AI Costs: Cheap to Use, Pricey to Build

Imagine paying less and less for a service that keeps getting cooler. Well, that's exactly what's happening with artificial intelligence for us users. "Tokens," those tiny units of computation you use to make a chatbot talk, are dirt cheap and getting cheaper.

The catch is, while the final price drops, the cost of building all this magic isn't following suit. In fact, the powerful chips that run AI still cost an arm and a leg. Bloomberg highlighted this AI cost misalignment during summer 2026, leaving investors scratching their heads. So, who's actually making money here?

Right now, the real party animals are the hardware makers. Companies churning out AI chips, like NVIDIA, are raking it in, selling their goods at premium prices. No chips, no AI, simple as that.

Meanwhile, those offering the AI services we all use daily are caught between a rock and a hard place. On one hand, they need to keep prices low to attract us. On the other, their infrastructure costs are insane. The AI chip market, dominated by companies like NVIDIA, has seen strong growth in demand and prices. Not exactly a fair fight, is it?

This gap between production costs and usage prices can't last forever. Either service prices will climb back up, or companies will have to become efficiency wizards to balance their books. We might even see mergers or closures for those who can't keep up.

Analysts predict that pricing pressure on AI services will increase the need for operational efficiency for providers. This could push for more innovative solutions or, more likely, a good old market shakeout. So, the future of AI isn't just about algorithms, but also a whole lot of cash.