Tesla Kills Dojo Supercomputer, Big on Next-Gen AI Chips

Tesla Ends Dojo Supercomputer Project, Shifts to AI5 & AI6 Chips

Tesla has officially pulled the plug on its Dojo supercomputer initiative, a high-profile project once billed by CEO Elon Musk as the key to unlocking fully autonomous driving. Instead, the electric vehicle giant will now concentrate its efforts on developing next-generation AI chips designed for both training and inference tasks.

The decision signals a major strategy shift for Tesla’s AI roadmap — and comes alongside a wave of high-profile departures from the Dojo team.

---

Key Leaders Exit for New AI Venture

Dojo program head Peter Bannon is leaving Tesla, along with roughly 20 engineers and executives, to join DensityAI, a startup founded in 2024 by former Tesla chip chief Ganesh Venkataramanan.

DensityAI aims to develop cutting-edge AI hardware and software for data centers, robotics, and automotive systems, directly overlapping with some of Tesla’s original Dojo goals.

---

From Supercomputer Ambitions to Chip Powerhouses

While Dojo was envisioned as Tesla’s in-house training powerhouse for its Full Self-Driving (FSD) neural networks, Musk confirmed the company will now double down on its AI5 and AI6 chip lines.

In a post on X, Musk said the chips will be “excellent for inference and at least pretty good for training,” hinting that Tesla sees more flexibility and cost-efficiency in scaling chip production than maintaining a custom supercomputer platform.

• AI5 chips are slated for mass production in late 2026.
• AI6 chips could arrive in 2027 or 2028.

These chips will power not only FSD-enabled Tesla vehicles but also the company’s Optimus humanoid robots, and potentially wider AI applications in edge computing and automation.

---

The Rise of Cortex: Tesla’s New AI Supercomputer

With Dojo canceled, Tesla’s large-scale AI computing needs will be met by Cortex, a massive supercomputer currently being built at its Austin, Texas data center.

Once completed, Cortex is expected to run on over 100,000 Nvidia H100 and H200 GPUs, putting it among the most powerful AI clusters in the world — and ensuring Tesla remains competitive in AI model training.

---

Why the Sudden Change?

While Musk had championed Dojo since 2021, signs of trouble surfaced in recent earnings calls, where he spoke of potentially merging chip architectures to avoid spreading engineering resources too thin.

Analysts say the pivot reflects a pragmatic move to invest in scalable, multipurpose AI hardware that can serve both Tesla’s automotive ambitions and emerging AI markets — rather than being locked into a single-use training system.

---

Bottom Line
Tesla’s Dojo dream may be over, but the company’s AI ambitions are not slowing down. By betting on advanced AI5 and AI6 chips and building the Cortex supercomputer, Musk is positioning Tesla to compete not just in autonomous driving — but in the global AI infrastructure race.