Tesla critical
Elon Musk says all of Tesla depends on one tiny — but very expensive — proposition
The CEO says the future of Tesla hinges on its next-generation microchips — but securing enough of them will be costly
All of Tesla Inc. depends on making a tiny silicon chip that will power everything from driverless technology to robots, Chief Executive Elon Musk said last week. But getting enough of those semiconductors will be an expensive task.
To start, Tesla
Samsung has a $16.5 billion deal that lasts from July through the end of 2033 to supply chips to Tesla. However, Musk has said that figure is “just the bare minimum” of what Tesla could pay Samsung for its chips.
Tesla initially planned to have Samsung and TSMC alternate making the different generations of Tesla’s hardware, but it now plans to have both companies make the AI5. That’s somewhat uncommon, since it will likely mean spending additional funds to adjust the chip’s design to match each manufacturers’ respective equipment.
“If you’ve designed a chip for TSMC, and now you want to bring it to Intel, it’s not just, you know, ‘Here, take it,’” explained Jack Gold, a technology analyst and the president of J.Gold Associates, a consulting firm. Rather it’s about “what do I have to do to make it compatible with your design processes and manufacturing processes?” Gold added.
Some of Tesla’s costs will likely be diminished because the company has streamlined the AI5 chip to match its needs. Musk has said Tesla has a “long list” of components it took out of the AI5 chip, which he said will improve performance and allow Tesla to fit the chip on a “half reticle,” or half the size of chip designs from Nvidia Corp.
Don’t miss: Nvidia earnings have become crucial to the stock market — and this time even more so
Generally, smaller chips also translate to a higher yield per silicon wafer used in the lengthy chipmaking process, because more potential products can fit on each wafer used. Simplifying the chips should also make it easier to adjust the design to match multiple chipmakers’ technology, Gold said.
“We’re going for radical simplicity,” Musk said in October. “The net effect is that I think AI5 will be the best performance per watt, maybe by a factor of two or three, and best performance per dollar for AI, maybe by a factor of 10” compared to rival chips.
A small number of AI5 chips will be made next year, with volume production only possible in 2027, Musk said recently.
According to Tesla’s timeline, that means the first Cybercab robotaxis and Optimus humanoid robots won’t have the chips, because production for those begins in 2026. And Tesla’s data centers, which currently use a mix of Nvidia chips and older Tesla hardware, will have to wait to get the AI5.
During Tesla’s third-quarter earnings call in October, Musk said he hoped for an oversupply of chips. But a few weeks later at Tesla’s annual meeting, he wondered aloud about a potential partnership with Intel Corp.
and a possible Tesla-built factory to supply even more. Intel declined to comment.“Even when we extrapolate the best-case scenario for chip production from our suppliers, it’s still not enough,” Musk said at the Nov. 6 annual meeting. “I think we’re probably going to have to build a gigantic chip fab.”
The production lines for Tesla’s Optimus robots in Fremont, Calif., are expected to generate 1 million units per year, according to Musk, who also thinks Tesla could eventually wind up making as many as 3 million Cybercabs per year. Tesla is currently capable of making more than 2.3 million vehicles across its two U.S. factories and facilities in Germany and China.
The company is on track to sell 1.6 million electric vehicles this year, according to a consensus of analyst estimates on FactSet.
Musk’s hypothetical “Terrafab” would be capable of making 100,000 wafer starts per month and eventually scale to produce about 1 million per month, he said. For context, TSMC — the market leader — made about 1.42 million wafers per month across its factories last year.
Generally, it can take between five and seven years to build a new chip factory, according to Ramiro Palma, a managing director at Boston Consulting Group who specializes in the technology industry. That includes the time it takes to pick a site and plan construction, in addition to actually building the factories.
Such facilities are also expensive, even with government incentives. Samsung and TSMC have been awarded up to $4.745 billion and $6.6 billion, respectively, by the U.S. government’s Chips and Science Act.
Samsung’s plan for a pair of fabs and a research-and-development lab in Taylor, Texas, was announced in late 2021 and is expected to cost the company at least $17 billion. The plant is set to begin operations in late 2026, about two years later than initially expected. Additional work must be completed by the end of December 2028, according to the city’s updated tax-abatement agreement with Samsung.
TSMC said it would invest $65 billion in Arizona to build three fabs set to be operational by the end of the decade. In March, TSMC said it would invest another $100 billion to develop three additional fabs, plus two advanced packaging facilities and a research-and-development center.
Both facilities will make chips for Tesla, although TSMC also has deals with customers like Apple Inc
Copyright © 2025 MarketWatch, Inc. All rights reserved.
Comments
Post a Comment