Tesla’s Dojo supercomputer is DOA — now what?

On Thursday, Bloomberg reported that Tesla CEO Elon Musk has ordered its Dojo supercomputer project to be disbanded after the project’s leader and most of its employees left to join another AI project. The news marked a major setback to the automaker’s efforts to develop its own in-house developed supercomputer to train AI models for self-driving cars. And it signaled yet another blow to Tesla’s ongoing efforts to retain its top talent.

For years, Tesla touted its custom built supercomputer meant to train machine-learning models for Tesla’s Autopilot, Full Self-Driving, and Optimus robots as the thing that will give it the leg up over other autonomous vehicle developers. While the rest of the industry was relying on external suppliers for compute and chips, Tesla would design its hardware in-house. According to Musk’s vision, a vertically integrated AI effort would allow Tesla to leapfrog over its more supply-constrained competitors.

But technical delays and a talent exodus seemed to push success further and further away. In 2018, Jim Keller, the famed AI chip developer who was first hired to lead Tesla’s chip making efforts, left. His successor, Ganesh Venkataramanan, left in 2023 to found Density AI. The latest to leave is Peter Bannon, who had been leading Dojo since Venkataramanan’s departure. He is also joining Density AI, along with 20 other ex-Tesla engineers.

Other Tesla executives to head for the exits in recent months include Milan Kovac, head of engineering for Optimus, David Lau, VP of software engineering, and Omead Afshar, one of Musk’s closest lieutenants.

After Bloomberg reported the developments, Musk confirmed Tesla’s change in approach, writing on X that the next-generation AI chips going into the company’s vehicles “will be excellent for inference and at least pretty good for training. All effort is focused on that.”

Depending where you stand on Tesla, the move is either a glaring sign that the company’s AI efforts are overhyped and underwhelming or it’s a savvy move to save money in an extremely capital-intensive race to build robotaxis and humanoid robots.

Either way, there’s no question the decision represents a major pivot from what Musk has been selling as Tesla’s winning formula for years.

Dojo was supposedly Tesla’s “secret sauce” for self-driving cars and humanoid robots, Gordon Johnson, founder and CEO of GLR Research, said in a note to clients. “The system was supposed to hoover up data from Tesla’s fleet and crunch it to supposedly make the algorithm smarter. Analysts even bought the hype with Morgan Stanley in 2023 slapping a $5o0 billion fantasy valuation to boost [Tesla] on Dojo’s potential.”

Moving forward, Tesla is expected to become more reliant on suppliers like Nvidia, AMD, and Samsung. Indeed, the automaker has already been making moves to shore up those relationships, striking a $16.5 billion deal with Samsung last month to source chips for its future EVs and robotaxis. The South Korean electronics giant is making Tesla’s AI6 chips, while TMSC is expected to deliver the automakers AI5 ones in the near future.

In the end, investors could end up rewarding Tesla for winding down its Dojo project and conserving its resources, especially as rising EV competition and the loss of government tax breaks continues to be a drag on the company’s core business. Indeed, Tesla’s stock is up over 2.5 percent in early trading Friday.

The talent exodus remains a bigger problem for Tesla — especially as Musk appears to be competing against himself with xAI. Tesla’s recent struggles seem to touch every corner of the company. Musk’s right-wing political views have created a brand crisis that has sent many of the company’s former liberal supporters running for the exits. If Tesla’s top executives start to follow them, than Musk may be left running this race for AI dominance all by himself.