By Stephen Snowden, head of fixed income at Artemis
A client recently expressed the view that Alphabet and Microsoft are great credits and that spreads on their bonds are too wide. So why don’t we own them? Here are just a few reasons.
The hyperscalers may be great companies but in the world of bonds, the best companies don’t always make the best investments. Investor concerns about spending are well founded. The four largest hyperscalers – Alphabet, Meta, Microsoft and Amazon – have committed nearly $2.4trn in AI-related spending over the coming years. Only part of that expenditure may be debt-financed, but it points to a substantial increase in issuance and a massive overhang of supply.
Even if the bonds of the hyperscalers do outperform the wider AI sector it seems probable to me that the whole complex will underperform the broader market. Owning the best credit in an underperforming sector still means underperforming.
What happens if I’m wrong and spreads on the hyperscalers’ bonds tighten? In my view, that could just encourage them to bring forward future planned issuance.
A recent survey of credit managers in US by Barclays showed that only 23% are significantly underweight in bonds issued by the hyperscalers. I suspect the average US corporate bond manager probably bought in too early and, as a result, is now sitting on losses, with limited capacity to keep on buying. This doesn’t help the technical backdrop.
Many of the hyperscalers are AA-rated today; I believe they might not be in a few years. As capex, lease commitments and debt issuance rise, today’s ratings might be called into question.
I’ve yet to see a definitive report laying bare the full scale of the hyperscalers’ future chip-purchase and lease obligations. But they have collectively signed billions of dollars in lease obligations – another form of debt – and those obligations are not always reflected in their financial statements.
Goldman Sachs calculates that the hyperscalers have taken on $1.5trn of lease commitments, of which about $1trn doesn’t yet appear on the books. US accounting standards allow for future lease obligations not to appear in accounts until they are ‘live’. The hyperscalers are also making use of special purpose vehicles (SPVs), keeping these lease obligations off their balance sheets.
At the same time, vendor financing in the AI space appears to be an increasingly complex and tangled web. Without wanting to be hysterical, there are at least hints of the type of financial engineering and complexity that we witnessed with Enron 25 years ago.
Although we are currently underweight in tech, we are not anti-technology. In fact, we recently added some direct exposure to technology opportunistically, buying investment grade credits from software provider Sage and data centre company Equinix.
Sage (Feb 2031, BBB+)
In early February, spreads on Sage’s bonds widened and its shares fell as Anthropic released a productivity tool targeting in-house lawyers. A few days later, it launched Claude Opus 4.6, an AI model capable of carrying out financial research, extending the threat to financial software firms.
The so-called SaaSpocalypse reflected fears that AI threatened all software companies, including Sage, with obsolescence. Our view then, and now, is that the core of Sage’s business, which is providing HR, payroll & accountancy software to smaller and medium-sized companies, will probably be safe over the medium term. These are areas in which no failure can be tolerated.
None of us knows what impact and influence AI will ultimately have. It may be transformational. And perhaps the pricing power of traditional software providers will collapse over time.
Equally, perhaps AI integration will make existing software packages more attractive and cheaper to improve. That makes calculating the value of Sage’s equity in 10 or 15 years’ time tricky. But we think it’s a reasonable bet that its business-critical software will still be heavily used four-and-a-half years from now, by which time these bonds will have matured.
Equinix (June 2031, A)
Some datacentre bonds in the US have performed badly of late. Many are financing datacentres yet to be built. But the Equinix bond we recently added is different: it’s secured against two existing datacentres in Slough, where constraints on new development have helped keep rents high and rising.
Publicly available information suggests that around 40% of the space is occupied by Amazon Web Services and, with less certainty, a further 40% by Microsoft. The loan-to-value on the assets is 55%, while the bond structure traps all property cash flow if the LTV rises above 75%, providing additional protection for us as bondholders.
The bonds trade at materially wider spreads than unsecured debt from companies such as Digital Realty Trust. Some datacentre bonds are highly speculative – but we believe this one sits at the opposite end of the spectrum.
All of this is written from the perspective of a bond manager. I fully acknowledge the bull case for the AI hyperscalers – AI is growing rapidly and adoption rates are still low and will surge. I accept too that capex is needed to help company’s take advantage of the explosive growth in demand.
But bond managers need to focus on downside risks in the hyperscalers’ credits rather than the upside potential in their equities. And, for now, those risks look too great for my tastes.
