Tesla accelerates next-generation AI chip strategy with Samsung and TSMC, while weighing its own ‘Tera Fab’

Tesla is moving to strengthen its supply chain for next-generation artificial intelligence (AI) chips, a key component in the company’s autonomous driving systems and humanoid robotics development. CEO Elon Musk said the company will produce its newly designed AI5 chips through Samsung Electronics in Korea and multiple Taiwan Semiconductor Manufacturing Company (TSMC) facilities, while also considering the construction of a dedicated chip manufacturing plant.

Speaking at Tesla’s annual shareholder meeting on November 6 (local time), Musk said, “AI5 chips will effectively be manufactured at four sites,” specifically naming Samsung Electronics, as well as TSMC’s production facilities in Taiwan, Texas, and Arizona. He added that even under optimistic scenarios, “the volume of chip supply from partners will not be enough.”

“We respect our partners, TSMC and Samsung, and we may collaborate with Intel as well,” Musk said. “However, the pace of chip production still will not meet the level of demand we foresee.” He suggested that Tesla may have to establish a “Tera Fab,” a chip factory potentially larger than Tesla’s existing Gigafactories.

While Musk did not disclose a timeline or location for the plant, he emphasized that chip production capacity is becoming central to Tesla’s strategy as AI development accelerates.

Musk also revealed that Tesla already has a roadmap for an AI6 chip that will follow the AI5. “We expect to transition to AI6 within a year of beginning AI5 production, doubling performance metrics on the same manufacturing footprint,” he said.

Musk highlighted that Tesla’s custom AI chip design emphasizes exceptionally high power-efficiency and cost-efficiency, arguing that it could achieve similar performance to Nvidia’s latest Blackwell chips at a fraction of the power consumption and cost. “We believe AI5 will consume roughly one-third the power at less than 10% of the cost,” he stated.

According to Musk, the advantage comes from developing chips tailored to Tesla’s own use cases, rather than designing for broad, generalized workloads. “Nvidia chips are excellent, but they must satisfy many different customers,” he said. “We only have to satisfy ourselves, which allows extreme simplification. That is where we are unique.”

Musk also reiterated that AI is central to Tesla’s future beyond electric vehicles, highlighting progress in full self-driving (FSD) software and Tesla’s humanoid robot, Optimus. He predicted that once mass-production scales to around one million units per year, Optimus could be priced at approximately $20,000 per unit.

Additionally, Musk confirmed that the autonomous ride-hailing vehicle “Cybercab” aims to begin production in April next year, the electric truck Semi will enter volume manufacturing next year, and the second-generation Roadster sports car will be unveiled on April 1, with production starting 12 to 18 months afterward.

The announcements reaffirm Tesla’s transition from a pure EV manufacturer into a vertically integrated AI-driven robotics and semiconductor company.