Morgan Stanley projects that AI-related bond issuance will reach $570 billion in 2026, doubling the $280 billion issued in 2025 and marking the fastest single-year credit expansion in any technology sector's history. The projection, published in the bank's Monday credit strategy note, frames the AI infrastructure buildout not as a technology cycle but as a credit cycle with its own leverage dynamics, refinancing risks, and default correlations [1].

The $570 billion encompasses bonds issued by data center operators, semiconductor manufacturers, cloud infrastructure providers, and the utility companies building power generation capacity to feed AI compute demand. Microsoft, Alphabet, and Amazon account for approximately 40% of the projected issuance, with the remainder coming from hyperscale-adjacent companies and specialized infrastructure funds [2].

Morgan Stanley's framing as a credit cycle is the divergence from MSM coverage. CNBC and Bloomberg lead with the scale of capital flowing into AI as a bullish signal. Morgan Stanley's note emphasizes the structural similarity to prior credit-driven buildouts — the railway boom of the 1840s, the telecom buildout of the late 1990s, and the shale energy expansion of the 2010s. Each produced real infrastructure. Each also produced a credit overhang that outlasted the initial demand cycle [1].

The Refinancing Wall

The $570 billion in 2026 issuance creates a refinancing concentration between 2029 and 2031, when the first wave of five-year bonds matures. If AI compute demand grows as projected, the revenue to service that debt exists. If demand plateaus — as it did after the telecom buildout — the refinancing wall becomes a stress event for the companies that issued the bonds and the banks that underwrote them [2].

The leverage ratio is the number that separates this from prior technology cycles. AI infrastructure companies are issuing debt at 4-6x EBITDA, compared to 2-3x for software companies in the 2010s. The higher leverage reflects the capital intensity of physical infrastructure — servers, chips, power plants, cooling systems — but it also means the margin for error is thinner. A 20% revenue shortfall against projections produces a coverage ratio that triggers covenant concerns [1].

The Utility Layer

The utility companies financing AI data center power demand are the most exposed layer of the credit stack. Data centers require 24/7 baseload power, and the utility bonds issued to build that capacity assume decades of consistent demand. If AI compute demand shifts — toward more efficient models, toward edge computing, toward different architectures — the utility bonds remain outstanding while the revenue assumptions change [2].

The $570 billion figure is not a prediction that the AI buildout will fail. It is a statement that the buildout is now a credit event with systemic dimensions. When a single sector issues half a trillion in bonds in one year, the refinancing risk does not stay contained to that sector. It propagates through the banking system, the bond market, and the pension funds that hold the bonds [1].

-- THEO KAPLAN, San Francisco