Separating AI Hype from Reality

2025 seemed to be a big year for AI hype, but it’s nothing compared to what we’re seeing this year. In April, we had everyone pushing to use as many tokens as possible (tokens are a way of measuring AI consumption), then, like clockwork, the bills came in during May/June, and suddenly all the talk was about how expensive tokens are. Then in July the talk is about open-weight models - as if on cue, a new Chinese model came out that appears to be almost as good as the best US models. 

A few points to manage the hype:
We are in another big technology change, just like the internet or the personal computer. The difference is that it is happening faster, so the hype gets compressed, and social media makes it ever-present. One way to think about all the hype is as the market figuring out rapid change, in real time. It’s healthy even if it can be annoying. 

The token cost issue has people assuming that the AI model companies are just subsidizing AI so we get addicted to it. This might sound like a satisfying explanation; history has shown that infrastructure builds like the data center build-out always overshoot, the market ends up with a glut, and prices crash. The lock-in almost always happens closer to the user, and anyway there are too many competitors in the AI model space to lock anyone in. Of much greater concern should be Microsoft, which is building products that increasingly take on AI and software functions. 

Finally, we are almost certainly at peak hype, and will be here for a little while longer, but in a year or so, it’ll settle down as all the other technology waves do; AI will be a tool you are used to, and we’ll be on to the next thing to gossip about.