Elon Musk Talks About Building “Gigantic” Tesla Fab, Partnering With Intel for AI Chips

Tesla CEO Elon Musk has said that he expects the company’s upcoming in-house AI processing units to be exceptionally powerful, and claims the company has mapped out plans for at least the next four generations. The upcoming AI5 chip is claimed to be 40-50X as capable as the current-gen HW4 platform, with 9X the memory and 8X raw compute power. 

Following a series of updates via social media over the past few weeks, Musk reiterated the importance of Tesla’s in-house chip designs at the company’s annual shareholders’ meeting, during which his $1 trillion pay package was approved. He also mooted the idea that Tesla might have to build its own “Terafab”, implying orders of magnitude greater scale than the car manufacturing plants he calls Gigafactories. 

Investing in establishing such a facility could cost tens of billions of US dollars, and would take several years to achieve. Acknowledging this, Musk also commented at the meeting that he could choose to partner with Intel, which has been trying to attract customers for its foundry business for several years. While he did state that no agreement is currently in place, it was not clear whether preliminary talks between the two companies have been held. 

Musk did not further expand on these comments or indicate whether such a deal would be temporary, till his own proposed facilities can be made operational. He has recently confirmed that the upcoming AI6 will be fabricated by both Samsung and TSMC within the US in order to promote the domestic semiconductor industry and hedge against potential supply chain disruptions. Plans for in-house manufacturing, or a tie-up with Intel, would therefore likely only apply to subsequent chip models. 

In previous comments, Musk had asserted that the AI5 platform would be ten times as powerful as HW4, which debuted in January 2023, but could consume up to 800W of power as opposed to 160W. The chip has also been delayed from January 2026 to the end of next year, which means AI6 will likely not enter production before 2028, and in-house fabrication or a deal with Intel would come into play even later.

Musk, who has made several hyperbolic statements and promises before, expects to need enough chips in 2026 for a Cybercab to roll off its factory lines every 10 seconds once production ramps up to full capacity. Similarly, he envisions over 10 million Optimus 3 robots built per year, starting with at least one million in 2026. He referred to Tesla’s production increases as “faster than anything’s ever been ramped up before in human history”.

Designed completely in-house, AI5 is said to be a “ground-up” redesign with several key AI acceleration features, and several redundant legacy components stripped out. Musk also recently stated that Tesla could deliberately overproduce this chip in order to have enough for its devices as well as datacentres. The company has not said anything about selling chips to other companies, which would put it in direct competition with Nvidia on the world stage. 

The hardware will power Tesla’s next-gen models, though it will not be ready in time for the announced initial rollout of the Cybercab, which is designed to be completely autonomous. Tesla cars currently use HW4 for all self-driving features, which the company calls Autopilot and Full Self Driving (Assisted). Future chips are also expected to power upcoming Optimus humanoid robots and other products that could heavily leverage on-device AI processing.