Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/elon-musk-lays-out-terafab-ai-chip-project-plan.25062/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2031070
            [XFI] => 1060170
        )

    [wordpress] => /var/www/html
)

Elon Musk lays out Terafab AI chip project plan

Daniel Nenni

Founder
Staff member
1778081190063.png


May 6 - SpaceX has proposed an initial $55 billion investment to build a semiconductor manufacturing facility in Texas, according to a filing made public on Wednesday.

Tesla (TSLA.O), opens new tab chief Elon Musk has said the electric vehicle maker plans to use Intel's (INTC.O), opens new tab next-generation 14A manufacturing process to make chips at its Terafab project, an advanced artificial-intelligence chip complex Musk has envisioned in Texas.

WHAT IS THE TERAFAB PROJECT?
Musk's SpaceX, its xAI unit and Tesla ‌will build two advanced chip factories at a sprawling facility in Austin, one to be used in Tesla vehicles and Optimus humanoid robots, and another designed for AI data centers in space.

"We either build the Terafab or we don't have the chips," Musk had said during a presentation in Austin in March, adding that current global chip production would meet only a small fraction of his companies' future needs.

Musk said he was grateful to existing chip suppliers, naming Samsung (005930.KS), opens new tab, TSMC (2330.TW), opens new tab and Micron (MU.O), opens new tab, but said demand from his companies would eventually exceed total global chip output.

He did not give a timeline for the project and has a track record ⁠of announcing highly ambitious projects, though several have faced delays or fallen away. SpaceX said in the filing there is no assurance it will meet its Terafab objectives within expected timelines, or at all.

Intel announced it would join the project in April, bringing established chip manufacturing expertise to the venture.

LOCATION AND SCALE
Musk has said the Terafab will handle each step of chip production, including the design. The facility is planned in Grimes County within a newly designated reinvestment zone, where local officials are expected to consider a property tax abatement agreement at a June meeting.

On Tesla's earnings call in April, Musk said that the details of the Terafab deployment are still being worked out.
In the near term, Tesla will be building the research fab on its Giga Texas campus in the Austin area. The initiative is expected to cost about $3 billion and "capable of maybe a few thousand wafers per month, but it's really intended to try out ideas," Musk said.

"What we figured out thus far is Tesla doing the research fab, SpaceX doing the initial ‌part of ⁠the large-scale Terafab. And then we got to figure out the rest," he said.

The company estimates total investment for the Texas semiconductor manufacturing facility could rise to $119 billion if additional phases are completed.

Terafab will eventually produce one terawatt of computing capacity a year, compared with about half a terawatt currently generated across the United States, Musk had said in March.

Building enough chip capacity to power one terawatt of annual compute would cost between $5 trillion and $13 trillion in capital expenditure, according to Bernstein estimates.

TECHNOLOGY
Tesla plans to use Intel's 14A manufacturing process to make chips at the ⁠Terafab project. The contract would mark Intel's first major customer for the technology, a breakthrough for the chipmaker which has struggled to stand up its contract manufacturing business essential for taking on top rival TSMC.

Musk said that by the time Terafab scales up, Intel's 14A manufacturing process "will be probably fairly mature or ready for prime time" and "seems like the right move."

Musk's staff have reached out ⁠to chip industry suppliers, including Applied Materials (AMAT.O), opens new tab, Tokyo Electron (8035.T), opens new tab and Lam Research (LRCX.O), opens new tab, and Samsung for the Terafab project, according to Bloomberg.

Staff have sought price quotes and delivery times for an array of chipmaking gear, Bloomberg reported in April, adding that in the past few weeks, they have contacted makers of photomasks, substrates, etchers, depositors, cleaning devices, testers and ⁠other tools.

Reuters reported that SpaceX is planning to make its own graphics processing units, or GPUs, the chips at the heart of training AI models.

THE "UNKNOWNS"
Though Musk has said Terafab would target chips for cars, humanoid robots and space-based data centers, many details are unknown, such as:
Who will pay for pricey chipmaking equipment
Who will operate the factory
When it will come online


 
Great news for intel. I'm still waiting for the SpaceX investment in Intel. Maybe after the trillion dollar IPO?

Intel 14A is HNA-EUV so this will be interesting. Since this is an IDM 3.0 type of deal maybe margins will not be a problem?
 
Great news for intel. I'm still waiting for the SpaceX investment in Intel. Maybe after the trillion dollar IPO?

Intel 14A is HNA-EUV so this will be interesting. Since this is an IDM 3.0 type of deal maybe margins will not be a problem?
How does Intel get revenue from this? we dont have any details so what is your model? Getting Tesla to pay Intel R&D costs would be HUGE benefit to Intel.

They are still getting approval for the site. So this is 2030 production at the earliest... 14A will be mature by then.
 
Last edited:
Great news for intel. I'm still waiting for the SpaceX investment in Intel. Maybe after the trillion dollar IPO?

Intel 14A is HNA-EUV so this will be interesting. Since this is an IDM 3.0 type of deal maybe margins will not be a problem?

Let's come back to this thread in five years! Everything now is for me too much a sales pitch from SpaceX/Tesla in view of the SpaceX IPO and the Chips Act 1 Jan 2027 deadline to benefit from 35% advance checks written to Musk by Uncle Sam/Main Street.

I will never invest in Musk related entities.........too many dead horses left behind his path through life......I just have to think back to his DOGE disaster and promises he made. Many people will need to clean up the mess he leaves behind. Easy to tear down stuff when you have no clue how a modern society works.

Musk seems to only care about his legacy in view of the history books, and seems to think that every person who does not follow him or even questions his "great entrepreneurial visions" is plain stupid and not worth his time and energy. He simply consumes too much oxigen......

In five years we can look back........Just my two cents.......
 
How does Intel get revenue from this? we dont have any details so what is your model? Getting Tesla to pay Intel R&D costs would be HUGE benefit to Intel.

They are still getting approval for the site. So this is 2030 production at the earliest... 14A will be mature by then.

My guess is this is too early to tell -- Musk has identified a "medium-term" need (more RAM, more chips, latest foundry node). He has somewhat limited choices - Full DIY, Full outsource to one or more of the Foundries, or use legacy nodes that are available.

He's decided to do a hybrid of DIY and partnering, with using Foundry capacity too (Samsung deal in Texas, and continuing to use TSMC for AI4, AI5, etc). I think the depedancies are too great to determine a real model:

- Optimus (robot) still a ways until production (i.e. capabilities and true demand 99.9% unknown)
- Future AI demand and technical requirements still unclear except for 'need to stay ahead of competitors'
- Cost effective AI datacenters still depend on Starship v4 or v5 (v3 hasn't launched yet)
- Car self-driivng demand (it's somewhat known on customer vehicles, but unknown how well the uptake actually will be for Robotaxi/Cybercab)

My guess is they'll take this year to break some ground, get some work started, and actually sketch out the long term plan. They'll (hopefully?) hammer out some technical details with Intel on the fab side, while working internally to align their internal roadmap to be agnostic between Terafab and outsourced Foundry businesss (fall back).

.. so maybe end of year we'll have a "path forward", and then another year before we can really begin guessing with >5% accuracy on volume details.

I'd appreciate other's thoughts here..
 
I'd appreciate other's thoughts here..

Today Tesla is using TSMC N3. Tesla signed a $16.5B deal for Samsung 2nm that runs through 2033. And now SpaceX is signing a deal with Intel for 14A.

Elon Musk should know by now that foundries build fabs based on customer commitments, not customer dreams.

Elon Musk should also know about the risk of delay or outright failure of process technologies. In order for Samsung and Intel to stay competitive with TSMC they must go the high risk route (HNA EUV) and blaze new trails.

Generally how companies choose fabs is based on the readiness of the PDKs. For big companies like Apple, Nvidia, QCOM, and AMD they help develop the PDKs so the risk of it not working is greatly reduced. To get into that TSMC inner circle you must have a heavy wafer agreement that will cover the costs of collaboration. Not only will TSMC have to build those fabs, they also need to keep them filled for many years to remain profitable. For example, TSMC has been moving fab capacity around a bit. I think we will see more of that in the future. Moving 7nm capacity to 5nm and 5nm to 3nm etc...
 
Back
Top