JPMorgan’s Andreas Michalitsianos outlines four ways he is investing in AI-related debt issuance.
Debt issuance is accelerating as the AI build-out accelerates and expands – and for Andreas Michalitsianos, manager of JPM Global Corporate Bond, that creates more opportunity for bond investors than most realise.
In 2025, the largest hyperscalers – including Amazon, Alphabet, Meta and Oracle – issued around $120bn in US corporate bonds versus an average of $28bn per year between 2020 and 2024, with Morgan Stanley forecasting that total AI-related global debt issuance will reach $570bn in 2026.
Recent debt issuance from the hyperscalers includes $14bn of Canadian dollar-denominated high-grade bonds, €14.5bn in euro-denominated bonds by Amazon and a 100-year bond in sterling markets by Alphabet.
Last month, PGIM’s co-chief investment officer of fixed income Greg Peters argued that bond investors in the unsecured space are “knowingly financing a bunch of losers”, given that the winner-takes-all nature of AI means the losers will overwhelm the one or two winners and, unlike equity investors, while bondholders may get their money back, they will never get the upside.
Michalitsianos also acknowledged the risk that underpins the AI bubble versus boom debate: What if the promises surrounding the AI build-out fall through?
“The obvious parallel is the dot-com bubble, when there were only a couple of winners and many more losers,” he said.
He emphasised the importance of being self-reflective but countered that the risk to bond markets is ultimately far less than in equity markets.
“If I use Meta and its Metaverse as an example, the company admitted this was not the right call and pulled back its capital expenditure (capex) on the project – I think this is what would happen with the hyperscalers, too,” he said.
“If a company decided that owning, building and maintaining a frontier model was not feasible after all, it would likely licence it from someone else and simply pull back on capex.”
While the consequences of this would likely send ripples across equity markets, Michalitsianos pointed out that this does not mean the company will be unable to pay back its debt.
“That decision actually makes it more creditworthy because it would be spending much less,” he said.
Given the growing pace of demand for AI and AI-powered products, it is also likely another hyperscaler would step in and commandeer the space should another pull back its financing plans, Michalitsianos added.
“In that scenario, where demand for AI is not on the trajectory people expected – and that is why the investee company is pulling back financing – from a bondholder’s perspective, we are still comfortable with the balance sheet, whereas an equity holder would be experiencing a lot of volatility.”
The four stripes of AI debt
Michalitsianos has been exploring the ways in which JPM Global Corporate Bond may invest in AI, noting there are “four main stripes” to AI-related financing.
The first is the debt of the hyperscalers themselves. “At times, they do provide opportunities because, as supply comes, they may present a new issue concession and they can be interesting tactically,” Michalitsianos said.
However, he was cautious about being too overweight at this time “because they are going to issue a lot of debt over the next several years but they remain interesting”.
The second way in which Michalitsianos expects to see AI debt manifest more is through investment grade construction bonds.
“The hyperscalers are funding a lot of chips and servers but they don’t necessarily want to pay for all the shells housing the data centres – so they are happy to let others do the heavy lifting there,” he said.
He explained that these bonds are typically issued from joint ventures – “typically non-recourse to a corporate parent and then secured by a data centre”.
“This would normally sit on a bank balance sheet, as it is construction lending, but because of the size, and because some issuers wish to secure funding in bond markets, we are seeing this become a new segment of our market,” he said, noting that he likes these bonds “when they tick all the boxes”.
A smaller sliver of the growing AI-related debt issuance comes from the utilities required to power all this new AI infrastructure.
“Of everything it takes to build a gigawatt of data centre capacity, only about 5% is the power build-out – yet this still needs to be funded,” Michalitsianos said.
In this area, he said he especially likes issuances from utilities with projects in southeastern states in the US, due to less political pushback to AI-related construction work.
The final area concerns bank lending.
“Bank lending in the US is up by around 11% year-over-year and a good portion of that is commercial real estate, which is now starting to inflect higher because of these data centres,” Michalitsianos said.
“So lending to banks – which primarily use deposits for funding but also borrow from bond markets – can present opportunities as well.”
However, AI is not the only secular theme prompting opportunities across debt markets, with Michalitsianos highlighting the “capex wave” surrounding the pharmaceutical revolution with GLP-1s, the global renewed desire for energy security, reshoring and nearshoring, and aerospace and defence spending.
“It points to a period where spending on capital equipment and factories will be very high, which is growth-supportive, and it will all require diverse funding,” he said.
“The AI story is like a cloud in the sense that you cannot see past it but there is an enormous amount happening behind it too.”