Tesla is ramping up its custom chip ambitions, with CEO Elon Musk confirming that the company’s Dojo supercomputer project—now dubbed Dojo3—is firmly back on track. The revival underscores Tesla’s growing determination to build its own high-performance, low-cost AI silicon and challenge incumbents such as NVIDIA.
Musk said Tesla plans to advance several generations of in-house chips, starting with the upcoming AI5 processor. He claims AI5 can match NVIDIA’s Hopper-class performance in a single-chip configuration, while a dual-die version could rival the latest Blackwell platform—at a significantly lower cost. By vertically integrating chip design, Tesla aims to gain a long-term cost and performance advantage across its AI ecosystem.
Dojo was previously sidelined as Tesla relied heavily on NVIDIA hardware for AI training. However, soaring compute requirements—driven by Full Self-Driving (FSD) development, Optimus humanoid robots, and large-scale AI model training—have prompted the company to recommit to its proprietary silicon roadmap. Dojo3 is expected to be built around AI5-based clusters, enabling a unified compute architecture spanning vehicles, robotics, and data centers.
Musk also outlined an aggressive development timeline, stating that Tesla intends to scale its chips through AI9, with a cadence of roughly nine months between generations. This approach mirrors the rapid innovation cycles seen in the broader AI semiconductor industry.
While Tesla’s renewed push positions it as a potential future force in chipmaking, translating ambition into reality will require consistent execution across design, validation, and manufacturing—an area where even established semiconductor players face formidable challenges.
