Tesla says AI5 tape‑out complete, aims for US-made chips and 2027 mass production

Tape‑out announced, Musk flags production partners and R&D push

Tesla has completed tape‑out of its next‑generation AI5 custom chip, CEO Elon Musk announced on X, handing the design to fabs to begin manufacturing with mass production slated for 2027. Musk said AI5 will replace the in‑house AI4 chip as the primary compute platform for Tesla’s autonomous‑driving stack and its humanoid robot program. He also revealed that Samsung Electronics and Taiwan Semiconductor Manufacturing Company (TSMC, 台积电) will split production duties — both using plants on U.S. soil — and that work on AI6 and a restarted Dojo‑3 program are continuing.

Performance claims and sources

It has been reported that AI5’s single‑chip performance could rival NVIDIA’s Hopper architecture, and that a dual‑chip configuration would approach the Blackwell class while delivering lower cost and power consumption than comparable NVIDIA parts. Musk has reportedly said AI5 will offer roughly 40× improvements in key metrics over AI4, including about 9× more memory and an 8× boost in compute. Industry outlets have also suggested AI5 is being positioned as an optimal inference engine for models under roughly 250 billion parameters, though those benchmarks have not been independently verified.

Geopolitics, supply chain and strategic stakes

The choice to produce AI5 in Samsung’s Texas facility and TSMC’s Arizona fab is notable. It aligns with U.S. policy incentives to on‑shore advanced semiconductor capacity (the CHIPS Act) and mitigates some risks from recent export controls aimed at limiting high‑end AI silicon flows to China. Can Tesla realistically challenge NVIDIA’s market dominance in datacenter inference? That remains an open question — technical claims are striking, but execution, yield and ecosystem software are decisive.

What to watch next

Key milestones to monitor are fab yields, samples for third‑party validation, and Tesla’s timetable to move from tape‑out to volume production in 2027. Musk briefly triggered confusion on social media by mis‑tagging a TSMC account after the announcement, a reminder that even big technical milestones can be overshadowed by small public missteps. If AI5 delivers on efficiency and cost, it could materially reshape Tesla’s autonomy and robotics economics — but independent verification will be essential.