You’ve probably seen the headline numbers doing the rounds: 100 million square feet, $16.8 billion, the biggest factory on Earth. Terafab, Elon Musk’s Texas chip factory, arrived in March 2026 as a joint Tesla, SpaceX and xAI venture, with Intel joining in April 2026 as the manufacturing partner. Musk framed it as a supply necessity: future chip demand across his companies would outstrip total global output.
The money trail is the first thing worth testing. The figures have shifted more than once, and none of them means what the headline numbers imply on their own.
Here are three tests to work through in order: what the first phase actually funds, what the footprint really measures, and what one terawatt of compute would take. By the end you’ll be able to read every Terafab headline against actual fab economics. For the full Terafab overview, start there.
What is Elon Musk’s Terafab, and what is the $16.8 billion first phase actually paying for?
Terafab is a 100-million-square-foot semiconductor campus announced by Elon Musk in March 2026 in Grimes County, Texas, a Tesla, SpaceX and xAI venture partnered with Intel. The $16.8 billion first phase funds the site, shell, cleanrooms, early tooling and utilities, against a possible $119 billion full buildout.
The money trail has shifted along the way. An early figure sat around $20 to $25 billion; a May 2026 filing then put the initial phase at $55 billion and the full buildout at up to $119 billion, before SpaceX and Tesla formally committed the initial $16.8 billion in August 2026.
What the first phase actually buys is smaller than the total. It covers land acquisition, the building shell, cleanroom construction, early lithography and process tooling, power and water utilities, and the setup of the Intel partnership. Process equipment typically runs 70 to 80 percent of a modern fab’s total cost, so $16.8 billion funds the foundation, with the full tooled factory still to come.
Tesla, SpaceX and xAI are both the backers and the buyers. Tesla wants inference processors for Optimus robots and Cybercab robotaxis, SpaceX wants high-power silicon for Starlink and space-based data centres, and xAI wants model-training compute. That captive pipeline is the point, and it is why the space-grade chip wedge matters.
Intel’s role is a partnership: Intel brings its 14A process technology and decades of fab-operations experience, and Tesla plans to use 14A for chips at Terafab, which would mark Intel’s first major customer for the node.
The site is part of the story too. Grimes County has about 30,000 residents, and locals raised transparency and environmental objections at a public hearing. Terafab will draw water from Gibbons Creek Reservoir rather than local groundwater, and Texas sweetened the deal with a $30 million Texas Enterprise Fund grant and JETI program qualification. Consolidating logic, memory, packaging and testing under one roof is the single-roof vertical integration bet made concrete.
Before the output, though, there’s the building itself, and the “biggest factory on Earth” label attached to it.
Why is Terafab being called the biggest factory on Earth, and how does it compare to Boeing’s Everett factory and the Pentagon?
Terafab’s proposed 100 million square feet of manufacturing space is roughly 15 times the Pentagon’s 6.6-million-square-foot floor area and about 23 times Boeing’s Everett factory at 4.28 million square feet. But “manufacturing space” is a much broader category than cleanroom space, and total footprint is the least meaningful metric for actual wafer output.
The 100 million square feet works out to about 9.3 million square metres across four buildings, and it is more than three times Samsung’s entire Pyeongtaek campus.
The catch is in the wording: “manufacturing space” covers far more than cleanroom area. In a chip site, the cleanroom typically accounts for around 20 to 25 percent of total floor space, with the sub-fab and mechanical systems taking much of the rest. An advanced fab might need 300,000 to 500,000 square feet of cleanroom. A bigger building does not automatically mean more chips.
That square footage still has to be powered, cooled, staffed and supplied, and those requirements sit far beyond the $16.8 billion first phase. Cleanroom area and wafer capacity determine output; total footprint does not. For how each piece fits together, see the wider chip-manufacturing story.
The footprint question only sharpens once you get to the output target.
Why does Musk’s 1-terawatt compute target make Terafab look unrealistic, and how many fabs would it need?
Musk’s one-terawatt target sits far beyond a single campus. It implies roughly a billion high-power chips a year, many multiples of what the leading foundry produces across its entire network, and would take hundreds of fabs and trillions in capex to reach.
One terawatt of computing capacity produced per year is a hard number to hold in your head. It means roughly one billion full-reticle chip equivalents annually, each at about one kilowatt of compute, which is about double the half-terawatt the entire United States currently generates. No single facility produces anything like that today.
For a baseline, look at TSMC. The leading foundry shipped 15.023 million 300mm-equivalent wafers in 2025, its biggest year ever, and that figure includes millions of legacy-node wafers. Terafab’s own stated path runs from 100,000 wafer starts per month to 1 million at full capacity, which would be about 80 percent of TSMC’s entire current global output. That is still short of one terawatt.
Bernstein’s estimate for closing the gap runs to hundreds of leading-edge fabs, at a capital cost of $5 to $13 trillion. For comparison, a single 2nm-class fab doing 50,000 wafer starts per month costs roughly $28 billion and takes about 38 months to build in the US. On the construction side, basic industrial shells run about $150 to $300 per square foot, while ISO 1-3 cleanrooms can run $1,500 to $2,500 per square foot.
Equipment adds another constraint. ASML shipped just 48 EUV systems in 2025, and its order book is fully allocated to TSMC, Samsung and Intel through 2027, which leaves little room for a greenfield campus to tool up quickly. For the full treatment, see whether the one terawatt target is realistic.
Put those three tests together and a different picture emerges. Step back, and the clearest way to read Terafab is as a captive-demand supply chain built around a single campus. The $16.8 billion first phase funds the site, the shell and the early tooling, with the full fab still to come. The headline footprint tells you little about wafer output. And the one-terawatt target describes an ambition with no binding plan behind it. The supply-chain logic is the part worth paying attention to.
The practical test is whether captive demand plus Intel’s 14A process and fab-operations expertise can justify the single-roof model. Square footage and terawatts do not answer that. SpaceX’s own May 2026 filing described Terafab as a “general framework” with no binding commitments. The wider chip-manufacturing story joins the pieces together, and whether one terawatt can ever be met is covered in whether the one terawatt target is realistic.
Frequently Asked Questions
Who actually owns Terafab, and is it a Tesla project?
Terafab is a joint Tesla, SpaceX and xAI venture, not a Tesla subsidiary. The three Musk companies pool their chip demand into one campus, with Intel joining in April 2026 as the manufacturing partner rather than an owner. That structure matters because it makes Terafab a captive-demand supply chain, where the customers are also the backers, rather than a conventional foundry selling to the open market.
When will Terafab start producing chips, and when could the first phase open?
No firm production date has been published. The project was announced in March 2026, the $16.8 billion first phase followed in August 2026, and a full buildout could run for years across multiple phases. A single modern fab typically takes about 38 months from construction start to first output, so even the initial wafer starts are unlikely before the end of the decade, assuming the shell, tooling and utilities come together on schedule.
Will Terafab replace TSMC, or is it competing with the leading foundry?
Not in the near term. Terafab is built around captive demand from Tesla, SpaceX and xAI, not around winning orders away from TSMC. Musk’s stated concern is that global chip supply cannot keep pace with his companies’ needs, so Terafab is an attempt to secure capacity rather than out-compete the foundry industry. On the numbers, TSMC shipped 15 million wafers in 2025, while Terafab’s first phase is still a shell.
What will Terafab actually make first: AI chips, robot processors, or something else?
The first products have not been locked in publicly, but the demand pipeline points to three groups: Tesla’s Optimus robot and Cybercab processors, SpaceX’s Starlink and orbital data-centre silicon, and xAI’s model-training compute. Intel’s 14A process would be the manufacturing foundation. The single-roof plan also folds in logic, memory, packaging and testing, so the initial output is likely a mix rather than one flagship chip.
Why is it called Terafab?
The name signals scale rather than a specific technology. “Tera” means a trillion, and “fab” is short for fabrication plant, so the branding points toward the one-terawatt ambition Musk has set for yearly compute production. It follows the same convention as Tesla’s Gigafactories, where the prefix advertises the intended magnitude before the capacity actually exists.
Is one terawatt of computing power even possible to run in Texas?
In energy terms it is an extraordinary ask. One terawatt is about double the half-terawatt currently generated across the entire United States, so powering that much compute would need a grid-scale buildout of generation, transmission and cooling, not just a bigger building. The target is best read as an ambition, because the physical infrastructure to feed a terawatt does not exist in Texas or anywhere else today.
What does “captive demand” actually mean for how Terafab works?
It means the customers are built in. Tesla, SpaceX and xAI are both the backers and the buyers, so Terafab’s output is spoken for before a wafer is produced. That inverts the normal foundry model, where a manufacturer chases orders from outside customers. The advantage is guaranteed demand; the risk is that the whole bet depends on Musk’s companies growing into the enormous capacity the campus is designed to hold.
What is the difference between a fab and a foundry?
A fab is a fabrication plant, the physical facility where chips are made. A foundry is a company that manufactures chips for other firms, like TSMC, rather than designing its own. Terafab blurs the two: it is a fab campus that serves a captive group of customers through Intel’s process technology, which is closer to an internal supply chain than a merchant foundry selling capacity on the open market.
What does Intel’s 14A process technology actually mean?
14A is Intel’s advanced process technology, essentially the manufacturing recipe that determines how small and efficient the transistors on a chip can be. Intel brings both that recipe and decades of fab-operations experience to Terafab. The arrangement is a partnership rather than a simple licence, which makes Terafab Intel Foundry’s first major anchor customer for producing chips at leading-edge scale for Musk’s companies.
What tax breaks and incentives did Terafab qualify for in Texas?
Terafab qualified for a $30 million Texas Enterprise Fund grant and the state’s JETI programme, which offers tax breaks for large job-creating projects. Those incentives are modest next to the $16.8 billion first phase, so they are best read as a site-selection sweetener rather than a reason the project exists. The bigger local draw was practical: Gibbons Creek Reservoir water, available land and a workforce pool across Grimes and Brazos counties.
