Can the bond market digest AI hyperscaler supply?

Another week, another hyperscaler deal. Market participants have by now become accustomed to the steady flood of issuance from the large tech players as they look to build out their Artificial intelligence (AI) infrastructure to manage the ever-increasing demands for compute. Amazon are the latest name in the book, raising $25bn of debt across eight tranches to take their tally over the past 12 months to $107bn, including March's $37bn deal, which was the fourth largest corporate bond issued in history. 

Whilst funding of these investments is spread across multiple sources, including (and primarily) operating cash flow, and (more recently) equity issuance, it is an interesting test for the bond markets . At this stage the test is not related to the credit quality of the issuers, even with an unknown return on investment (ROI) on what are very high levels of capital expenditure (CapEx), but the ability of the market to continue absorbing large levels of net supply at spreads that are close to all time tights. 

It's worth noting that issuance volumes so far this year have been easily absorbed. Rolling six-month gross supply of $1.2tn is close to the record set in the period after the Covid reopening, whilst net supply for AA issuance is running at record levels, driven by the types of deals we have just mentioned. This is as US IG spreads of +76 basis points (bp) hover just 4bp above the lows of the past 10 years (reached in January). 

But concession levels and book oversubscription rates are pointing to some indigestion as we move into the (admittedly quieter) summer months. Amazon's $25bn multi-tranche deal this week, with maturities ranging from three to 40-years across eight tranches and priced with a final order book of $41bn, or an oversubscription rate of 1.6x. This is half the rate seen in the company's March bond, which had a final order book of $118.7bn, whilst the concession to the existing curve (the so called new issue premium) equated to 12-22bp across the tranches vs an average this year across IG of 4bp. 

Whilst Amazon have said they are largely done in terms of USD-denominated issuance this year, they will remain opportunistic depending on the market. They and other issuers are increasingly tapping foreign currency markets, with non-dollar bonds accounting for 33% of hyperscaler supply this year, up from 14% in 2025. Indeed, Alphabet even issued the largest ever Swiss franc bond and a 100-year GBP bond, whilst Amazon issued a record sized deal in Canadian dollars. 

For fixed income investors, these deals are impossible to ignore. They demand close attention, given the size of the supply wall this year, but also in the years after, along with the increasingly large weight the sector carries in global credit benchmarks. AI-related issuance (if one includes utilities and industrials related to the build out) is already the largest sector in the US IG index according to J.P. Morgan and will likely continue to expand (they expect $2.1tr of AI financing in the IG market to 2030). 

For active investors, this presents an opportunity, particularly for those who are currency agnostic and can invest across the duration spectrum. The view on the multi-sector desk  from a fixed income point of view is that it is worth having some small exposure to AI, but with significant "dry powder" given the supply coming down the line and given where spreads currently trade. It is also worth mentioning investors have a few different ways of gaining exposure to the AI trade in fixed income. There are hyperscaler, highly rated, senior unsecured bonds such as the Amazon example. But there are also high yield (and IG deals) structured as project finance where the exposure is to a single or group of data centres. Suitability will depend on the mandate given by investors and their and risk return profiles. Needless to say, the way to go is in a highly selective manner, in structures that protect us in the downside scenario via a strong well rated issuer or valuable collateral, and in structures where we can be confident debt service is not at risk if the ROI of the AI investments is not the one equity holders expected. After all, as preferred in any sector, the best case scenario is to receive our coupon and principal payments in a timely manner.