What's this about?
An unusually shaped acquisition. From what's understood, Nvidia bought the "model factory" part of poolside — the model-training operation — and many of its employees, reportedly researchers, received Nvidia offers. But the founders are staying at poolside. The open question raised is that with no mention of poolside's infrastructure company, whatever remains might simply become a neocloud compute provider.
What exactly is a "model factory," and why buy only that?
It's the machinery and expertise for training models repeatedly and reliably — data pipelines, training infrastructure, the evaluation loop, and crucially the people who know how to make a run succeed. That capability is scarce and hard to build, and it's separable from the product a company was selling. Buying it and leaving the founders is close to an acquihire of a capability rather than a company, and it suggests Nvidia wanted the training expertise rather than poolside's coding product.
Why is this worth noticing?
- Founders staying while the researchers leave is backwards from the usual pattern, and the most informative detail in the report.
- It fits Nvidia's broader positioning. A company that sells the hardware acquiring the expertise to train on it moves it further up the stack — the same week it reduced its OpenAI infrastructure backstop.
- The neocloud question is the real one for poolside. Without the training operation, a company holding infrastructure becomes a compute provider, which is a very different business from the one it raised money to build.