テックトレンド

Is Terafab Really the Largest Building Ever? Running the Numbers on 100 Million Square Feet

Published2026-08-08Ryuta Hamamoto

On August 6, 2026, SpaceX and Tesla announced Terafab, a semiconductor plant in Grimes County, Texas. One hundred million square feet, $16.8 billion in phase one alone. I check the Texas governor's official release and Musk's own post, then actually do the arithmetic on the claim that this will be the largest building ever. Includes the official concept film.

Is Terafab Really the Largest Building Ever? Running the Numbers on 100 Million Square Feet
Share

Hello, this is Ryuta Hamamoto from TIMEWELL.

Watch this first. It is the concept film SpaceX released for Terafab.

For the record, that is CGI, not aerial footage of something already standing1. Over 53 seconds a Tesla drives toward a glowing TERAFAB sign, passes through a structure with pedestrian levels, and then the camera pulls up until a symmetrical complex reaching the horizon comes into view, ending above the clouds. It is well made. And precisely because it is well made, I wanted to check the numbers myself.

That is what this article does. How far does the "largest building ever" claim actually hold? I went to the official release and to Musk's own post, and did the arithmetic.

What was actually announced

Start with the record. On August 6, 2026, the office of the Texas governor issued an official announcement2. SpaceX will build a vertically integrated semiconductor manufacturing facility in Grimes County. The name is Terafab. Phase one is a capital investment of more than $16.8 billion, creating 3,000 new jobs spanning engineers, technicians and plant operator personnel.

State support is specific too. A $30 million Texas Enterprise Fund grant has been extended, and the facility is a qualified project under the state's JETI program, Texas Jobs, Energy, Technology, and Innovation2. This is a deal the state paid to attract.

Then the scale. The governor's release itself describes a "100-million-square-foot facility"2. That is the number that matters most here, and it is worth stressing that it appears in an official document rather than only in secondhand reporting.

Musk's comment is carried in the same release, closing on producing AI chips at scale for use on Earth and in space2. The stated purpose is to bring logic, memory, packaging and testing into one location, making chips for Tesla's Optimus and Cybercab and for SpaceX's space-based data centers3. I wrote about that orbital compute concept separately in SpaceX's orbital data center plan. Terafab is the part that makes the components for it.

Phase one is $16.8 billion, and reporting puts the potential total across phases at around $119 billion3. The state release only carries the phase one figure and does not say how many phases there are in total, so that number belongs in the reported-not-official column.

Take AI-driven development all the way to production

WARP is a hands-on program for teams who want more than headlines. Former enterprise DX and data strategy leads work alongside you until it runs.

Putting "50 times the Pentagon" on a calculator

Now the arithmetic. Musk posted on X that Terafab will be 50 times the size of the Pentagon when complete4. That line is being quoted widely, so it is the first thing to check.

The Pentagon has long been described as one of the world's largest office buildings, with roughly 6.5 million square feet of floor space. Sources put it anywhere from 6.5 to 6.64 million5, but the conclusion holds whichever you pick. Divide Terafab's published 100 million square feet by that and you get about 15.4 times. Not 50. Going the other way, 50x on floor area would need 325 million square feet, more than three times the official figure.

So it is one of two things. Either he is describing a future state including later phases, or he means something other than floor area, site acreage for instance. The published material does not let me decide which. The honest way to write it is that Musk said this, and it does not reconcile with the published floor area. The bigger the number, the more the person quoting it owes everyone one division.

What about the headline claim itself? There, the arithmetic supports it.

One hundred million square feet is about 9.29 million square meters. The cleanest comparison is Boeing's Everett factory, which Guinness World Records certifies as the most capacious building on Earth: 13.4 million cubic meters of volume and an indoor floor area of 39.8 hectares, about 398,000 square meters5. Terafab would be roughly 23 times that floor area. Against Tesla's own Gigafactory Texas, reporting puts it at about 10 times3. "A different order of magnitude" is not hyperbole in this case.

A building and a structure are two different claims

A word on terminology. I have seen this described as the largest structure in human history. I understand the impulse, but it is not accurate.

What the published figures support is the largest building by floor area. Widen the frame to structures and you are up against the Great Wall, large dams, airport complexes. On measures like length, volume or footprint, plenty of things are bigger than Terafab. "Largest building ever" and "largest structure in human history" are not paraphrases of each other. They are different claims.

This may sound pedantic, but I care about it, because numbers like these start walking on their own the moment someone drops them into an internal deck. Trace "50 times the Pentagon" and "largest ever" back to source and their evidentiary weight is completely different. The first is a direct quote that does not match the published figure. The second is a definitional slip. Nor is the answer to dismiss it as hype: state the measure and it is impressive on the merits. Adding the yardstick makes the claim accurate without making it any less striking.

Whether an organization habitually states the yardstick is a decent proxy for how well it converts technology news into decisions. The AI literacy assessment is a quick way to see where yours sits.

Why build it themselves

Enough about size. Why do this at all?

Semiconductors have long run on a division of labor between companies that design and companies that fabricate. Fabless and foundry. That split held because fabrication equipment is extraordinarily expensive and only pencils out when demand from many customers is pooled. This announcement is a decision to move to the building side of that line.

The reasoning is not hard to follow. When AI chip supply is tight, depending on outside capacity means you cannot secure the volume you need when you need it. Producing Optimus at volume. Running Cybercabs. Putting data centers in orbit. Every one of those makes "how many chips" a precondition of the business plan, and this is a decision not to leave that precondition inside someone else's allocation. I think the logic holds. It is consistent with how SpaceX integrated vertically from launch through communications, which I traced in SpaceX's IPO and vertical integration.

The siting is worth a look too. Reporting places the build on the site of the former Gibbons Creek coal-fired power plant3. Fabs consume enormous amounts of electricity, and advanced AI-class processes more so. Taking over a decommissioned power plant already tied into the grid reads less like acquiring land and more like acquiring power. Floor area gets all the attention when people talk about factory scale, but in practice the binding constraints are usually electricity and water. Securing that first deserves credit.

Logic holding and execution landing are different things, though. Ramping semiconductor volume production is decided more by people and yield than by capital. Three thousand jobs looks conservative against a facility of this size. I would guess heavy automation is assumed, but that is not something the published material tells you. Nor does the state release give a production start date.

Honestly, that is the number I want. Not 100 million square feet. The date the first wafer comes out. A factory produces nothing on the day it is finished.

Three things worth watching

First, whether the AI chip procurement structure shifts. If vertical integration works, SpaceX and Tesla take their own demand out of the merchant market. That loosens supply overall, and at the same time concentrates more advanced manufacturing know-how in one company. For buyers elsewhere, the question is whether the menu of suppliers gets longer or shorter.

Second, that space and AI keep converging inside one balance sheet. The orbital data centers here, the unlicensed spectrum opening for satellite direct-to-device I wrote about the other day, and Terafab are all happening inside the same capital structure. Read as separate stories you miss it. Put together, the shape is one company holding compute and connectivity, on the ground and in orbit, vertically. That is an economic security question as much as a technology one.

Third, and most practical: when a big number shows up, check the yardstick. One division was enough to show "50 times the Pentagon" does not match the published figure. This is not about catching anyone out. It is about not copying a number into an internal deck and regretting it later. Trace the source, align the units, do the division once yourself. The quality of what circulates internally changes immediately.

Hold the order of magnitude, wait for the wafer

To recap. On August 6, 2026 the Texas governor's office officially announced SpaceX's semiconductor plant in Grimes County. Terafab, more than $16.8 billion in phase one, 3,000 jobs, a $30 million state grant. The scale stated in the official release is 100 million square feet of floor area, about 9.29 million square meters, more than twenty times the floor area of the building certified as the most capacious on Earth. The "largest building ever" claim holds once you name floor area as the measure. "Fifty times the Pentagon" does not reconcile with the published figure, and "largest structure in human history" is the wrong word.

The film at the top is CGI, a rendering of the finished state. So two things are required: don't watch it and think this already exists, and don't dismiss it as just a render either. A state that has committed a $30 million grant and qualified the project under JETI has made this an administratively real plan.

Personally I take the absence of a start date as a good sign. Committing to an aggressive date on a fab rarely ends well. Hold the order of magnitude and wait quietly for the first wafer. That feels like the right distance to stand at.

If you want help turning technology news like this into investment and planning decisions, that is the work we do through WARP. For specific questions, get in touch.


Footnotes

  1. The concept film embedded at the top was released by SpaceX and is a CGI rendering of the completed facility, not footage of an existing structure.

  2. Office of the Texas Governor, "Governor Abbott Announces SpaceX Expansion In Grimes County" (August 6, 2026). The vertically integrated semiconductor manufacturing facility in Grimes County, the phase one capital investment of more than $16.8 billion, the 3,000 new jobs (engineers, technicians, and plant operator personnel), the $30 million Texas Enterprise Fund grant, the qualification under the Texas Jobs, Energy, Technology, and Innovation (JETI) program, and the description of a "100-million-square-foot facility" all come from that release, as do Governor Abbott's comment ("Texas is where big ideas grow even bigger…") and Elon Musk's comment, which closes "…enabling us to produce AI chips at scale for use on Earth and in space." https://gov.texas.gov/news/post/governor-abbott-announces-spacex-expansion-in-grimes-county 2 3 4

  3. The potential total of roughly $119 billion across phases, the combination of logic, memory, packaging and testing in one location, the intended output for Tesla's Optimus and Cybercab and SpaceX's space-based data centers, the comparison to Gigafactory Texas at about 10 times, and the siting on the former Gibbons Creek coal-fired power plant about 100 miles from Austin and 20 miles east of College Station all come from press reporting. The state release gives neither a phase count nor a production start date. https://techcrunch.com/2026/08/06/tesla-and-spacex-will-invest-16-8b-to-start-building-terafab-chip-factory-in-texas/ / https://electrek.co/2026/08/06/tesla-spacex-terafab-grimes-county-16-8-billion/ 2 3 4

  4. Elon Musk on X: "Terafab will be 50 times the size of the Pentagon when complete." https://x.com/elonmusk/status/2085530890759749760 / The Pentagon is generally cited at roughly 6.5 million square feet of floor space; dividing the published 100 million square feet by that gives about 15.4x. Reaching 50x on floor area would require about 325 million square feet, which does not reconcile with the published figure. This article treats the gap as possibly referring to later phases or to a measure other than floor area; the published material does not allow a determination.

  5. Building records are per Guinness World Records. The "most capacious building" is Boeing's Everett factory in Washington State, with a total volume of 13.4 million cubic meters (472 million cubic feet) and an indoor floor area of 39.8 hectares (98.3 acres). https://www.guinnessworldrecords.com/world-records/69361-most-capacious-building / Figures for the Pentagon's floor area range from roughly 6.5 to 6.64 million square feet depending on the source; the conclusion in the text (roughly 15x, not 50x) holds across that range. https://www.globalsecurity.org/military/facility/pentagon.htm / One hundred million square feet is converted here as about 9.29 million square meters. 2

How well do you understand AI?

Take our free 5-minute assessment covering 7 areas from AI comprehension to security awareness.

Share this article if you found it useful

Share

Newsletter

Get the latest AI and DX insights delivered weekly

Your email will only be used for newsletter delivery.

Turn AI-driven development into something you can run

WARP is a hands-on program that takes teams from reading about AI to shipping with it. Former enterprise DX and data strategy leads run the sessions.

Related Articles