There’s a company down 17.6% this year because investors are afraid AI is going to disrupt it.
That same company spent the last two years building the product AI agents cannot legally operate without.
I want to be careful how I say that, because it sounds a little too neat. But an AI can write you a better analysis than most humans on the street and it still cannot write itself into a pension fund’s investment mandate, and that’s what makes this a unique opportunity.
What they own is the specific data that regulation, contracts and investment mandates already point at by name, which is a good deal harder to compete with than simply having better information.
Below is how that works, and why the machines make it bigger instead of smaller.
The company is S&P Global, ticker SPGI, and most people know it for exactly one thing without knowing what the business underneath is.
You know the S&P 500. That’s an index they own, and when trillions of dollars track it, the funds running those products pay licensing fees, because the index is S&P’s intellectual property and there isn’t a way around that.
The ratings business works the same way, only more so.
When a company issues a bond it pays S&P for a rating, and it does that because most institutional money is either prohibited by mandate or effectively barred by regulation from buying unrated debt.
The pension fund never chose S&P. It followed a rule that says the paper needs a rating from a recognized agency, and there are only two of those that seriously matter.
Same story in commodities, where a great many oil, gas and metals contracts settle against S&P’s Platts benchmarks, which are written into the contracts themselves.
What S&P has isn’t about holding better data than the next guy, it’s about being the data that regulation and contract language already point at, and that’s a different animal entirely.
They May Be About To Sell The Part You’re Worried About
Bloomberg reported last week that S&P is weighing a spinout of Capital IQ Pro, which is their data and research platform, and the stock rose 4% on the news.
Now think about which piece of this company that is. Capital IQ Pro competes head to head against FactSet, the London Stock Exchange’s data arm and Bloomberg. It’s a terminal business, and a terminal business is precisely where somebody with better software could come along and take your customers.
It’s also the piece the market has been marking the whole company down over.
Take it out and what remains is ratings that regulation requires, indices written into mandates, and benchmarks written into contracts. The optional business leaves and the mandatory business stays.
Bloomberg said this is early innings, S&P declined to comment, and they may well decide against doing anything at all. But they already spun off the Mobility segment in July, so the direction of travel is at least consistent.
The Numbers That Come Out Of A Business Like That
And a dividend that’s been raised for 53 consecutive years, on a payout ratio around 20%, which tells you the dividend isn’t stretching anything.
You can find plenty of good businesses. It’s harder to find one where the customer is required to show up.
Now What’s Changing
Financial work is starting to migrate from people to software, and that changes what the data has to look like.
A human analyst can hold an approximate number in their head and be fine, and an agent can’t do that. It needs the number, the source, and a citation trail that survives an audit, and that requirement comes from the regulator sitting behind whoever deployed the thing.
S&P has spent the last couple of years building precisely that. They launched something called Adaptive Retrieval in July through their AI Data Portal, which lets a customer’s AI agents query licensed S&P data directly, in natural language, with citations attached. Their own language on it is that agents and language models need data that is properly cited, verifiable and auditable.
They’ve wired it into Google Cloud, with the data unified on BigQuery and their retrieval agent running inside Gemini Enterprise. They’ve wired it into Microsoft 365 Copilot as of August. AWS, Snowflake and Databricks as well.
So as the volume of financial work done by software goes up, the number of lookups against authoritative data goes up with it, and S&P sits on one end of that pipe.
The Second Leg, And I’d Say This Is The Cynical One
The AI buildout itself is being financed with an enormous amount of new corporate debt.
Every tranche of it needs a rating before pensions and insurers can buy it, because those buyers are the same ones who can’t touch unrated paper. S&P gets a fee sized to the issuance, with no balance sheet exposure to whether any of those GPUs hold their value.
And if the AI credit cycle turns out badly, downgrades are billable too. Surveillance fees continue, workout activity generates ratings actions, and they get paid on the round trip.
I’m not saying that’s admirable. I’m saying it’s how the business is constructed, and it’s the reason this is the one name in the whole AI capital structure that collects on the way up without taking the hit on the way down.
What You’re Paying And What Goes Wrong
The stock isn’t cheap on any screen you’d run. You’re paying a premium multiple for a premium business, and I’d sooner say that plainly than dance around it.
Near-term earnings are cyclical with debt issuance, and issuance slows when rates rise, which they may well do this month. So the ratings line can get soft at exactly the moment the macro gets loud.
And this is a slow compounder. Low teens annually from earnings growth, buybacks and the dividend is the realistic expectation, not a double.
The interesting part is that the stock has been down close to 18% on the year, and the stated reason has been fear that AI disrupts the data business. Which is a reasonable fear in the abstract and a strange one to apply to the company selling the agents their citations.
Everybody in the AI fight issues debt, licenses indices and needs verified data, and this is the way to collect from all of them without having to pick which chipmaker or model or cloud comes out on top.
There’s a number of ways to play this one, long LEAPS, buying the outright common, or even selling puts.
If I take a trade, it will be in Special Situations.
Andrew Giovinazzi