Last week, Meta announced it was planning to sell excess computing capacity to third-party customers, an initiative reportedly dubbed Meta Compute. Read between the lines and this is a concession. This announcement is essentially corporate language for: we're not winning this race.

Meta spent years and tens of billions trying to build a frontier model that could go toe-to-toe with ChatGPT and Claude. Now, the company is repositioning as a toll-taker instead. Turns out, AI slop doesn't need frontier intelligence. It took Mark Zuckerberg some time to figure out that less investment is needed to keep the core algorithm + advertising machine humming, and now he needs a way to monetise the excess inventory.

Spare a thought for the neoclouds (that’s retal favourites like Coreweave, IREN, and Nebius). Over the past few months Meta agreed to pay neoclouds roughly $48 billion for GPU capacity, but it turns out they may soon compete in this space directly.

There was always a defensive logic to those contracts: if you take all the compute, competitors like Snap or TikTok can't get access to build their own AI features. But now that Meta has surplus of its own, the obvious question is what happens when those CoreWeave, Nebius, and IREN contracts come up for renewal in a few years. We wouldn't want to be on the other side of that negotiation…

Step back and the end game comes into focus. For the giants (Amazon, Google, Meta, Oracle) compute is becoming a commodity, and the price of commodities trends toward their marginal cost. Imagine a world where compute costs converge on the price of the electricity needed to serve it. That is, more or less, Google's long-term strategy. In that world, the margin migrates to the semiconductor layer… which is exactly why every hyperscaler is racing to design its own chips and escape the Nvidia toll booth.

This dynamic leaves precious little room for the neoclouds to thrive. Renting out someone else's chips at a markup is a fine business while compute is scarce. It's a brutal one when your biggest customers become your biggest competitors.

FactSet reported earnings last week, and the headline numbers were very good. Revenue growth re-accelerated to 7%, the fifth quarter in a row of improvement. Customer retention is still sky high at 95% or more. And when clients renewed, they signed contracts that were on average 30% longer than the ones they replaced… with no discount given in exchange. Clients locking themselves in for longer, at full price, is about the strongest vote of confidence a subscription business can receive.

But the numbers that matter most in this report are the AI ones, because they settle (at least for now) the biggest question hanging over the entire financial data industry.

Quick primer for anyone unfamiliar: FactSet sells financial data and analytics to investors, banks, and wealth managers. Think of it as one of the big utilities of finance, alongside Bloomberg, S&P Global, and the London Stock Exchange Group. Its customers pay recurring subscriptions, year after year, to access clean, reliable market and company data. The bear case since ChatGPT arrived has been simple: if AI can find, process, and summarise information on its own, why would anyone keep paying tens of thousands of dollars per seat for a data terminal?

Here is what FactSet's quarter actually showed. More than 90% of its top 50 clients now use four or more of its AI products. Subscription growth among clients using those AI tools was 50% higher than among the rest of the client base - that was the killer stat of the whole report. And FactSet's newest offering, a pipe that feeds its data directly into clients' AI agents, already has 450 clients signed up or trialling it, with usage growing 13-fold in a single quarter.

Pause on what this means. AI isn't cutting out the data provider. It's making the data provider more valuable. The reason is intuitive once you see it: the quality of an AI's output depends entirely on the quality of the data going in. Garbage in, garbage out.

An AI agent doing investment research consumes far more data than a human analyst ever could, and it needs that data to be accurate, up to date, and traceable, because no fund manager wants to explain to clients that a trade was based on something an AI hallucinated. The companies that own trusted, clean, verified data are the ones AI has to be plugged into.

FactSet’s report is the first hard, numerical evidence that AI is a demand accelerant for financial data rather than a threat to it. That reads well for the other data businesses in our coverage. S&P Global owns exactly the kind of proprietary datasets (credit ratings history, company fundamentals, commodity benchmarks…) that AI agents need to be grounded in. The London Stock Exchange Group has made the same architectural bet through its Microsoft partnership that FactSet just validated with Google. If FactSet's quarter is any guide, AI is about to make the moats of these businesses finally pay.

Upgrade to unlock full access to the RatedA Portfolio (with weekly updates on every buy/sell and rationale), and the complete database of companies + ratings.