TSMC and Samsung give a boost! Tesla AI5 chip completes design approval, targeting mass production in mid-2027

Tesla CEO Elon Musk (Elon Musk) announced on the X platform today that its in-house developed next-generation AI chip, AI5, has completed the final design (tapeout). Performance is five times that of the current dual SoC AI4 setup. The mass-production target is set for mid-2027, and it is expected to be used for Full Self-Driving (FSD) autonomous driving technology and the Optimus humanoid robot program.

Congrats to the @Tesla_AI chip design team on taping out AI5!

AI6, Dojo3 & other exciting chips in work. pic.twitter.com/hm54TdIzBx

— Elon Musk (@elonmusk) April 15, 2026

AI5’s performance comprehensively surpasses AI4, matching Nvidia H100, Blackwell GPUs

AI5, a system-on-chip (SoC) designed by Tesla specifically for real-time AI inference in vehicles and robotics applications, will be used to replace AI4, which has been installed in its vehicles since early 2023. According to what Musk said in a post, AI5’s computing power is about eight times that of AI4, memory capacity increases by nine times, and bandwidth expands by five times. Overall performance is estimated to reach between 2,000 and 2,500 TOPS, representing a major leap compared with AI4’s 300 to 500 TOPS.

Musk revealed that the inference performance of a single AI5 chip is about comparable to the Nvidia H100 GPU, while a dual-chip configuration can rival Nvidia Blackwell processors—but at far lower cost and power consumption than the latter. In terms of architecture design, AI5 is deeply optimized for low-precision inference workloads, pursuing Musk’s so-called “extreme simplification” design philosophy.

Musk personally stays on site to oversee: AI5 is crucial to Tesla’s survival and success

In January, Musk said that the AI5 plan is a life-or-death strategic priority for Tesla. The chip’s performance for Tesla’s own application scenarios will be better than any alternative solution: “Solving the AI5 problem is a survival-level issue for Tesla, which is why I have to personally oversee it, spending every Saturday for several months working on this chip.”

AI5 is at the core of Tesla’s vertically integrated AI strategy. Hardware and software adopt a co-design approach, aiming to maximize the utilization of every circuit resource.

From the product impact perspective, the model parameters used by the current FSD software are about one billion; the next-generation v15 version will expand to about ten billion, fully relying on AI5’s compute power to support it. The Optimus humanoid robot will also gain real-time inference capabilities through AI5, without needing to rely on cloud connectivity to quickly process sensor data.

Partner with TSMC and Samsung for mass production; advance Terafab, a self-built fab, in parallel

In its manufacturing strategy, AI5 adopts a dual-factory parallel model. It simultaneously commissions production at TSMC (TSMC)’s Arizona plant and Samsung (Samsung)’s Texas plant to ensure supply-chain resilience and mass-production capacity. Musk explained: “Although the two foundries use their own process technologies, the chip designs being produced are completely identical.”

At the same time, Tesla is building its own fab, Terafab, in Texas. In the future, it will take on larger-scale production capacity. The company also plans capital expenditures for 2026 of up to $20 billion for this purpose, including investments in Terafab construction and projects such as the Cybercab robotic taxis and Optimus robots.

(Musk announces Terafab rollout in Texas: combining SpaceX and Tesla xAI to accelerate chip manufacturing processes)

AI5 is expected to go into mass production in mid-2027, while the AI6 final design target is set for end of 2026

On the mass-production timeline, small-batch engineering samples for AI5 are expected to come out by the end of 2026 for early Optimus testing or for development vehicles. Mass production for vehicles is targeted for mid-2027. It is worth noting that Tesla’s dedicated robotic taxi, Cybercab, will not wait until AI5 is ready; instead, it will be launched first using the current AI4 hardware.

In subsequent planning, Musk has set an aggressive iteration schedule: the goal is to release a new chip design every 12 months and reach mass production, ultimately compressing the design cycle to nine months. The AI6 final design target is set for December 2026. AI7 and subsequent generations have already entered the planning stage as well, showing Tesla’s determination to build an advantage in its own chips.

This article, “TSMC and Samsung help out! Tesla’s AI5 chip completes the final design, targeting mid-2027 mass production,” first appears on Chain News ABMedia.