AI bonds’ weak points are becoming clearer

By the Editorial Team · Thursday, October 22, 2026

The bond market is increasingly differentiating credit risks among AI infrastructure financiers, showing greater caution toward smaller AI computing firms like CoreWeave while favoring data center bonds tied to major cloud providers such as Microsoft. This nuanced risk assessment reflects growing investor attention to factors like construction delays, lease security, and local regulatory challenges, signaling a more segmented and sophisticated approach to financing AI-related physical infrastructure.

The bond market is increasingly distinguishing risks among players involved in financing artificial intelligence (AI) infrastructure, particularly highlighting vulnerabilities in some AI computing providers while showing comparatively less concern about the data centers that host this computing power.

Credit market activity this year reflects significant growth in financing for AI-related data centers. Through October 16, issuers associated with data center development raised over $100 billion in U.S. dollar bonds, accounting for roughly 25% of AI-related bond issuance in U.S. currency—a notable increase from about 16% the previous year. This surge underscores the capital-intensive nature of building the physical infrastructure needed to support growing AI workloads.

Data center bonds linked to hyperscale tenants—major cloud providers such as Microsoft—generally command lower yields than the bonds of the tenants themselves. Microsoft, for example, maintains a triple-A credit rating and is considered a reliable lessee, reducing default risk on leases. However, financing data centers carries additional risks, including potential construction delays or cancellations, which typically result in their bonds trading at higher yields compared to tenant corporate debt.

The credit risks diversify further when examining smaller or more leveraged players like CoreWeave, a relatively new entrant in AI cloud computing, sometimes described as part of a group dubbed "neoclouds." CoreWeave competes with established giants such as Amazon and Microsoft, relying heavily on leased data center capacity to meet rapidly expanding demand. Unlike these tech behemoths, CoreWeave’s bonds are rated below investment grade and trade in the high-yield segment, reflecting greater credit risk.

Recent market data illustrate this divergence: CoreWeave’s bonds due 2030 yield close to 15%, whereas bonds issued last year to fund Applied Digital’s North Dakota data center—leased to CoreWeave—offer yields near 8%. The lower yields on data center bonds reflect their secured nature; these bonds often are backed by the physical assets under construction, unlike CoreWeave’s unsecured corporate debt. However, the actual security depends on the successful completion of the facility and its usability either by the current tenant or alternative users in case of tenant default.

Some data center bonds incorporate structures that further mitigate risk, such as timing principal repayments to coincide with the lease term, reducing dependence on securing subsequent tenants to cover outstanding debt. This was evident when CoreWeave restructured a lease for an Applied Digital facility, spinning off the lessee into a separate investment-grade entity, a move that resulted in a notable tightening of the corresponding bond spreads.

Despite these mechanisms, uncertainties remain over risks related to construction, lease commitments, and political factors. Rising local resistance to data center projects in certain communities may be influencing investor concerns about completion risks and demand viability. Paradoxically, if regulatory hurdles limit new data center construction, existing facilities may gain in value due to constrained supply.

Overall, as the AI data center financing market matures, investors are expected to continue differentiating credit risk more granularly across various business models and bond structures rather than treating the sector as a homogeneous group.

This article was reported by synthesizing coverage from multiple independent news outlets to present a balanced, unbiased account. Read about our editorial standards. Spotted an error? Let us know.