Recently, Tesla (Tesla)CEO Elon Musk (Elon Musk) announced the launch of the largest ever fab "TeraFab" program, the goal to achieve an annual output of 1 TW (Terawatt) of AI computing power, and release in space PW level (Petawatt) computing power. The first TeraFab advanced technology factory will be located in Austin and will integrate logic chips, memory chips and advanced packaging in the same factory.

Musk said that the TeraFab plan will be jointly participated by xAI, Tesla and SpaceX. The planned production capacity covers key links such as logic chips, memory chips and advanced packaging, of which about 80% of the computing power will be used in aerospace-related fields, and about 20% of the computing power will be used for ground applications.

Musk pointed out that the TeraFab plans to build a super-large facility that integrates chip design and manufacturing capabilities, mainly to meet the rapidly growing computing power needs in the fields of aerospace, AI and robotics in the future. For example, only for the humanoid robot "Optimus" chip demand scale will reach 100~200 gigawatts (GW) level, which is the current semiconductor industry as a whole supply difficult to meet the gap.

Musk revealed that TeraFab expects to make two kinds of chips: the first will be used for edge reasoning, mainly in Tesla cars and Optimus humanoid robots. He predicts that humanoid robots will be widely used in the future. According to the current global annual output of about 0.1 billion cars, the annual output of humanoid robots can reach 10 to 100 times that of cars, that is, between 1 billion and 10 billion.

Musk emphasized that the other is a high-performance chip specifically for space AI systems. It needs to fully adapt to the more stringent conditions in the space environment, such as extreme temperatures, irradiation of high-energy ions and photons, and electron charging effects. It is expected that in the next two to three years, the deployment cost of space AI systems will be lower than that of ground systems.