We tried training a small model on a laptop. It mostly worked.
You don't need a data center to get started — you need patience and a smaller ambition.
The pitch for local, laptop-scale model training has always been “you can do this at home.” The honest version is: you can, if your expectations are sized to match a laptop GPU rather than a server rack.
The bottleneck was never the math. It was waiting for checkpoints to save to a drive that wasn't built for this.
Still, for small, task-specific models — a classifier, a style-matched text generator — a modern laptop is genuinely enough. The ecosystem of tools that assume a single consumer GPU has grown a lot this year.