Elon Musk says Tesla could build a gigantic chip fab and work with Intel to meet demand

[midian182]

What just happened? Fresh from being awarded his $1 trillion pay package by Tesla, Elon Musk has announced that the EV giant will probably have to build "a gigantic chip fab" to ensure its AI and robotic ambitions are fulfilled. Musk also said he was considering working with Intel to help meet chip demand.

Speaking at the annual shareholders meeting this week, Musk said, "One of the things I'm trying to figure out is – how do we make enough chips?"

"You know, maybe we'll, we'll do something with Intel," he added. "We haven't signed any deal, but it's probably worth having discussions with Intel."

Tesla currently relies on TSMC and Samsung as its contract chipmakers for the hardware powering its next-gen full-self-driving systems. The automaker signed a $16.5 billion deal with Samsung earlier this year for its future A16 chip to be produced at the Texas fab. It now seems Intel could be added as another chip-building partner.

A deal with Intel would give its foundry business a boost. Reports in August claimed its 18A process has been experiencing low yields and quality issues, but since then we've heard that it could begin manufacturing x86 chips for AMD, and reports suggest it has signed Microsoft as a major client for 18A.

Intel's share price jumped 2.2% following Musk's comments, while Tesla's shares were up 4% in after-hours trading.

Intel's year-to-date share price

Musk said this week that Tesla is finalizing the design of its A15 chip, which will apparently see a 40x improvement compared to the current A14. A small number of A15 units will be produced in 2026, with high-volume production only possible in 2027.

A16, meanwhile, is predicted to double the performance metrics of its predecessor and go into volume production in 2028. Musk added that it would be inexpensive, power-efficient, and optimized for Tesla software. He claimed it would probably consume about a third of the power used by Nvidia's Blackwell chip, but cost 10% less to make.

And there is a clear path to a doubling of performance on all metrics for AI6 within 10 to 12 months of AI5 shipping https://t.co/ziI6dNkrKk

– Elon Musk (@elonmusk) November 5, 2025

"But even when we extrapolate the best-case scenario for chip production from our suppliers, it's still not enough," Musk warned. As such, Tesla will probably need to build a "gigantic" chip fab, which Musk described as a "Tesla terra fab."

"I can't see any other way to get to the volume of chips that we're looking for," he said. The CEO claimed the Tesla fab would have an initial capacity of 100,000 wafer starts per month and eventually scale to 1 million. For comparison, the world's largest contract chipmaker, TSMC, has a capacity of around 1.42 million per month.

Permalink to story:

Elon Musk says Tesla could build a gigantic chip fab and work with Intel to meet demand

 

[Plutoisaplanet]

It sounds like the 40x jump in performance from AI4 to AI5 has the caveat of only being applicable in “certain tasks” which refers to Softmax operations. Meanwhile, in other areas the performance is expected to jump by 8x.

What’s notable about Tesla’s strategy is not using generalized chips from Nvidia that excel at AI, but designing custom ones that have low power requirements designed for the car. Elon Musk previously cited a 100W target.

In addition, at some point he started talking about using the chips for both cars (inference) and data centers (training) if they ever ordered too many. Considering that supply is lacking, Nvidia accelerators the primary goto in data centers.

 

[Revolution 11]

This doofus can't make car panels align properly, now he is going to make semiconductors, the most demanding precision product on the planet?

Give me a break. This is just another attempt to keep the stock price high. Nothing more, nothing less.

 

[BobDolesGhost]

The entire article's reference to Tesla's "a16" chip is driving me crazy. The next gen Tesla chip is "ai6" (lower cased here to emphasize the correction).

Making state of the art cutting edge silicon is not for the feint of heart. It takes many many billions of dollars to design and construct a successful fab, an enormous investment of upfront capital. And its HARD. AMD's fab was spun off into a new entity "Global Foundries" because it was unable to keep up. GFs also gave up on modern nodes and has since just concentrated on mature technology nodes.

Intel is famous for its trouble in modern nodes. I'm sure everyone here remembers the 14nm++++++++ series processors because their 10nm fab was so troublesome. They're again having problems with their 18A (angstrom, they swapped marketing names). Samsung is the world's second best fab, and they also have yield issues and performance not on par with the world leader, TSMC.

TSMC is the world premier fab. Basically 100% of TSMC's modern production is spoken for. It is hard to squeeze your way in to get your chips made. Apple and Nvidia swallow enormous portions of TSMC output, about 45% together. Pretty much every Nvidia chip manufactured is sold and gobbled up. If they could get more production, I'd imagine they would.

I can't see a Tesla + Intel, partnership doing well. Intel would have to get its fab in order first. Tesla buying silicon cylinders at exorbitant cost and getting an awful yield, would only yield irritation in the relationship. Perhaps they get a PR boost and maybe get tax breaks from the US Gov to make it worthwhile, as it would be a large modern chip partnership on US soil. Well, could be on US soil. Intel has fabs elsewhere as well.

Rambled enough I guess. Good Friday everyone!

 

[wiyosaya]

No news here. fElon is just passing gas, yet again.

Move along.

 

[James321]

He and he's cronies need gag orders against them because you get too many *****s following every word they say and praise it like they are gods.

Stock market has become a shitting ground with these type of comments.

 

[Plutoisaplanet]

I can't see a Tesla + Intel, partnership doing well. Intel would have to get its fab in order first. Tesla buying silicon cylinders at exorbitant cost and getting an awful yield, would only yield irritation in the relationship. Perhaps they get a PR boost and maybe get tax breaks from the US Gov to make it worthwhile, as it would be a large modern chip partnership on US soil. Well, could be on US soil. Intel has fabs elsewhere as well.

To be fair, Elon did say Tesla will buy tens of billions of dollars worth of training compute over the coming years, and they plan to buy these as American-made chips. They have just over $40B currently and are about $2B flow positive every quarter. So there is a chance they will actually do it despite paying a premium. Tesla has said they've on manufacturing with both TSMC and Samsung and it definitely sounds like Elon knows Intel is a last resort. He said Samsung and TSMC's production rates cannot keep up with how many they want to buy.

Tesla plans to increase the number of inference chips required over the coming years from 1.8M per year currently to 5M+ per year (dependent on the ramp of the upcoming "cybercab"). Then they plan to increase it further with a $25k+ humanoid robot over the coming 10+ years to 100M+ per year. These numbers were all shared in the annual shareholder meeting yesterday: https://x.com/I/broadcasts/1LyGBXwkNlYxN?t=55m24s

 

[Septimus]

For once, a comment that literally isn't bat **** insane from this asshat. But, yeah, as mentioned, the engineering from his exceedingly harassed team isn't the best. Not sure they will do intel any favors beyond money.

 

[bviktor]

Le good old

should
could
would

After a decade of empty promises, it's getting old.

 

[OortCloud]

Dear God, can a single day go by without that loathsome reptile Herr Musk spouting off on something?