AI labs test the rigour of credit rating agencies

Is there a world where OpenAI and Anthropic have the same credit rating as Oracle? Bankers for frontier AI labs, soon to go public, are lobbying major credit rating agencies in hopes of securing an investment-grade score. In other words, they are coveting the BBB- rating from S&P at which Oracle, the database group founded by Larry Ellison, sits after it was downgraded this summer.

The two types of businesses have very different profiles. The labs have huge private market valuations — approaching $1tn each — and perhaps soon public market capitalisations past $1tn. That thick “equity cushion” could reassure rating agencies. But while these companies have achieved “run rate” annualised revenue of at least $40bn, they remain lossmaking. Compute, R&D and employee pay costs are enormous.

Oracle, meanwhile, is an established company, with a healthy market capitalisation of nearly $500bn. In the year ending in May 2026, it generated more than $20bn in operating profits, for an operating margin of over 30 per cent, primarily from its core business of selling enterprise application and database software. Its risk stems primarily from its push into building data centres, with heavy investments resulting in $168bn of debt and negative free cash flow. At the same time, Oracle has $260bn of off-balance-sheet data centre lease liability.

On these numbers, at least, the labs would seem to be aiming high. But it isn’t hard to see why they would. Credit ratings, particularly investment-grade ones, matter a great deal. Even at the bottom rungs, high-grade companies are able to sell their debt to a broader set of asset managers including insurance companies and banks who are limited to conservative bets.

While their bonds may slip to trade at wider spreads than the ratings assigned to them would imply, as those issued by SpaceX this summer have done, an investment-grade rating is still a significant imprimatur.

There is a way in which the fortunes of frontier labs and Oracle are intertwined. Just think that half of Oracle’s “remaining performance obligations” — customer pre-payments — of $638bn are owed by OpenAI, on S&P estimates. What makes the ecosystem function will be high demand for AI models which allows the labs to pay their high infrastructure costs to the likes of Oracle, which are in turn investing heavily to meet that ostensible demand from OpenAI and Anthropic.

Still, while equity investors can be more creative in their valuation approaches, debt investors must be more sober. And rating agencies should find a way to have consistent criteria across revolutionary and established companies.

[email protected]