Tesla files trademark for 'Megapod,' a modular AI data center system, as it tries to build a compute hardware business after killing its Dojo supercomputer.

What happened: Tesla filed an intent-to-use trademark application for 'Megapod' this month, describing a modular data center system bundling servers, networking equipment, power distribution, and cooling for AI workloads. The company killed its Dojo supercomputer in August 2025, with CEO Elon Musk calling the Dojo 2 design 'an evolutionary dead end.'

Why it matters: Tesla is trying to enter a market already dominated by Nvidia's GB200 NVL72 and similar rack-scale systems from Dell and Supermicro. More fundamentally, Tesla runs roughly 67,000 Nvidia H100-equivalent GPUs in its own training cluster and has no track record selling compute hardware as a merchant business. Tesla's chip schedule is also slipping—AI5 taped out nearly two years behind schedule, and AI6 has slipped about six months, with mass production pushed toward late 2027.

What to watch: Tesla's strength is in power and cooling, not chip design. Its Megapack and Megablock energy storage products are already selling into AI data centers; xAI has bought roughly $1 billion（約1600億円） of Megapacks. The question is whether Megapod leans on that genuine power-management business or attempts to compete directly with Nvidia on servers and chips.