The market for AI-driven corporate debt is showing its first signs of strain as a record $285 billion in hyperscaler bond supply tests investor appetite.
The AI-fueled bond issuance boom is hitting a speed bump as credit markets struggle to absorb a record wave of hyperscaler debt, with Amazon's latest offering drawing less than half the demand of its March sale, according to UBS.
"The market is repricing the long-term monetization risk of AI capital expenditure," a team of credit strategists at UBS said in a July 27 report.
Amazon's July investment-grade bond drew a subscription ratio of just 1.6 times, down from 3.4 times in March. Multiple long-dated tech bonds have widened 20 to 50 basis points in secondary trading after issuance. The six largest hyperscalers — Alphabet, Amazon, Microsoft, Meta, Oracle and CoreWeave — are projected to spend a combined $849 billion on capital expenditure in fiscal 2026, with UBS forecasting that figure could exceed $1 trillion in 2027.
The deteriorating demand matters because hyperscalers are increasingly reliant on debt markets to fund their AI buildouts. UBS raised its forecast for hyperscaler bond issuance this year to as much as $285 billion, up from a prior estimate of $240 billion, with bonds maturing in more than 10 years accounting for 32% of issuance — well above the 19% historical average since 2024.
Demand Deterioration Signals Market Shift
The subscription ratio collapse for Amazon — the most creditworthy of the hyperscaler cohort — is the clearest signal that investor appetite is reaching its limit. UBS attributed the cooling to three factors: the shift by enterprise customers toward cheaper AI models, which is compressing revenue growth at frontier labs; growing uncertainty about when massive capital spending will translate into measurable returns; and a 30-basis-point rise in 30-year Treasury yields over the past month, which has amplified price pressure on long-duration bonds.
The credit curve for hyperscaler bonds has steepened sharply. The average spread between 10-year and 30-year hyperscaler bonds now stands at about 42 basis points, nearly double the 22-basis-point average across all investment-grade issuers. While that is below the 80-to-85-basis-point peak during the 2013-to-2015 M&A wave, UBS expects further widening of 10 to 15 basis points over the next six months as capital expenditure continues to overshoot and free cash flow remains negative.
CDS Divergence Reveals Investor Favoritism
Credit default swap markets show investors are discriminating sharply among hyperscalers. Google and Amazon trade 10 to 15 basis points wider than Microsoft in the CDS market, while Meta — which has provided less evidence of capital expenditure monetization — trades about 20 basis points wider than Google and Amazon. Oracle, with leverage exceeding four times earnings, was downgraded to BBB-minus by S&P and is not expected to issue new bonds this year, though off-balance-sheet financing remains possible, UBS said.
The divergence matters for the broader investment-grade market. The five largest investment-grade hyperscalers now account for about 4.5% of the investment-grade index by market capitalization, but their duration-adjusted DV01 weight has reached 6%, up from 3.3% in 2024. If current issuance forecasts hold, that share could rise to 7% or 8% by year-end, making the performance of AI-related bonds increasingly consequential for the entire investment-grade credit market.
UBS warned that the flood of long-dated tech supply could trigger rotation selling into other sectors. In the portion of the investment-grade index maturing in 20 years or more, technology bonds sit alongside consumer staples at about 19%, utilities and communication at about 14% each, energy at about 8% and capital goods at about 5%. Wider spreads on hyperscaler long bonds are likely to first spill into technology and communication, then into utilities and capital goods as investors rebalance, the bank said.
This article is for informational purposes only and does not constitute investment advice.