Tesla has declared that it will make AI Semiconductor chips in-house. Chief Executive Officer Elon Musk said at Tesla's annual shareholders meeting on the 6th (local time) that the company is seeking partnerships to significantly expand AI Semiconductor production capacity. It is moving away from a structure that relied entirely on Taiwan's TSMC, adding Samsung Electronics as a major foundry partner, leaving the door open to potential collaboration with Intel, and even considering building its own manufacturing facilities. It is unusual for automakers to enter semiconductor manufacturing.
The AI5 chip Tesla is developing is a next-generation high-performance semiconductor that will serve as the brain for Full Self-Driving (FSD) and the Humanoid Robot "Optimus." It is being developed to target up to 2,500 TOPS (Tera Operations Per Second; 1 trillion operations per second) and will enter mass production late next year, with full-scale volume production slated for 2027. Musk said, "The AI5 chip will basically be produced in four places," pointing to Samsung Electronics and TSMC plants in Taiwan, Texas, and Arizona. He also signaled mass production of the next-generation "AI6" chip in 2028.
However, Musk judged that outside partners alone cannot meet demand. "The biggest concern right now is how to secure enough chips," he said, adding, "TSMC and Samsung are excellent partners, but even if suppliers produce under the best-case scenario, it is still not enough." He went on to say, "In the end, it seems Tesla will have to build a 'Tera Fab' itself." The company has set a goal of full internalization, handling both semiconductor design and production directly.
Not only Tesla but also major big tech corporations are jumping one by one into "AI chip self-sufficiency." Amazon handles AI training and inference on Amazon Web Services (AWS) with its in-house "Trainium" and "Inferentia" chips, Google has its "TPU (tensor processing unit)," Microsoft has "Maia," and Meta directly develops and operates "MTIA." OpenAI is also preparing a "ChatGPT-only chip" in partnership with Broadcom.
Structural factors are at work as big tech moves to achieve AI chip independence. More than 90% of the world's advanced semiconductor production is concentrated in Taiwan, creating significant geopolitical risk. At the same time, from the perspective of TSMC or Nvidia, companies like Tesla are closer to "lower-priority customers." Each time GPU supply has fallen short, product shipments have been delayed repeatedly. In addition, chips are no longer just hardware; they have become the "platform" that governs the entire AI service. To achieve full efficiency from model training to service deployment, an integrated system that ties chip design and data infrastructure together is necessary.
Nvidia's dominance is not about to crumble immediately. However, the moves by big tech companies such as Tesla, Amazon, and Google to design chips directly and even control parts of production herald a shift in the industry landscape. Competition in the AI chip industry is moving from "who can buy more" to "who can make it directly." Samsung Electronics joining as Tesla's foundry partner is a strategy that emerged amid this trend.
That said, semiconductor manufacturing requires massive investment and technical prowess. The industry believes that while big tech's in-house chip development and production may reduce dependence on Nvidia, it will be difficult to change the structure in a short period. Whether AI chip self-sufficiency by Tesla and other big tech companies will lead to supply chain diversification or result in overinvestment is likely to be decided over the next few years.