing is inherently more complex than data center financing, and many deals would have gone to private credit anyway. Some privately issued bonds are also ultimately intended for sale to traditional investment-grade bond buyers such as insurers and asset managers.
GPU financing currently appears to have maturities of less than 10 years. The pool of fixed-income capital is larger at shorter maturities, but investors also have more products to choose from, so GPU financing faces stronger competition on relative pricing.

So far this year, loans in related deals have represented 70% to 95% of project costs: emerging cloud providers are at the low end and hyperscale cloud providers at the high end, with the average close to 90%. The bank expects lower leverage and a higher equity share in future deals.
More equity means less debt needs to be issued for the same project cost. JPMorgan believes this helps ease bond supply pressure and support spreads; the stabilizing effect of equity issuance on financing costs over the past week also partly reflects market expectations of lower future credit bond supply.
This adjustment coincided with market volatility: in October, high-performance computing high-yield bond spreads widened by 123 basis points to 535 basis points, and the sector’s total return was negative 2.9%. As of the report, however, its year-to-date total return remained positive at 2.05%, above the overall high-yield bond index’s 0.55%.