AsianFin – Tesla is expected to adopt Samsung’s cutting-edge 2nm (SF2) process for its next-generation AI6 chip, which is slated for mass production in 2027, according to prominent analyst Ming-Chi Kuo.

However, uncertainty looms over whether the chip can be delivered on schedule due to low current yields.

Samsung’s SF2 node, which leverages GAA (gate-all-around) transistor architecture similar to the SF3 process, currently achieves a yield rate of only 40–50%, Kuo noted. This lags behind TSMC’s N2 node, which boasts a yield rate above 70%, and Intel’s 18A process, estimated at 50–55%.

While GAA technology is expected to help improve yield and support scalability for advanced AI chips, Kuo cautioned that it remains difficult to predict whether the AI6 chip can enter volume production as planned.

Tesla’s move to design custom AI chips comes as it deepens its ambitions in autonomous driving and AI infrastructure, where performance and energy efficiency are critical. The company previously unveiled its Dojo supercomputer, built with in-house D1 chips, to accelerate its full self-driving (FSD) training capabilities.