Who makes containerized "AI factories" for AI data centers

A containerised system arrives as a sealed unit with compute, power distribution and cooling already installed and commissioned in a factory. The pitch is time: the integration work that would otherwise happen on site, in sequence, with several trades, has already been done, so deployment becomes a delivery and a set of connections. The honest reading of the recent "AI factory" branding is that it is mostly the prefabricated data center idea, which has existed for years, applied to dense accelerator racks and sold by the server makers rather than by the modular specialists in the parent part. That is a real shift in who sells it, and it reflects the fact that a dense liquid- cooled rack is hard enough to install correctly that the vendor would rather do it in its own factory. The trade-offs are the ones prefab has always had. Capacity comes in fixed blocks rather than continuously, the unit's cooling and power design is the vendor's rather than the site's, and the site still needs power, heat rejection and network brought to the pad — which is usually the long pole regardless. What to look for is what the unit actually contains, how it interfaces to site utilities, and whether the deployment is genuinely edge or remote — where prefab is clearly right — or a way of accelerating a conventional build.