The Hard Questions for AI Bonds Investors

SpaceX has given equity and bond investors two rollercoaster weeks. After at one point exceeding US$3 trillion at its intraday peak before a multi-day tumble, it then issued US$25 billion of investment-grade (IG) bonds.

While it appeared that investors happily piled into the deal initially, three days later they wanted to be paid as if they were lending to a non-IG rated company, with the 30-year paper yielding 2.01 percentage points above equivalent US government debt against 1.67 percentage points for the average speculative-grade borrower.

The disconnect between an almost US$3 trillion market cap and bonds trading closer to high-yield territory could invite at least three readings: is it a market comment on governance and management concentration, a genuine reassessment of default probability beneath the headline valuation, or simply the reality of a near-term gold rush in which equity investors chase the price jump while bond investors price the 30-year reality?

We’ve pointed out on more than one occasion that SpaceX might not be your run-of-the-mill AI play, but investors are looking to it for clues to gauge how other upcoming IPOs and bond offerings may fare. There is a lot of fear around being underexposed to the “up-crash” potential in the sector and the category.

The hard questions

Those answers are unlikely to come from SpaceX. For us, what it does instead is force some harder questions about how credit markets should price the AI wave behind it. Such as: when is IG not actually IG and just an enormous high-yield issue? Is a historic record level market cap actually relevant to creditworthiness?

Under the Merton model, the quant framework that underpins structural credit analysis, market capitalisation serves as a proxy for firm value, and thereby default probability. Under the model, a valuation that is large relative to a company's liabilities drives the probability of default to close to zero. Merton et al would say there is enough market capitalisation that the credit spread should be basically zero. Until it wasn’t.

Two companies squeezed into one bond deal

Looking at what the rating agencies revealed may help explain the disconnect. SpaceX reported a net loss of US$4.9 billion on revenues of US$18.7 billion last year. That is the consolidated picture. Underneath it, the satellite connectivity business Starlink generated US$7.17 billion in segment adjusted EBITDA (earnings before interest, taxes, depreciation and amortisation) on US$11.4 billion in revenue in 2025, a 63% margin.

Moody's said it applied a telecom services methodology to underpin its investment-grade rating, anchoring on Starlink's cash flows. Moody’s also highlighted the uncertainty around the AI segment's scalability, timing and economic durability.

In other words, credit investors anchored on Starlink while equity investors priced xAI, the AI business that recorded a US$6.35 billion operating loss in 2025 on US$3.2 billion in revenue, consuming US$12.7 billion in capital expenditure.

That distinction raises a key point for the AI bond market more broadly. SpaceX had a profitable satellite business that appears to have supported the case for an investment-grade rating. Many of the companies expected to tap bond markets over the coming months do not have an equivalent.

What about the AI pure plays?

Data centre operators, neo-cloud companies and leveraged software businesses coming to market are, in many cases, pure plays on the AI thesis, without a cash-generating business underpinning the credit case. For those issuers, any Merton cushion may rest entirely on speculative equity value.

We have been cautious about this dynamic. Technology's share of the investment-grade credit market has already grown from less than 1% at the start of 2015 to 4% today. In our H2 Credit Outlook last week, we noted that the sheer scale of AI bond supply risks pushing spreads wider across the credit complex through what we called a substitution effect: as investors absorb ever-larger volumes of AI paper, compensation for lower-quality risk has to rise to clear the market.

Morgan Stanley estimates AI and hyperscaler issuance could reach US$400 billion in investment grade and a further US$65 billion in high yield and loans this year, potentially accounting for nearly 20% of total issuance. For context, internet and tech issuance at the peak of the dot-com era averaged 14.5% of total issuance across 2000 and 2001.

Figure 1. AI driven issuance is surging

Source: Dealogic, Pitchbook | LCD, Morgan Stanley Research forecasts, as of May 2026.

Problems loading this infographic? - Please click here

There is some serious asymmetry in that trade. Those buying equity in the AI buildout stand to benefit if the boom plays out as bulls expect, but it needs to happen soon. Those buying the bonds carry full exposure to execution risk, construction delays, cost overruns and competitive disruption, for a fixed coupon in return.

SpaceX's order book made the market's unease visible during the primary sale, which showed demand was strongest at five years and fell sharply at 20 and 30. In secondary trading, the longest-dated bonds widened the most. Investors appear willing to lend over the near term. Whether the AI thesis holds over 30 years is much harder to call.

One would imagine that tensions will become more pronounced as the pipeline grows. And investors need to ask the hard questions about risk, timing, consolidation and failure. Is this about competing for the launch (equity) or managing the landing (debt)?

All data Bloomberg, unless otherwise stated.

Author: Matt Rowe, Managing Director, Solutions at Man Group.