Tesla has dismantled its Dojo supercomputer team and ordered the programme shut, according to Bloomberg and Reuters reports citing people familiar with the matter. Vice President of hardware design Peter Bannon, who led Dojo, is leaving the company after nine years, and remaining engineers are being reassigned to other compute and data-centre projects. Dojo was developed around Tesla-designed D-series chips to train the vast volumes of driving data that underpin the carmaker’s Autopilot, Full Self-Driving software and Optimus robots. The initiative had been touted by Chief Executive Officer Elon Musk as a cornerstone of the company’s artificial-intelligence ambitions. With the in-house effort shelved, Tesla plans to rely more heavily on external suppliers. The company will source compute from Nvidia and Advanced Micro Devices and tap Samsung Electronics for chip manufacturing. Samsung last month secured a US$16.5 billion deal to make Tesla’s forthcoming AI6 inference chips. Roughly 20 Dojo engineers have already left to form DensityAI, a data-centre hardware startup. Musk wrote on X that “there is no sense for Tesla to scale two different AI chip designs,” adding that future work will focus on the company’s AI5 and AI6 chips, which he said will handle both inference and, to a lesser extent, training workloads.