The AI bubble refers to the concept that the excitement about AI has led to excess investment. Many areas are contending with new data centers being built. The idea is that huge models, with trillions of parameters, will be used and everyone will need to access a data center for their AI.
Recently I developed a program that works with a 350-million parameter model. The model is only 219 MB on disk; it works well for what I use it for, and I can use it as a tool-calling chat bot that actually accomplishes tasks.
If small models become more popular, could this cause the AI bubble to burst? • No one will need a data center to run 219 MB models; no one even needs a GPU.
• Small models could lead to lower demand for AI subscriptions.
• This could lead to less revenue for frontier model companies and thus pop the bubble.
It is interesting to think about at least. My own interest in owning a bigger GPU has diminished after designing a system that works with a 350M model. After all, if I had a more powerful GPU I would just want to run the same 350M model, as it works fine for my use. Other tasks, like agentic coding, still require more powerful models—so small models will not completely erase the demand for frontier models.