AI: How deep are the bond market’s pockets?

For much of the past year, the AI story in markets has been one of unrestrained optimism. Firms have been racing to spend on chips, infrastructure, and data centres, and equity valuations have generally rewarded those with the boldest capital expenditure plans.

However, in recent weeks both equity and credit markets have pulled back from what look like frothy valuations. The S&P 500 has dropped around 5% below its all-time high of late October, while US investment grade (IG) corporate bond spreads have risen around 10bp off the 28-year low they hit in early October. 

After a wave of jumbo deals from the likes of Oracle, Meta, Alphabet and Broadcom, there is growing unease about the “capex treadmill” the AI boom has created. Credit investors don’t have the luxury of waiting years for returns on AI investments to materialise the way equity investors do. Many of the growth assumptions feeding into net present value (NPV) models justifying today’s enormous capex are, at best, educated guesses. Equities are still happy to price in the upside, despite certain names taking a breather, but bondholders are increasingly asking when and how those returns will materialise and what the consequences might be for leverage in the meantime. 

Staggering supply puts brakes on AI bonds

For now, the bigger issue in credit isn’t so much growth but rather supply. AI-related issuance estimates are in the region of $1-3tr over the coming years, a volume of debt that may well prove manageable, but timing is a concern; investors are asking why they should buy today’s deal when another mega-deal is likely just around the corner. 

We’ve already had a run of jumbo bond issuance linked to AI spending, and as a result four tech firms now feature in the top 10 US IG bond issuers year-to-date, versus zero in 2024 (see Exhibit 1).

This is in addition to substantial private market financing. Meta alone raised around $30bn privately before turning to public markets for a $30bn multi-tranche deal. That deal, despite coming a day after Meta’s shares fell more than 10% on higher-than-expected capex, still attracted about $125bn of orders, the largest orderbook on record. It was a reminder that demand is still deep, but also that the market is being asked to fund staggering sums in a very short period.

Oracle is the clearest example of how supply pressure can overwhelm fundamentals. Its $18bn deal in September drew roughly $90bn in demand, but the glow faded quickly as the market began to grapple with the looming Vantage Data Centers financing, reportedly around $38bn, tied to Oracle through 15-year data centre leases as part of its $300bn cloud compute agreement with OpenAI. The news of the Vantage plans hit shortly after the September deal and has hung over the market ever since. Oracle’s spreads are now 30-40bp wider than where that deal priced, with the 10-year around 135bp, a significant move for a BBB rated issuer with a capital stack of around $100bn.

Given Oracle and other tech names are set to represent a large and increasing share of corporate bond indices (see Exhibit 2), it’s unclear how much capacity investors have to absorb the heavy pipeline. While most of the widening reflects the expected supply overhang, questions remain about how manageable the near-term rise in leverage will be and how quickly capex can translate into earnings.

A reality check for AI capex

There is, in our view, something healthy about markets beginning to impose discipline on AI-related capex. Syndicate desks will think more carefully before pushing multiple multi-tranche AI deals into a market asking some tough questions. Appetite remains strong, but investors rightly fear this is only the beginning. Capex is heavily front-loaded ahead of earnings, and only a handful of companies (the likes of Alphabet, Meta, Microsoft) have the balance sheet strength to justify spending at this scale in our view. Even then, private credit is likely to play a larger role as corporate cash burn accelerates, and issuers look to avoid repeated blockbuster bond deals.

Some of the early behaviour in markets likely reflected investor FOMO, the fear of missing out. When Oracle came first, many investors saw it as one of the few liquid, IG opportunities to gain direct credit exposure to AI, a theme thus far dominated by equities. Since then, expectations for future supply have ballooned, and credit buyers are more aware of the multi-year issuance cycle taking shape as AI capex accelerates across the sector.

Tech companies are on track to raise around $210bn from public bond markets this year, almost double the prior year with Wells Fargo projecting this figure to rise to $350bn in 2026. This increase looks manageable while credit conditions remain supportive, with persistent inflows and tight spreads. But if markets become less accommodating, this level of spending becomes costlier and harder to justify, and prior NPV calculations may not look so attractive.

AI becoming unavoidable in investment grade

Beyond tech, AI is touching almost every area of the corporate landscape. It is already a central theme in virtually every investor roadshow we attend. For utilities, the story is about enormous data centre power needs and the unprecedented growth opportunity this creates. A traditionally low growth, regulated sector is now talking about double-digit revenue growth off the back of multi-year data centre contracts. That will mean heavy issuance, both in senior unsecured and likely corporate hybrid format, as balance sheets stretch. Telecoms, healthcare, and even industrials are all framing their strategies around AI efficiency, productivity and optimisation. Avoiding the theme altogether will become increasingly difficult for IG investors at the more cautious end of the spectrum. 

While the AI theme continues to dominate the fundamental landscape, the macro consequences remain uncertain. There are clearly contrasting forces at play: AI-driven innovation could boost productivity and growth, ultimately pushing yields higher, while the opposing view highlights disinflationary pressures and potential job losses.

Do bonds and equities see AI differently?

A recent study conducted by two MIT professors for the National Bureau of Economic Research looked at 15 major AI model release dates between January 2023 and December 2024 from OpenAI, Anthropic, Google DeepMind, xAI, and DeepSeek, and examined how US Treasury (UST) yields responded.  They found UST yields typically fall by more than 10bp, on average, in the 15 trading days following a release. The decline begins a few days before the official release as the news begins to leak and persists through at least 15 trading days afterward. 

The authors stress that the study is early, narrow and highly uncertain. They also note that hard macro data – inflation, labour markets, growth surprises – overwhelmingly drive the US yield curve, far more than speculative interpretations of AI developments. With no comparable studies yet and only a small sample of events, it is too early to draw firm conclusions. Still, the contrast between how equities often price AI growth as upside while bonds lean toward caution is noteworthy. More time, more releases and more data will be needed to determine whether AI consistently influences UST yields or whether these early reactions were noise around a handful of high-profile events. But it is nonetheless an intriguing early signal of how the bond market may be interpreting the macro implications of rapid AI progress.

Frenzy meets funding

We’re not here to declare the AI boom a bubble, or to second-guess the boldest growth forecasts – that is a job for equity analysts. As fixed income investors, our strategy is stay disciplined in our AI-related exposure.

While sectors with very uncertain, yet potentially huge, upside are not typically what rocks our boat, given our focus is on collecting coupons and principal at maturity, there are pockets where we see value. In particular, we prefer to play the theme through AI infrastructure, such as data centres and utilities, assets which are essential to the industry regardless of whether Meta’s or OpenAI’s growth forecasts prove correct. Of course, assessing how levered structures are, and what counterparty risk is being assumed, are key components of credit analysis that may override how valuable or important the asset is.

In our view, the pause in AI enthusiasm is less a verdict on the technology and more a reminder of the financing realities beneath it. With multi-year issuance pipelines forming and capex front-loaded ahead of earnings, credit investors are rightfully asking tougher questions.

AI will continue to reshape corporate balance sheets, but its next phase will be driven not just by innovation, but by the market’s capacity, and willingness, to fund it.