After four years of development, Tesla has pulled the plug on its ambitious Dojo supercomputer project. Bloomberg News reported in August 2025 that the company disbanded the entire Dojo engineering team. The shutdown came just two years after Tesla launched the supercomputer in July 2023.
Tesla abandons Dojo supercomputer after four years, disbanding engineering team despite July 2023 launch.
The Dojo project faced challenges from the start. In September 2022, a cabinet test drew 2.3 megawatts of power and tripped a substation. Tesla’s projections for the system didn’t pan out. The company had aimed for 100 exaflops of computing power by October 2024 and a top-five global ranking by February 2024. Neither goal was met.
CEO Elon Musk’s confidence in Dojo appeared to waver over time. In August 2022, he promised to “phase in Dojo” to reduce GPU purchases. But by January 2024, Musk called Dojo a “long shot” with a low chance of success. He acknowledged the trade-off between potential and probability. Tesla’s 2025 earnings call notably made no mention of Dojo, signaling the project’s declining importance.
While Dojo struggled, Tesla expanded its use of NVIDIA hardware. The company operated 10,000 NVIDIA H100 GPUs alongside Dojo in August 2023. By April 2024, that number grew to 35,000 units. Tesla’s new Gigafactory Texas cluster, started in June 2024, was designed to use both Tesla and NVIDIA hardware equally. The facility, known as Cortex, represents Tesla’s shift toward proven GPU technology for AI training.
The financial stakes were high. Tesla committed over $1 billion to Dojo through 2024. By April 2024, the company projected $10 billion in total spending on self-driving training systems. The investment covered custom chip design, manufacturing, and production equipment.
Tesla’s decision to abandon Dojo signals a strategic shift. The company is moving away from exclusive custom training systems. Instead, it’s adopting a hybrid approach that relies heavily on proven NVIDIA technology. The new Texas facility plans for 130 megawatts of initial power capacity, eventually expanding to over 500 megawatts.
The Dojo shutdown shows Tesla’s willingness to cut losses on projects that don’t deliver. While the company continues to develop some custom hardware, NVIDIA GPUs have become its primary training infrastructure. This pivot allows Tesla to focus resources on data pipelines and software development for its self-driving technology.
