Terafab 21 March 2026

[user nl]

Elon Musk has announced the launch of a massive semiconductor project in Austin, Texas, dubbed "Terafab," aimed at internalizing chip production for his expanding ecosystem of AI, robotics, and aerospace ventures.

The facility, which will be jointly operated by Tesla Inc (NASDAQ:TSLA) and SpaceX, is designed to bypass what Musk characterized as a sluggish global semiconductor industry unable to meet his aggressive scaling requirements.

The project targets the eventual support of a terawatt of computing power annually, facilitating the transition toward autonomous driving, humanoid robotics, and space-based data centers.

Around 33 minutes starts the compute needed and the Terafab chip buildout:

https://x.com/i/broadcasts/1yKAPMzlvgWxb

https://finance.yahoo.com/sectors/t...nces-terafab-internalize-tesla-062638546.html

https://www.basenor.com/blogs/news/tesla-terafab-inside-the-20b-chip-factory-that-changes-everything#:~:text=available for production."-, UPDATE — March 22, 2026,xAI, located near Gigafactory Texas.

 

[DanX]

"Assume that Columbus the navigator monopolized ships, monopolized new energy (oil vs. coal), and monopolized collaboration opportunities among navigators. How should the New World be valued?"

 

[DanX]

"500 years later, the GDP of the New World (North America + South America + Oceania) is more than twice that of the Old World (Europe)."

 

[Daniel Nenni]

Here is the disconnect: Elon said that the foundries were not comfortable with committing to the output that he requires. Current chip output is at 2% what they need so it is significant. Elon said he would buy all of the chips they can make etc...

To me this does not pass the sniff test. If Elon was able to write a big enough check, which I assume he can, TSMC would build the fabs. They do this for Apple and Nvidia, they can do this for Tesla/SpaceX.

The key here, and in other Elon Musk ventures, is that Elon thinks he can do it better than TSMC. I certainly hope this is true. This will push TSMC and the other foundries to work harder and that is a good thing.

I'm a big fan of disruption. I met Steve Jobs when he founded NEXT. Steve was a disrupter. Elon Musk even more so. We should, however, remember that disrupters make mistakes too. There are a lot of pivots in semiconductor disruptions. The laws of physics still apply with semiconductor disruptions. The chances of Elon Musk being successful here within a fixed amount of time is infinitesimal. Elon/Tesla promised out of the gate fully autonomous cars and he did not deliver and still has not delivered. It was a much bigger problem than even Elon imagined.

TSMC has 30+ years of experience building fabs and wafers. TSMC's success today is largely due to introducing disruptive technology in a safe and sane manner where customers can trust TSMC to meet their capacity requirements. That is the key to the foundry business model, trust.

So you have to ask yourself: Self, do you trust Elon Musk in regards to building a Terafab and yielding leading edge wafers in a fixed amount of time?

Nope.

It was interesting that Elon never mentioned Samsung even though they have a multi billion dollar agreement at 2nm.

 

[Jert]

The chances of Elon Musk being successful here within a fixed amount of time is infinitesimal. Elon/Tesla promised out of the gate fully autonomous cars and he did not deliver and still has not delivered. It was a much bigger problem than even Elon imagined.

1. Completely agrees on the first point. As a shareholder, I'm worried he's gonna burn all that cash for nothing. Creating another TSMC is indeed harder than going to Mars.
2. Do not agree on the second point. The latest Tesla cars with FSD is basically autonomous driving or 99.99% there. It drives itself out the garage and will auto park when arrives destination, all by itself. The only time you may touch the steering wheel is when you get in and get out . The main barrier for autonomous driving is gov regulations and general public fear (but once tried, I think most people will let go their hands on the wheel)

 

[user nl]

So you have to ask yourself: Self, do you trust Elon Musk in regards to building a Terafab and yielding leading edge wafers in a fixed amount of time?

Nope.

It was interesting that Elon never mentioned Samsung even though they have a multi billion dollar agreement at 2nm.

https://www.basenor.com/blogs/news/... 22, 2026,xAI, located near Gigafactory Texas

UPDATE — March 22, 2026​

Terafab ≠ Giga Texas Fab: Elon Musk has clarified that the Advanced Technology Fab announced for Giga Texas is a separate, smaller facility focused on iterating chip designs — not the full Terafab project. The true Terafab will require thousands of acres and over 10GW of power at full scale, and no location has been confirmed yet, with several sites still under consideration.

 

[blueone]

Do not agree on the second point. The latest Tesla cars with FSD is basically autonomous driving or 99.99% there. It drives itself out the garage and will auto park when arrives destination, all by itself. The only time you may touch the steering wheel is when you get in and get out . The main barrier for autonomous driving is gov regulations and general public fear (but once tried, I think most people will let go their hands on the wheel)

I think FSD is already safer than most human drivers. (Especially the jackasses who live in my neighbor and run stop signs and speed on roads with cyclists and pedestrians running in the bike lanes. ) The problem is that FSD is expected to be essentially perfect. I think FSD as originally envisioned by Musk is probably unachievable, given the camera-only strategy and transformer neural networks. I read that Tesla is rumored to be developing world models, but even if that's the case a new strategy would be years away from general availability.

 

[Daniel Nenni]

1. Completely agrees on the first point. As a shareholder, I'm worried he's gonna burn all that cash for nothing. Creating another TSMC is indeed harder than going to Mars.
2. Do not agree on the second point. The latest Tesla cars with FSD is basically autonomous driving or 99.99% there. It drives itself out the garage and will auto park when arrives destination, all by itself. The only time you may touch the steering wheel is when you get in and get out . The main barrier for autonomous driving is gov regulations and general public fear (but once tried, I think most people will let go their hands on the wheel)

I'm not a Tesla owner so your opinion is much more qualified than mine. The point I was making is that Elon Musk made the FSD promise many years before he actually delivered:

Major FSD promises

• 2015 — Musk said Tesla would have complete autonomy in ~2 years.
• 2016 — He pledged a Tesla could drive from Los Angeles to New York by end of 2017.
• 2019 — He said FSD would be feature-complete by end of 2019, and by 2020 drivers could “snooze” while the car drives.
• 2019 (again) — Musk predicted 1 million autonomous robotaxis by mid-2020.
• 2020–2025 — He continued to promise full autonomy “this year” multiple times.
• 2025 — He said unsupervised robotaxis would launch in Austin soon.
• 2026 — He again discussed needing more data after missing another deadline.

Bottom line: TSMC has nothing to worry about.

 

[Xebec]

2. Do not agree on the second point. The latest Tesla cars with FSD is basically autonomous driving or 99.99% there. It drives itself out the garage and will auto park when arrives destination, all by itself. The only time you may touch the steering wheel is when you get in and get out . The main barrier for autonomous driving is gov regulations and general public fear (but once tried, I think most people will let go their hands on the wheel)

The exact goal post for FSD matters; Based on thousands of miles of first hand experience, I'd rate it :

99.99% for : Good/Great weather conditions, well marked roads, no potholes, well mapped cities - probably better than an average human
<highway driving in here somewhere>
99.9% for: Rain and foggy weather, most road obstructions - probably better than an average teenager driver
95% for: Parking garages (negotiates the gates well, but parks in 'don't park here' spots occasionally, home garages (returning home), parking lots (need more options for parking spots)
85-90% for: Pothole avoidance (i.e. northeast roads at the end of winter) - worse than someone whose had two beers
30% for: Snow and Ice condiitons (it does some basic slowdowns, but it's like v0.2 beta here), cameras need manual cleaning often in some conditions - only for the very brave
20% for: Speeding cameras (ask my wife, and no that's not an internet deflection ).

 

[Xebec]

Here is the disconnect: Elon said that the foundries were not comfortable with committing to the output that he requires. Current chip output is at 2% what they need so it is significant. Elon said he would buy all of the chips they can make etc...

The output could mean how fast he wants to rapidly change steppings or spins of a product, too - not just the quantity. I assume (perhaps incorrectly) that the foundries are a lot slower at allowing you to iterate on chips that are produced on wafers, than say -- Intel could do for itself as an IDM?

(I'm happy to be wrong - but I suspect his -- fail fast philosophy is another dimension that may not be well suited for the foundry model, since they have so many other commitments).

 

[user nl]

I'm not a Tesla owner so your opinion is much more qualified than mine. The point I was making is that Elon Musk made the FSD promise many years before he actually delivered:

................................................................................................

Bottom line: TSMC has nothing to worry about.

Indeed, I'm sure TSMC's Fab 25 will be running many kwspm at 1.4 nm by the time Musk has his first EUV-tool installed in his Terafab,

Maybe someone should put a bet up somewhere (on polymarket?) on

Who will have produced the most wafers with >70% yield by 2030: Musk at Terafab in 2 nm, or TSMC in 1.4 nm



https://www.trendforce.com/news/202...tedly-under-construction-or-starting-in-2026/

 

[Brady]

Tom’s Hardware says it’s “reportedly $20B” but that’s barely a single leading edge fab module these days.

Zero chance an end to end fab ecosystem to output “1 terwatt of compute” (why do we keep using power as the target now…) is going to cost anything less than $50-100B just to start.

 

[blueone]

The output could mean how fast he wants to rapidly change steppings or spins of a product, too - not just the quantity. I assume (perhaps incorrectly) that the foundries are a lot slower at allowing you to iterate on chips that are produced on wafers, than say -- Intel could do for itself as an IDM?

(I'm happy to be wrong - but I suspect his -- fail fast philosophy is another dimension that may not be well suited for the foundry model, since they have so many other commitments).

Much to my surprise, I'm told TSMC allows "hot lots", just like Intel fabs did/does for CPUs. I'm also told that needing one is so expensive at TSMC it could get an engineering leader demoted or fired, but the fact TSMC supports them is different than I expected to hear.