"Intel is proud to join the Terafab project with @SpaceX, @xAI, and @Tesla to help refactor silicon fab technology"

[tooLongInEDA]

Sometimes I wonder if there is a master plan here that is going on behind the scenes. Then I realize that Elon Musk and other tech titans are just human. The difference being that Elon thinks big and he thinks out loud. While most of us geniuses think before we speak Elon and Sam Altman speak first. Instead of saying "what if we can't make enough wafers" he states it as a fact to get the discussion amongst us professionals started. I guess this is how disruption happens? Has Elon even spoken with TSMC about this? That would have been my first call. Clearly he has already spoken to Samsung and Intel. Save the best for last?

There will be a Q&A at the TSMC event next week. Toss me some Qs and I will ask.

It's all very well to assume that Elon Musk is a genius and that this somehow gives him the right to open his mouth without much thought and brainstorm in public. But not if he's running a publicly listed company and these pronouncements are materially affecting stock prices (whether intentionally or not). There are rules here and we shouldn't be cutting Musk some slack just because he's a "genius". Off topic, DJT exhibits very similar behaviour and no one's accused him of being a genius (at least not that I recall).

Going further, I'm not convinced that Musk says such things with the primary intention started "among us professionals" ! The bloke's got an IPO to push later this year IIRC.

 

[Xebec]

And if TeraFab is built with Intel’s help, will it eventually become a fierce competitor to Intel?

If Intel had sufficient financial strength, should it simply walk away from this deal instead?

Agree with your comments in general; and you're right, Intel is not in a position of strength right now.

Without TeraFab, Intel gets 0% of SpaceX and Tesla's business, and leaves the door open for TSMC and Samsung to partner and potentially profit from TeraFab. (TSMC and Samsung have been really good at starving Intel of Foundry business.. )

Honestly, any involvement in TeraFab is better than zero involvement for Intel, as long as Intel is smart about how they engage and what they agree to. I hope LBT is looking at all viewpoints, and receiving contrarian feedback to whatever the Board is thinking to make the right decision here.

 

[Xebec]

View attachment 4439

Some adjustments to consider for your model:

- Space Data Center is 0 until ~ 2029 or 2030

+ However, AI6 (2028) should start going into an Earth-based datacenter for "next gen Tesla Dojo"

+ Tesla autonomous driving is potentially much higher in just 2-3 years; they sell 2M vehicles currently, and need 4M AI4 chips as of 2025. By 2027 they (should be?) producing a large number of Cybercabs on top of their existing lineup, potential for 4M+ vehicles in 2027, and maybe 8M+ in 2030 if they really get FSD unsupervised working in at least North America and EU.

+ AI5 chip is rumored to be significantly larger than AI4 die-wise

+ TeraFab intends to bundle/fabricate memory for AI chips; with a rumor of "9X AI4" memory per AI5 chip - so something on the order of 128GB or 144GB for each of those AI5 chips. Maybe assume 256GB/288GB for AI6.

+ Tesla sells a non trivial amount of battery storage for infrastructure and home (~ 47 GWh in 2025), and of course vehicles. They could re-use AI chips or make new chips to support/control these products rather than using COTS.

+ With an Intel partnership, Intel could win back Tesla's infotainment business back from AMD -- i.e. 4-5 million Intel SoCs by 2030 could be made at this fab (up from ~ 2M Ryzen today).

+ Tesla's goal is ultimately 10 million robots per year, but I think your numbers are much closer to reality.

 

[ai268]

Sometimes I wonder if there is a master plan here that is going on behind the scenes. Then I realize that Elon Musk and other tech titans are just human. The difference being that Elon thinks big and he thinks out loud. While most of us geniuses think before we speak Elon and Sam Altman speak first. Instead of saying "what if we can't make enough wafers" he states it as a fact to get the discussion amongst us professionals started. I guess this is how disruption happens? Has Elon even spoken with TSMC about this? That would have been my first call. Clearly he has already spoken to Samsung and Intel. Save the best for last?

There will be a Q&A at the TSMC event next week. Toss me some Qs and I will ask.

TSMC to Musk: There's No Shortcuts in the Chip Business​

TSMC to Musk: There's No Shortcuts in the Chip Business

[Opinion] CC Wei is playing tortoise as Elon Musk takes the role of hare in the global AI chip race.

www.culpium.com

Of course, Musk’s move might be a massive headfake, designed to pressure TSMC to expand even quicker so that he can get on with the task of launching AI data centers in space. If so, Wei isn’t buying it, and he’s certainly not about to accelerate TSMC’s growth trajectory to just to stave off a possible loss of orders to Intel and Terafab.

“We repeatedly say that we prepare capacity to meet customers’ demand, not because of our competitor and not because of other considerations,” he said. “The most important one is our customers’ demand, and they work with TSMC so that we plan our capacity and our capital expense.”

Of course, this might be TSMC’s own psyops. The flipside to the “expand now, pay for it later” thesis is that extra industry capacity would give clients more leverage when a slowdown comes, especially since a chipmaker’s capital costs are baked in to its balance sheet and income statement. Seeding doubts into Musk’s mind, Inception style, could pay off bigtime if the world’s richest man later gets cold feet."

 

[Barnsley]

Lets compare what is happening vs exagerrated stories on Terafab.

Musk has committed to build a 25B factory in Texas with colocations of logic, memory packaging and mask making. this is called Terafab. This makes sense althought low volume IDM is a very bad idea, but only if you care about cost effectiveness.

Musk has said he want Billions of AI chips, 1M space station AI datacenters, and capacity 50x todays capacity. Cost is 2-5 Trillion. This is also called Terafab
This is nonsense and a dream.... we need to stop calculating financial impact of it. If he is wildly successful, he will have 4 Fabs 20 years from now making Chips for his companies.

I wonder why people make wild dream comments like this .... then I remembered: Musk is the richest man in the world. He apparently watches Bond movies and says "See rich people do this stuff all the time". Maybe Moonraker and Die Another day are his models. Maybe Austin Powers?????

Please correct me if I am missing something

 

[MKWVentures]

Need a AI meme of Musk saying how he will use a ""LASER"" to blow up the world. Maybe a "Mini Musk" too. Playing the part of Robert Wagner has to be Trump.

 

[Xebec]

TSMC to Musk: There's No Shortcuts in the Chip Business​

I was at a Lockheed Martin all hands meeting in Texas when they said the same thing about how hard it was to get to space, and that SpaceX would fail at re-usable rockets.

Musk doesn't like monopolies -- he ran over ULA with a (space) train, and the car industry is now in major upheaval (fossil fuel cars and legacy auto models being the monopoly). Occams Razor - Musk wants more capacity at a lower price than TSMC is willing to do (look how expensive their margins are becoming), rather than a feint.

It remains to be seen if Musk can keep raising capital to make something like this succeed (will the AI bubble bust first?), but this is a missed opportunity for CC Wei. It's also the kind of opening Intel really needed as TSMC rarely mis-steps.

 

[Xebec]

Somewhat related - revealed specs on Tesla AI6:

• - AI6 doubles AI5 performance, despite AI5 already being a 5× jump over the dual‑SoC AI4 hardware.
• - AI6.5 is a surprise mid‑cycle upgrade, shifting production from Samsung’s 2nm Texas fab to TSMC’s 2nm Arizona fab for extra efficiency.
• - Both chips move to LPDDR6 memory and allocate ~50% of TRIP accelerators to SRAM, giving order‑of‑magnitude higher effective bandwidth than DRAM.
• - AI6/6.5 are aimed at future autonomy platforms—Cybercab, Optimus, and Tesla/xAI’s TERAFAB—rather than current consumer vehicles.
• - Tesla’s vertically integrated TERAFAB pipeline targets a nine‑month chip iteration cycle, accelerating hardware evolution for unsupervised autonomy.

Elon Musk Shares Specs for Tesla's AI6 Chip, Teases AI6.5

Just hours after confirming that the next-generation AI5 chip has finished its design phase, Musk also shared some eye-watering specs for its successor

www.notateslaapp.com

 

[KevinK]

I was at a Lockheed Martin all hands meeting in Texas when they said the same thing about how hard it was to get to space, and that SpaceX would fail at re-usable rockets.

Beating Mil-Aero companies is easy (my wife has all kinds of stories from her first job about how screwed up procurement is and how conservatively the military plans and manages projects). My take is that disrupting the entrenched fossil/auto industry was also somewhat easy, because underlying cost and technological dynamics were changing underneath the industry (higher fuel costs, drops in solar and battery pricing due to scale, growing electronic content in autos to the point of app-ification of autos). Not clear that silicon chips are at a disruption point, though multi-die packaging is clearly changing cost and integration contours, and AI is driving a completely new design point for system density / scale / speed.

 

[Barnsley]

How many Cybertrucks doea Intel have to buy to help close out any deals

 

[Xebec]

Beating Mil-Aero companies is easy (my wife has all kinds of stories from her first job about how screwed up procurement is and how conservatively the military plans and manages projects). My take is that disrupting the entrenched fossil/auto industry was also somewhat easy, because underlying cost and technological dynamics were changing underneath the industry (higher fuel costs, drops in solar and battery pricing due to scale, growing electronic content in autos to the point of app-ification or autos). Not clear that silicon chips are at a disruption point, though multi-die packaging is clearly changing cost and integration contours, and AI is driving a completely new design point for system density / scale / speed.

From a technical standpoint - I agree & that's true, both auto and defense were ripe for competition. They stagnated by design.

But both have immense lobbying organizations, with the defense contractors being literally in bed with the government. (They hire a lot of government employees at high prices as consultants, whom have connections back into the military and government. ). In addition, certainly a lot of capital was required to shake these markets up - which is why we didn't have much disruption over time.

I think there's a chance that the advanced node industry could be ripe for some disruption too:

Business-wise/Economics:
- Moore's Law has slowed down -- being a node ahead means a lot less than it used to
- A lot of people don't want to see a TSMC monopoly, that means a competitor no longer has to 'beat' TSMC, just be a good enough alternative
- Memory seems to be an area of opportunity for profit over the next 5 years given AI demands

Regarding shortcuts, are there really zero options?
- Intel has a lot of technology that isn't being fully utilized
- (I believe) Intel has fabs that could be expanded upon - both shells and not fully filled fabs, i.e. TeraFab doesn't have to start 'at scratch'
- TeraFab partnering with Intel doesn't preclude them from sourcing IP from Samsung or Rapidus
- Is there a chance Intel or Samsung have EUV equipment they're not using?

This is an admittedly bullish take on the situation, and I think the advanced fab industry is a much harder target than auto.

 

[siliconbruh999]

- Is there a chance Intel or Samsung have EUV equipment they're not using?

Samsung Maybe Intel nope they didn't invest in EUV till later point of time