Global AI-related bond issuance totaled only about $23 billion in September, marking the second-lowest month of the year. AI issuance in the US investment-grade market dropped to zero.

Morgan Stanley characterized this slowdown as a "pause rather than a retreat," expecting issuance to pick up in the fourth quarter but not to replicate the explosive growth seen in the first half.

As of the end of September, total global AI-related bond issuance for the year had reached $466 billion, more than double last year's full-year total of $216 billion.

In its latest report, Morgan Stanley attributed September's cooling not to deteriorating fundamentals or capital shortages, but to a combination of three factors: a significant front-loading of earlier issuance, regulatory and political headwinds facing data center construction, and sharply rising interest rates forcing project-level terms to be renegotiated.

Why September cooled down

The pace of AI financing this year has been highly uneven. June set the annual peak with $113 billion in single-month issuance, followed by a monthly decline thereafter.

High-quality hyperscalers including Google, Amazon, Meta, and Microsoft issued approximately $132 billion in the investment-grade market this year, a roughly 25-fold year-over-year increase. Adding Oracle and SpaceX, the six major hyperscalers' total issuance across all currencies has reached approximately $254 billion, heavily concentrated in the first half. Morgan Stanley noted that "after a busy summer, given the volume already completed and issuers' efforts to establish a cadence, we expected near-term supply to slow."

US investment-grade AI issuance was zero in September. Non-dollar markets saw slight activity, with Amazon issuing a 4.25 billion pound four-tranche deal. In leveraged finance, SoftBank Group issued approximately $10 billion in high-yield bonds, and AI data center operator Crusoe issued about $500 million in loans, bringing September's total leveraged finance volume to approximately $14 billion.

The second factor is regulatory and political constraints. New data center construction faces permitting, power supply, and political headwinds—which Morgan Stanley previously summarized as the "three Ps": people, power, and politics. These constraints are transitioning from potential risks to real obstacles.

The third factor is interest rates. The sharp rise in yields means project-level transactions may need to be renegotiated, with lower-rated issuers and project financings being more sensitive to funding costs.

Q4 outlook: Hyperscalers to return, but at a different pace than last year

Morgan Stanley expects fourth-quarter issuance to exceed September levels but not to be as concentrated as last year, when most supply appeared in Q4.

High-quality hyperscalers are expected to return to the US investment-grade market while continuing to raise funds in non-dollar markets. Non-dollar hyperscaler issuance has reached approximately $72 billion this year, nearly one-third of total global hyperscaler issuance, with currencies expanding from just dollars and euros last year to Canadian dollars, British pounds, Swiss francs, Australian dollars, and Japanese yen. Morgan Stanley expects euro-denominated deals to also enter the market in Q4.

The pace of data center project financing is harder to predict. Morgan Stanley lowered its ABS and CMBS issuance estimate for the year to $25-30 billion, implying roughly $5-10 billion of remaining supply space in Q4.

An important driver of the Q4 rebound is the 2027 capital expenditure outlook. Morgan Stanley forecasts combined capex for the six major hyperscalers in 2027 at approximately $1.4 trillion, significantly above the market consensus of about $1-1.1 trillion. Third-quarter earnings reports, expected to roll out in October, could bring another round of capex upgrades, directly driving financing demand.

Morgan Stanley believes high-quality hyperscalers are "largely insensitive" to higher rates—with strong return on invested capital (ROIC) prospects and debt financing costs still below equity financing costs. These companies have overall leverage of only 1.3 times (net leverage 0.4 times), a cash-to-debt ratio of 132%, a median credit rating of AA-, and ample balance sheet capacity.

Macro test looms larger than supply shock

Morgan Stanley emphasized that for the investment-grade credit market, "the macro environment—not excess supply—is the biggest test before year-end."

Investment-grade bonds have posted a year-to-date total return of negative 3%, with quarterly returns down 4%, a drawdown that has reached levels that could trigger large-scale redemptions from mutual funds and ETFs. The 10-year US Treasury yield has risen from 1% to above 5%, and multiple rate repricings over the past five years have repeatedly impacted credit spreads and fund flows.

However, Morgan Stanley believes the credit market as a whole can absorb rising yields, as nominal growth remains above 6%, corporate earnings growth is even stronger, and the rate rise is supported by solid fundamentals. The firm's economists expect the Federal Reserve to hike only twice more, more dovish than market pricing of nearly four.

On credit differentiation, Morgan Stanley reiterated its preference for secured assets, which account for about 30% of AI-related debt. Spread volatility in secured data center bonds is primarily driven by construction risks—permitting delays, power supply, and lease uncertainty—while already-operating ABS and CMBS assets are less affected by such shocks. Chip financing's asset characteristics allow for faster cash flow generation and amortization, but publicly traded products remain limited.

Looking beyond the fourth quarter, Morgan Stanley believes most of the hyperscaler spread compression trade may already be done: issuance cadence is becoming more predictable, capital expenditure is shifting toward shorter-duration assets like chips, and strong returns on AI investment continue to validate the soundness of the spending rationale.