Terafab Is More Interesting Than Its 100-Million-Square-Foot Headline

Terafab has been getting attention because Elon Musk says it will eventually become the largest building on Earth.

That part is certainly hard to ignore.

The proposed semiconductor complex in Grimes County, Texas, could eventually contain more than 100 million square feet of manufacturing space. Tesla’s entire Gigafactory Texas is around 10 million square feet.

But the size isn’t actually what I find most interesting.

Terafab is an attempt to take an industry built around extraordinary specialization and pull an enormous portion of it back under one roof.

That is a much bigger experiment.

The semiconductor supply chain is usually the opposite of this

Modern chips are rarely the product of one company doing everything.

A company might design a processor, another company fabricates the logic, another supplies memory, other companies manufacture substrates, somebody else performs advanced packaging, and an entire ecosystem supplies lithography systems, deposition equipment, etchers, chemicals, masks and testing equipment.

The specialization is part of what makes modern semiconductor manufacturing possible.

Terafab wants to collapse a surprising amount of that chain.

SpaceX describes the plan as a closed-loop facility covering the design of lithography masks, fabrication of both logic and memory chips, advanced packaging and testing.

That means they don’t just want their own foundry.

They want to be able to design a chip, make it, package it, test it, discover what is wrong, change the design and start another iteration without moving the work through several different companies around the world.

That is very SpaceX.

One of the reasons SpaceX has been able to iterate rockets as aggressively as it does is because so much of the vehicle is manufactured internally. Tesla has followed the same philosophy with batteries, castings, motors, electronics and manufacturing equipment.

Applying that philosophy to semiconductor fabrication is considerably harder.

A rocket factory is complicated.

A leading-edge semiconductor fab is one of the most complicated machines humanity has ever built.

They are starting with a smaller factory in Austin

There are actually two Terafab-related projects happening.

The first is a research fab on the north side of Gigafactory Texas in Austin.

Tesla broke ground on it in April.

Musk has described that facility as roughly a $3 billion research operation capable of a few thousand wafer starts per month. That isn’t intended to provide the enormous production volume promised for Terafab.

It is there to learn.

The idea is to run enough wafers through a process that engineers can determine whether an experimental manufacturing technique actually works reliably rather than merely producing one successful chip in a laboratory.

I think this may end up being one of the most important parts of the entire project.

Before attempting to build a semiconductor factory an order of magnitude larger than anything Tesla has built before, they are building a smaller place where they can learn how to be a semiconductor manufacturer.

That is considerably more believable than simply announcing a 100-million-square-foot fab and pretending the manufacturing expertise appears with the building.

The enormous Terafab is going to Grimes County

The production site is near Gibbons Creek Reservoir in Grimes County, east of Bryan and College Station and northwest of Houston.

This had been uncertain for months.

That changed on August 6 when SpaceX, Tesla and the State of Texas formally announced the location.

The first phase represents more than $16.8 billion in committed investment and is expected to create at least 3,000 permanent jobs.

Earlier tax-incentive filings describe something much larger if every proposed phase is built: as much as $119 billion across four phases, more than 22,000 acres and construction extending into the 2030s.

I would not treat $119 billion as money already committed.

These are long-range development plans and incentive filings. The scope will almost certainly change as engineers discover what actually works, demand changes and the economics of semiconductor manufacturing move around.

Even the first-phase numbers have changed considerably during planning.

What matters right now is that the project has crossed several important lines.

The site has been selected.

The state and local agreements are signed.

The incentive agreements with the school districts are active.

SpaceX has told local officials it is moving into civil work and foundation preparation.

And land clearing around the future site is already underway.

So Terafab isn’t a factory yet.

But it isn’t just a rendering anymore either.

Intel may be one of the most important parts of this

Tesla and SpaceX are not pretending they already know how to operate a leading-edge semiconductor fab.

Intel joined the project in April.

Musk has said Terafab plans to use Intel’s upcoming 14A manufacturing process, while Intel has described its role more broadly as helping Tesla, SpaceX and xAI explore ways to “refactor” semiconductor manufacturing.

That partnership makes this project much more interesting to me.

Intel has been manufacturing semiconductors for more than half a century. It knows things about process control, yield, contamination, lithography, packaging and high-volume production that cannot simply be hired into existence overnight.

Intel 14A itself is also ambitious. It uses Intel’s next generation of RibbonFET gate-all-around transistors and backside power delivery, with Intel currently targeting high-volume manufacturing later in the decade.

That gives us another useful clue about the Terafab timeline.

The buildings may start appearing soon.

A truly scaled leading-edge production operation is still years away.

What would it make?

The Terafab site currently identifies three major chip families.

AI5 is intended for Tesla’s vehicles and Optimus.

AI6 follows it for future generations of Optimus and other Tesla AI systems.

D3 is intended for SpaceX.

The SpaceX side may eventually be the stranger and more interesting part.

SpaceX is developing the idea of putting large amounts of AI compute into orbit. Its argument is that space eventually offers effectively unlimited solar exposure, avoids consuming terrestrial land for data centers and can radiate heat directly into space.

There are enormous engineering problems hiding inside that idea, but it explains why SpaceX suddenly cares so much about owning part of the semiconductor supply chain.

If you intend to manufacture thousands of AI satellites and put enormous amounts of compute into orbit, buying GPUs from the same supply chain as every cloud provider on Earth eventually becomes a constraint.

SpaceX also wants chips specifically optimized for the radiation, thermal conditions, power limitations and reliability requirements of space.

That is where an internal research fab starts making a lot more sense.

The one-terawatt target needs some skepticism

Terafab’s stated goal is to produce more than one terawatt of compute per year.

That number gets repeated frequently because it sounds enormous.

It is enormous.

It is also not a conventional semiconductor manufacturing metric.

Fabs are normally discussed using things like wafer starts per month, process nodes, die size, yield and packaged-device output.

“One terawatt of compute” describes the aggregate power scale of the computing hardware they hope to manufacture rather than directly telling us how many wafers or chips the factory can produce.

Until we know considerably more about the chips, their power consumption, yields and packaging, I don’t think it makes sense to compare that number directly with conventional fab capacity.

It is a goal.

It is not yet production.

There is a very large distance between those two things.

Power and water may be just as difficult as the chips

A semiconductor facility this large also becomes an infrastructure project.

SpaceX says it intends to generate much of Terafab’s electricity itself using natural-gas power generation paired with very large battery systems rather than simply placing the entire load onto ERCOT.

Water is another major reason the Gibbons Creek location was selected.

Chip fabrication uses enormous amounts of highly purified water. SpaceX says it intends to use closed-loop systems that recover and recycle water, with Gibbons Creek Reservoir providing storage primarily from surface water and stormwater rather than relying heavily on local groundwater.

Those promises are going to deserve continued scrutiny as the project develops.

People already living around the site have legitimate concerns about water, traffic, pollution, emergency services and what happens when thousands of acres of rural land suddenly become one of the largest industrial developments in the world.

Some nearby residents are already watching land around their homes being cleared.

It is possible to think Terafab is an extraordinary engineering project while also expecting SpaceX to be a good neighbor.

Those aren’t contradictory ideas.

Terafab will not replace the existing chip industry overnight

There is another important detail buried in SpaceX’s own filings.

SpaceX says it still expects to obtain a significant portion of its compute hardware from third-party suppliers even after Terafab exists.

That is important.

Terafab should not be understood as Tesla and SpaceX immediately replacing TSMC, Samsung, Micron, Nvidia and the rest of the semiconductor industry.

At least initially, this is additional capacity and an internal development loop.

If it works, that balance may shift over time.

But semiconductor supply chains are too complicated to recreate overnight, even with tens of billions of dollars.

This is one I want to watch being built

There are certain factories where seeing the building is part of understanding the technology.

This is one of them.

I would love to see Terafab in person someday.

Not because it is an Elon Musk project, but because actually walking through a semiconductor facility attempting this level of vertical integration would be extraordinary.

I want to see how they move wafers through it.

I want to see how the logic and memory manufacturing are separated.

I want to understand where the advanced packaging happens.

I want to see the cleanrooms, automation, material handling systems, power infrastructure and the machines that make the machines work.

No public Terafab tour program exists yet, and semiconductor fabs understandably tend to be much more restrictive than ordinary factories.

But Tesla does occasionally allow scheduled tours of Gigafactory Texas.

I hope somebody involved with Terafab eventually recognizes the educational value of doing something similar, even if visitors can only look into the cleanroom through glass.

A 100-million-square-foot building would be impressive to visit.

Watching an attempt to rebuild a substantial portion of the semiconductor supply chain inside it would be far more interesting.

That is the part of Terafab I want to follow.