Credit issued by data centers under long-term lease to Amazon, Microsoft, and Google is clearing at yields 150 to 200 basis points above the corporate bonds of those same hyperscalers, despite identical counterparty risk and contractual revenue streams extending ten to fifteen years. The divergence surfaced in European fixed income desks this week as allocators began comparing yields on recent data center project finance against the September corporate issuance from the hyperscalers themselves.
The structural setup is straightforward. Data center operators — typically infrastructure funds or specialized REITs — issue senior secured debt backed by triple-net lease agreements with hyperscalers. The leases are absolute obligations, inflation-indexed, with minimal tenant optionality before year ten. Covenant packages mirror investment-grade corporate structures. Yet the debt trades at yields consistent with high-yield real estate, not the A to AA ratings the underlying obligors carry. A $600 million tranche from a Northern Virginia facility leased to Amazon Web Services cleared at 6.8% in October; Amazon's own ten-year corporate debt issued the prior month yields 4.9%.
The spread exists because bond investors apply structural subordination logic even when operational risk is contractually eliminated. Hyperscaler corporate bonds sit senior in the capital structure to every obligation the company holds, including real estate leases. Data center debt, despite being secured by physical assets and lease cash flows, ranks junior to the corporate bondholders if the hyperscaler enters distress. That theoretical subordination commands a premium, even though the probability of a scenario where Amazon defaults on corporate debt but continues paying rent is effectively zero. The market is pricing legal hierarchy over economic reality.
What makes the dislocation actionable now is supply. Hyperscalers are adding 40 to 50 gigawatts of data center capacity globally through 2026 to support AI training infrastructure, much of it financed off-balance-sheet through sale-leaseback or build-to-suit structures. Credit desks estimate $18 to $22 billion in new data center project finance will print in the next eighteen months, concentrated in Dublin, Frankfurt, Northern Virginia, and Singapore. Simultaneously, corporate bond desks at the hyperscalers themselves are issuing conservatively — Microsoft has $21 billion in undrawn revolver capacity and no near-term need for benchmark issuance. The result is excess supply in the project finance channel and scarcity in the corporate channel, widening spreads independent of credit fundamentals.
Allocators should monitor three follow-on events. First, whether any hyperscaler consolidates a data center portfolio onto its balance sheet in the next six to nine months, which would immediately compress spreads by eliminating structural subordination. Second, rating agency treatment of these leases as contingent liabilities — if Moody's or Fitch begin penalizing hyperscaler corporate ratings for off-balance-sheet real estate exposure, the theoretical spread justification collapses. Third, whether private credit funds begin warehousing these assets for CLO issuance, which would formalize the spread as a permanent feature rather than a temporary arbitrage.
The fixed income desks that moved early on this trade are running $400 million to $1.2 billion positions, levered two-to-one in total return structures. They are not betting on hyperscaler credit improvement — they are betting the market stops paying 200 basis points for a legal distinction without an economic difference.