AI as utility: settled product, unsettled economics (OpenAI losses + SpaceX compute landlord)

We treat AI like we treat electricity: something that simply works, will keep working, and will hopefully cost roughly what it does today.

This cannot continue indefinitely.

Consider the commercial layer first. According to leaked financials, OpenAI, the firm behind ChatGPT, took in around thirteen billion dollars last year and spent close to thirty-four. The reported operating loss, roughly twenty-one billion dollars in a single year, is not a slow start or a rounding error.

It means the service is still being sold on economics that are nowhere near settled. The gap is being carried by capital markets, infrastructure partners and investors, not by a clean, proven margin model.

Now consider the ground it runs on. SpaceX has become a major AI infrastructure landlord almost overnight, with reported compute-access deals with Anthropic and Google worth roughly twenty-six billion dollars a year combined.

Several of those agreements carry ninety-day termination clauses, exercisable from the end of 2026. Useful capacity, but not the same thing as owning the ground beneath your feet.

Put the two together and the picture is clear. What we experience as a dependable utility is, underneath, loss-making, investor-subsidised, partly rented and commercially unproven.

None of that makes the tools less useful. It is not an argument against adoption.

It is an argument against assuming the terms are fixed. Price, availability, data retention and even ownership of the infrastructure behind the product are more conditional than the interface suggests.

The sensible response is not to retreat. It is to treat AI as a capability you adopt, not a fixture you can blindly lean on. Build with value in mind, but also with portability, fallback options and an exit route.

Avoid, if at all possible, being locked into a single vendor.

A product can feel like infrastructure long before its economics have earned the name.