Musk's Dual Game: 2nm Tesla AI6 Chips and NVIDIA Purchases

On March 19, 2026, Elon Musk revealed the hybrid hardware strategy of his companies. By December, the "tape-out" (completion of design) of the custom Tesla AI6 chip, which will be printed at Samsung facilities using a 2nm process, is expected. Simultaneously, Musk assured the market that Tesla and SpaceX AI will continue to purchase NVIDIA GPUs at scale.

There is no contradiction here; it is the classic division of labor in AI. NVIDIA's architecture (B200 clusters) is critical for *training* colossal multimodal models in data centers. However, proprietary, highly energy-efficient AI6 silicon is needed for *inference* directly at the "edge"—in the onboard computers of electric vehicles for Full Self-Driving and in the heads of humanoid Optimus robots. By controlling the Edge infrastructure, Musk frees his physical products from dependence on third-party vendors.

Source: Reuters / CNA