Bond market could define AI's winners and losers

There has been no shortage of bond issuance by hyperscalers this year as the big US tech firms have repeatedly hit the bond market in search of funds for their massive artificial intelligence capital expenditure plans.

At least once a month since February, investors in a variety of currencies have had the chance to buy a huge deal from the likes of Alphabet, Amazon.com, Meta, Oracle and SpaceX.

And although all these issuers have been able to scoop sizeable funding out of the market, with the smallest dollar deal weighing in at $20bn, the difference in aftermarket treatment of these jumbos is stark.

The bond market, could in theory, help define the winners and losers in the hyperscaler arms race.

So far, investors are saying 'we were right already'. There have been no role reversals — the credits that were stronger on paper are doing better.

Alphabet, Amazon and Meta all have double-A ratings. Investors appear to trust those ratings, as well as the companies' massive market capitalisations and strong cash reserves. Alphabet is the highest rated of the three, with S&P placing it on par with the US government itself.

Oracle and SpaceX, meanwhile, are languishing in the triple-B bracket. Earlier this month, S&P cut Oracle to BBB-, leaving it just a notch above junk.

Oracle was the first mover this year among the hyperscalers. In early February it brought a $25bn eight tranche dollar bond, saying this would be part of a $45bn-$50bn funding plan, together with equity and convertible bonds.

Investors adored the deal, pouring in a record-breaking $129bn of orders, so Oracle could price the tranches 5bp to 15bp through fair value.

Five months later, the shine has worn off. Oracle's five year credit default swap spread hit 203bp on Tuesday this week, an all time high, according to S&P Global. Its five year bond was priced at 115bp over Treasuries in February.

Meanwhile, Alphabet, Amazon and Meta's five year CDS were trading between 57bp and 78bp on Tuesday, according to S&P.

Looking at the credit metrics, it's surprising the difference isn't wider — and that the gap has only recently got as large as it is.

The three double-A companies all have more cash than debt — Oracle has four times as much debt as cash, and $98bn of net debt.

Yet just a year ago, before rolling out its massive borrowing plans, Oracle's CDS traded at 38bp — just 4bp wide of Alphabet.

Source: CreditSights

The newest and brashest member of the club is SpaceX, which raised $86bn with a record-splintering IPO at a $1.77tr valuation in early June, then 10 days later asked the bond market for $20bn.

It came away with $25bn from the 144A/Reg S sale, at spreads between 110bp and 175bp over Treasuries, inside where Oracle's bonds were trading.

But the bonds, like SpaceX's shares, have tumbled since. The shares are now at $124, 8% below the IPO price of $135 and 39% below their peak.

SpaceX’s CDS have climbed from 137bp when the bonds were launched to 164bp. The 10, 20 and 30 year bonds have fallen worst, by up to 10%, as the table below shows.

Source: Dealogic, Tradeweb

SpaceX believers will be quick to lay the blame at the volatility and sell-off in long dated US Treasuries since the US reignited its war with Iran in early June.

But if this was the driving force, then Amazon’s long bonds — launched a fortnight ago in a $25bn deal that found a reception distinctly cooler than hyperscaler issues earlier this year — would have gone through a similar rout.

However, the Seattle-based firm’s 30 and 40 year bonds have only fallen by two percentage points, with minimal spread widening, according to Tradeweb.

As with SpaceX's IPO, initial appetite for its bonds was strong, with a $90bn book. But demand has not carried through into the secondary market, where spreads have widened by up to 66bp across the curve, increasing the spread on the longer tranches by almost a third.

Oracle and SpaceX are likely to return to the bond market some time in the next 12 months. Unless their spreads have tightened by then, they will face wince-making financing costs.

Oracle's operating cashflow in its financial year to May was an ample $32bn, up 54%. But because it is investing so much, free cashflow was negative $23.7bn.

SpaceX made an operating loss of $1.9bn in the first quarter. Its maiden results as a public company are not due until August 4.

With metrics like these, the cost of debt matters, and investors' willingness to lend becomes less automatic.

As in past bubbles, there will be winners and losers in the AI arms race.

#For much of the US tech boom of the 1990s and early 2000s, Netscape was the dominant platform for browsing the web, with a 90% market share at its peak. It was subsumed into AOL, in its turn the pre-eminent player in the US. Then AOL itself fell from the firmament.

AOL's downfall was partly due to hubristic, debt-fuelled expansion. In 2001 it bought Time Warner, attempting to become a global digital media player, before its core business shrivelled.

In a curious echo, Paramount, led by Oracle chairman Larry Ellison's son, has just acquired Warner Bros Discovery, supported by Ellison with loans secured on Oracle shares.

Who wins and loses the AI technology race will depend ultimately on which companies succeed in devising business models that actually meet customers' needs and are profitable.

But the bond market will have a meaningful say in the outcome. As things stand, most of the leading players believe they need vast sums of capital to compete. If they cannot retain access to that capital, they may be forced out of the race.

And if one player appears to be stumbling or straggling, bond investors will sniff failure as sharply as equity investors. The weakling could be punished harshly.

This year, among the hyperscalers, a bifurcation has already appeared in their access to credit. The weaker ones — and their investors — must hope it narrows.