The Rethought — The Math: $3.5 Trillion That Might Not Exist

Gartner published something this week that got buried under TSMC’s record profits and Anthropic’s $2 trillion valuation chatter.

Their projection: more than $3.5 trillion in AI infrastructure spending could face impairment by 2030 — if AI revenues don’t keep pace with the capital being deployed.

Read that again. $3.5 trillion. Potentially impaired.

The math no one wants to do

TSMC just reported 69% profit growth. Nvidia is printing money. Hyperscalers are expanding data centres faster than they’re building roads. Every earnings call has the word “AI” in it at least a dozen times.

And yet Gartner is quietly saying: most of the specialised AI cloud providers operating today might not exist independently by 2030. The infrastructure being built on top of a demand curve that has not been proven at scale.

This isn’t a doomer take. It’s an accounting one. When you spend capital faster than you earn returns on it, at some point the books have to reconcile.

Why developers should care

Here’s the part that doesn’t get discussed enough: the platforms we build on are downstream of this.

The API you call. The vector database you depend on. The managed inference endpoint your production service hits — these are funded by capital that’s making a bet on future revenue. A bet that Gartner is now saying has a real chance of not paying off.

That’s not a reason to stop building. But it is a reason to think about what happens when your infrastructure layer gets acquired, pivoted, or shut down. It’s happened before — not at this scale, but the pattern is familiar. The .com bubble didn’t kill the internet. It did wipe out a lot of the companies developers had built on top of.

The uncomfortable question

Is AI infrastructure different this time?

Maybe. The demand is real in a way that pet food delivery never was. Enterprises are genuinely adopting these tools. Revenue is growing — Anthropic’s $16.8 billion quarter is not imaginary.

But “revenue is growing” and “the returns justify the capital” are two different statements. TSMC’s customers are already planning for 2029 and 2030. That’s confidence. It’s also a very long runway for a bet that hasn’t fully paid off yet.

What I’m actually thinking

Gartner’s warning isn’t that AI is fake. It’s that the infrastructure layer is being built at a speed that the application layer hasn’t caught up to yet.

We’re in the phase where the pipes are being laid faster than the water is flowing. That’s fine — until it isn’t.

The developers and teams who come out of this well are the ones who aren’t just building on AI, but building things people actually pay for. Abstraction on top of uncertain infrastructure is a risk. Abstraction that delivers measurable value is a different story.

The question isn’t whether to use AI. It’s whether what you’re building with it would survive if three of your AI vendors disappeared next year.

Would yours?