AI Data Center Debt Offers Investors Huge Yields — and New Risks

SoftBank's data-center-related bond sale is pricing at yields near 10%, signaling that financing costs for AI infrastructure are rising sharply. Higher borrowing costs add to already escalating data-center construction expenses and could weaken expected returns on the trillions of dollars being committed to the AI buildout.

Analysis

The relevant repricing is not simply higher interest expense; it raises the hurdle rate for every marginal AI campus and shifts bargaining power toward counterparties with investment-grade balance sheets, contracted tenants, and access to unsecured debt. DLR and EQIX should gain share as smaller developers face wider project-level financing spreads, while APLD, CORZ and other capital-intensive HPC hosts become more dependent on pre-leased capacity, customer deposits, and dilutive equity. The first-order effect on VRT and ETN is likely order-timing risk rather than outright demand destruction: customers can defer construction starts while preserving multi-year compute plans.

Over the next 1-3 months, the key transmission channel is whether debt-market stress forces revisions to announced campus capex or changes contract terms from take-or-pay to more contingent structures. A 200bp increase in all-in funding cost materially erodes project equity returns for assets with long construction periods and utilization ramps, particularly where power interconnection is not secured. This should favor CEG, VST and other power suppliers with contracted load growth over data-center landlords whose returns depend on refinancing and terminal-value assumptions.

Consensus may be too quick to extrapolate a financing shock into an AI-demand collapse. Hyperscalers with net cash and strategic need for capacity can absorb higher financing costs or self-fund projects; the pain is concentrated in merchant developers and sponsor-backed intermediaries. The structural consequence over 6-18 months is likely consolidation and tighter supply, which can ultimately support rents and utilization at scaled incumbents, even as equipment bookings become more volatile near term.

Key Decisions for Investors

• Initiate a 3-6 month pair: long EQIX / short APLD. EQIX has diversified recurring revenue and superior capital-market access; APLD is more exposed to construction funding, customer-concentration and execution risk. Reassess if APLD secures fully committed non-recourse financing plus long-duration take-or-pay contracts.
• Maintain VRT and ETN as watch-list longs rather than add aggressively before the next earnings cycle. Add only if management confirms backlog conversion and customer funding are intact; reduce if book-to-bill or delivery schedules weaken, as financing pressure will hit equipment timing before end-demand.
• Overweight CEG or VST versus data-center development exposure over 6-12 months. Contracted electricity demand and scarce interconnection capacity can capture the scarcity premium even if individual campus developers fail; thesis is falsified by material power-demand forecast reductions or adverse state-level rate/regulatory outcomes.
• Monitor high-yield spreads and new-project financing terms as the actionable trigger. A sustained widening of roughly 100bp or more from current levels, combined with delayed project announcements, would justify increasing shorts in leveraged HPC/data-center developers; absent that confirmation, avoid treating one issuer's funding cost as sector-wide evidence.