The filing describes a complete, self-contained computing system for AI workloads — and it lands less than a year after Tesla killed Dojo, its only in-house AI training computer.
The problem is that this market already has a dominant product, and it isn’t Tesla’s.
Nvidia’s GB200 NVL72 is the reference design for modular AI compute today — a liquid-cooled, rack-scale system packing 72 Blackwell GPUs and 36 Grace CPUs that behaves like a single giant GPU. Nvidia’s DGX SuperPOD stacks those racks into clusters that scale past 9,000 GPUs. Dell builds its PowerEdge XE9712 on the same platform, and Supermicro ships its own GB200 NVL72 SuperCluster.
That’s the competitive set Megapod would enter: established, liquid-cooled, rack-scale systems from the company whose chips power essentially all of it.