Tesla Taps Samsung and Intel for Dojo 3 Chip Production

According to Korean media outlet ZDNet Korea, Tesla is making a major shift in the development of its AI supercomputer, Dojo, which powers its Full Self-Driving (FSD) systems. While previous generations of Dojo chips were manufactured solely by TSMC, Tesla is reportedly building a new dual-track production strategy for the third-generation Dojo—partnering with both Samsung and Intel.

This move marks an industry-first collaboration between two chipmaking giants under the direction of a single end-user. Tesla is in active talks with Samsung's foundry division for front-end chip production and with Intel for advanced packaging using its EMIB (Embedded Multi-die Interconnect Bridge) technology.

At the heart of the Dojo system is Tesla's self-developed AI chip architecture. Earlier versions used 7nm D1 chips from TSMC, but the upcoming Dojo 3 will feature a new "D3" chip integrated with an "AI6" chip, intended to support next-gen FSD, robotics, and data center workloads. Tesla CEO Elon Musk previously noted that D3 and AI6 chips will be "essentially the same," with scalability depending on use—ranging from two chips for EVs or humanoid robots, to 512 chips for servers. The AI6 chip is expected to be based on a 2nm process.

Behind the shift lies a mix of technical and strategic reasoning. Tesla's unique chip design relies on extra-large packages, previously made possible through TSMC's System-on-Wafer (SoW) technology. But SoW is better suited for low-volume, high-performance chips—raising questions about its scalability. In contrast, Samsung and Intel are said to be offering more flexible and scalable options to support Tesla's roadmap.

Samsung already signed a major deal—worth around $16.5 billion—to mass-produce AI6 chips at its upcoming fab in Taylor, Texas. Meanwhile, Intel’s EMIB technology enables efficient 2.5D packaging without the need for large interposers, potentially offering better flexibility for chip layout and size scaling.

That said, EMIB may still face challenges in supporting ultra-large designs like Dojo 3. The industry speculates Intel may need to upgrade its EMIB platform or invest in new equipment to meet Tesla's needs. Samsung is also accelerating its efforts in advanced packaging for large AI chips, which could lead to its deeper involvement in future phases.

This shift in Tesla's chip manufacturing approach highlights the growing demand for customized high-performance computing hardware in the AI era. It also signals a new model of collaboration across the semiconductor industry—setting the stage for deeper integration between chip foundries, packaging specialists, and end-application innovators.