Artificial intelligence capital spending is flooding credit markets with debt, repricing valuations and creating selective opportunities
Watch time: {{ formattedDuration }}
View transcript
Welcome to the latest edition of The Weekly Fix. My name is Andrzej Skiba.
For better or worse, credit markets have largely moved on from following Middle East-related headlines. All of the focus is on the upcoming tech earnings and sector issuance expectations. In this edition, we'd like to share our latest thoughts on the topic.
From an earnings performance standpoint, there's not too much to complain about. Data is showing increasing AI adoption, robust pipeline of business and strong delivery so far. What is however concerning, is the never-ending increase in capex expectations. Part to do with cost inflation, part to do with huge data processing needs, the race to raise capital to fund AI development is only accelerating. We now see a good chance of AI-related capex approaching $1 trillion in 2026 alone, with further increases slated for 2027.
This presents credit markets with an issue. In the absence of large-scale equity raising, all of this needs to be funded by internal cash flows and debt. As capex budgets move upwards, debt has to play an ever-increasing role in this exercise. This is what we're witnessing right now.
Earlier in the year investors sought relief in the knowledge that hyperscalers are starting to tap markets other than US $ to fund their needs. They also welcomed Google's large equity raise. However, since then issuance fears returned as capex budgets keep moving higher, Google was not followed by other issuers with large-scale equity raises and hyperscaler issuance is now happening in addition to large deals from specific data center projects and, pretty soon, GPU (graphics processing unit) financing vehicles.
This recognition led to a sell-off in AI-related debt across both investment grade and high yield. Issuers were forced to offer higher concessions to get their deals done and that, by extension, repriced spreads of existing deals. When you get a flood of deals hitting the tapes, markets need wider spreads to find a new equilibrium. This process might last for a while as there is a large pipeline of deals yet to hit the market.
So, where does it leave our portfolios? Moderately overweight the space as valuations are getting to undoubtedly attractive levels, however with a few important caveats. In investment grade, we're underweight most high quality hyperscalers, as we see the least resistance to their spreads converging towards the likes of Oracle with each successive wave of issuance. We like select data center deals, but only those with either fully amortizing structures where you face no refinancing risk in the future, or those with strong residual value guarantees from high quality tenants. In high yield, our preference is for deals where we strongly expect these to be called in two years' time, once the projects are built and refinanced in either investment grade corporate or ABS (asset backed securities) markets.
We could easily see extended periods of elevated volatility in the AI-related space ahead and for that reason, we want to size our positions conservatively, run a diversified book of holdings and avoid at all costs weaker structures irrespective of their valuation appeal. We are confident that when we look back at the markets few years from now, this time will be seen as a unique opportunity to pick up high quality assets at attractive valuations and to be rewarded for in-depth credit underwriting work, separating the "haves" from the "have nots".
Thank you for your attention.
Key takeaways
AI capital expenditure is approaching $1 trillion in 2026, with more expected in 2027. Without large-scale equity raises to absorb the cost, companies are funding this spending through internal cash flows and debt, putting persistent issuance pressure on credit markets.
A flood of AI-related debt issuance has triggered a sell-off across both investment grade and high yield. Issuers had to offer higher concessions to get deals done, repricing spreads of existing bonds wider as markets search for a new equilibrium.
The team is moderately overweight AI-related credit, but with important caveats. They favor fully amortizing data center structures and high yield deals they expect to be called within two years, while avoiding weaker structures no matter how attractive the valuation.