Tesla says its factory footprint exceeds 30 million square feet; adding the sites up individually gets to roughly 34 million. SpaceX's manufacturing floor — Starbase, Hawthorne, McGregor, Redmond, Bastrop — is on our estimate around 5 million, a fraction of Tesla's, which is the first thing worth saying out loud about these two companies.
Then there is Terafab: the vertically integrated semiconductor campus Tesla and SpaceX are building together in Texas — logic, memory, packaging and test under one roof, aimed at AI silicon for Optimus, Cybercab and SpaceX's orbital data centres, with Intel involved. Stated full build-out is more than 100 million square feet. Initial phase investment is $16.8 billion; filings and earlier references run the multi-phase total toward and past $119 billion. Power draw at scale is quoted in gigawatts. The first phase alone is at least 3,000 jobs in Grimes and Brazos counties.
One announced campus, roughly two and a half times everything both companies have physically built since 2002.
The number is easy. The question underneath it is not: Musk's implicit claim is that building complex things at scale has become a repeatable, transferable skill inside his companies. Terafab is that claim's largest test. This piece measures the claim rather than repeating it, and the measurement needs a common yardstick — because "under construction" covers everything from a signed incentive agreement to a roof going on.
Five stages of a production building
Every site below is placed on the same ladder, and the second chart shows where each one currently sits:
| Stage | What it means | What you can verify |
|---|---|---|
| 1 — Permits & land | Land assembled, incentives and utility agreements signed. Nothing physical. | Filings, county and school-district agreements |
| 2 — Foundations | Clearing, grading, piles, footings, grade beams. | Aerial imagery; concrete pours |
| 3 — Structure | Steel, envelope, roof, cranes installed. | Steel count, roof trusses |
| 4 — Tooling | Lines, cleanrooms, tools installed and commissioned. | Tool deliveries, hiring, commissioning notices |
| 5 — Producing | First units off the line. Says nothing yet about rate. | Company announcements, first-unit photos |
The reason to be strict about this is that stages 1–3 are the ones Musk's companies are demonstrably world-class at, and stages 4–5 are where every schedule has historically slipped. Collapsing them into "building a factory" is what makes the announcements sound the same as the deliveries.
Tesla: where each building stands
| Building | Approx. floor space | Stage | Evidence |
|---|---|---|---|
| Giga Texas, vehicle plant | >10M sq ft | 5 | Model Y, Cybertruck, Cybercab in production; ~16,500 employees; ~2,500-acre campus |
| Giga Nevada + Semi plant | ~5.4M + 1.7M | 5 | Semi reached volume production ~April–May 2026; "many thousands" targeted in 2026, 50k/yr long term |
| Fremont | ~5.5M | 5 vehicles / 4 Optimus | Model 3/Y running; first-gen Optimus lines installing where S/X lines were; start expected later in 2026 |
| Giga Shanghai | ~4.5M | 5 | >950k units/yr installed capacity; export hub |
| Giga Berlin | ~3M | 5 | Step-up to ~7,500 vehicles/week (~390k annualised) targeted by October 2026; cells toward ~18 GWh |
| Megafactory Texas (Brookshire) | ~1.65M | 5 | Megapack 3 rolling off the line since early August; 50 GWh designed; ~1,500 jobs |
| Megafactory Lathrop / Shanghai | ~0.44M / — | 5 | ~40 GWh and ~20 GWh Megapack |
| Giga Texas, Optimus factory | ~1.5M planned | 3 | Multi-storey steel rising fast, several cranes, roughly a quarter of planned length; 2027 orientation |
| Giga Texas, precursor chip fab | — | 2 | Footings poured, perimeter grade beams forming |
| Project Crystal Sun (Fort Bend) | ~3,000 acres | 1 | ~$10.1B filed, ingot-to-module, construction 2026–2028, operations ~Q1 2029, ~9,700 jobs |
| Mexico | — | 1, paused | No change |
SpaceX: a different machine, pointed at a different problem
SpaceX's independent build-out is not a copy of Tesla's. Tesla is adding product categories; SpaceX is adding rate to one vehicle programme and holding two very high-volume consumer-scale lines beside it.
| Building | Approx. floor space | Stage | What it does |
|---|---|---|---|
| Starfactory, Starbase | ~1M sq ft | 5 | Starship and Super Heavy components from steel coil — barrels, nosecones, flaps, tanks. Replaced the tent era |
| Mega Bay 1 & 2, Starbase | included above | 5 | Vertical stacking, engine install, outfitting |
| Gigabay, Starbase | ~700k sq ft | 3 | Structural steel essentially complete, roof trusses finishing mid-August — first column set roughly ten months earlier. ~24 work cells, cranes to ~400 tons |
| Hawthorne, CA | ~1M sq ft | 5 | Falcon 9/Heavy stages, Dragon, Merlin, much Raptor production, Starlink terminals |
| McGregor, TX | — | 5 | Raptor test and production. Raptor 3 serials into the 200s; 800–1,000 engines/yr referenced |
| Redmond, WA | ~0.5M sq ft | 5 | All Starlink satellites including V3 — ~70 satellites a week, ~3,640/yr, expanding |
| Bastrop, TX | >1M sq ft | 5 | User terminals at ~15,000 a day; expanding into more vertical integration |
| Starbase Pad 1 | — | rebuild | Trench, new mount, chopstick and diverter upgrades; return to service targeted 2027 |
| Starbase Pad 2 + ASU | — | 5 | Active for recent flights; tank farm expanding; on-site air separation for local LOX/LN2 |
| KSC LC-39A / SLC-37 | — | 2–3 | Starship infrastructure on the Space Coast; 39A chopstick load testing and deluge, 39A operational readiness possibly still in 2026 |
Right now Starbase has multiple vehicles in flow — ships in the 41/42 range, boosters 21/22 and successors — with Flight 14 in a possible late-August window and testing continuing at Massey's.
The two companies are solving opposite problems
| Tesla | SpaceX | |
|---|---|---|
| Unit of output | Vehicles, packs, and now robots and chips | Rockets, engines, satellites, terminals |
| Direction of the build | Wider — four industries at once | Faster — one vehicle programme to high rate, plus two consumer-scale lines |
| Highest-rate line today | ~950k vehicles/yr installed at Shanghai | 15,000 terminals a day at Bastrop; ~70 satellites a week at Redmond |
| Newest big building | Megafactory Texas — 16 months to first Megapack 3 | Gigabay — ~10 months from first column to structural steel complete |
| What gates the next step | Ramp of genuinely new products: Optimus, Cybercab deployment, Semi rate | Reuse and cadence: Pad 1 back in 2027, Gigabay fit-out, engine supply |
| Where the floor space is going | Robots and chips | Starship integration and refurbishment capacity |
The Bastrop line is worth pausing on, because it is the least-discussed number in either company: 15,000 terminals a day is roughly 5.5 million units a year, an order of magnitude more units than Tesla builds vehicles. Whatever "can Musk's companies do high-volume manufacturing" means, that line already answers it — for a product with a fraction of a car's complexity.
The claim, measured: ground to first unit
If "scaling complex production is repeatable" means anything checkable, the interval from breaking ground to shipping the first unit should be short and getting shorter. On the public record:
| Facility | Ground broken | First production | Months |
|---|---|---|---|
| Giga Nevada (cells) | mid-2014 | early 2017 at volume | ~30 |
| Giga Shanghai | January 2019 | December 2019 | ~11 |
| Giga Berlin | early 2020 | March 2022 | ~25 |
| Giga Texas | July 2020 | April 2022 | ~21 |
| Starfactory, Starbase | 2023 | 2024 | ~15 |
| Megafactory Shanghai | May 2024 | February 2025 | ~9 |
| Semi plant, Nevada | ~early 2024 | April–May 2026 | ~28 |
| Megafactory Texas | ~April 2025 | August 2026 | 16 |
| Gigabay, Starbase | ~October 2025 | steel complete Aug 2026 (stage 3) | ~10 to structure |
Dates are public record and company statements rather than filings, and the slow European and Nevada rows were paced substantially by permitting and product readiness rather than by construction. Read the pattern, not the row: the most recent purpose-built plants are the fastest, and stages 1–3 now run on a nine-to-sixteen-month clock at both companies. Shanghai's eleven months in 2019 was treated as an anomaly. It is now roughly the house standard.
That skill is real and badly underrated. Most industrial companies cannot do it once.
What does not repeat
Apply the same test to products rather than buildings:
| Product | Unveiled | Volume production | Years |
|---|---|---|---|
| Model 3 | March 2016 | mid-2018, after "production hell" | ~2.3 |
| Semi | November 2017 | April–May 2026 | ~8.5 |
| Cybertruck | November 2019 | 2024 ramp | ~4.5 |
| Optimus | 2021–2022 concept | Fremont lines installing; start expected late 2026 | ~5 |
| Cybercab | October 2024 | first unit Feb 2026, continuous April 2026 | ~1.4 |
| Megapack 3 | 2025 | August 2026 | ~1 |
The two fast rows are the two derivative products — a pack on an existing pack line, a vehicle on an existing autonomy stack in an existing plant. Everything genuinely new took two and a half to eight and a half years.
SpaceX's version of the same gap is cadence rather than first units. Starship flies, and the programme's stated ambition runs to daily launches by the end of 2027; the pad that most recent flights depend on is a rebuild returning in 2027, and Gigabay — the building that unlocks the throughput — is at stage 3. The vehicle exists. The rate does not yet.
So the honest form of the thesis is narrower than the marketing version and much stronger than the dismissal:
What Musk has industrialised is stages 1 through 3 — permits, dirt, steel — and it now runs in parallel across twenty-odd buildings and four industries. What has not become repeatable is stage 4 and what follows it: tooling a genuinely new product and ramping it to rate. We put numbers on the far end of that this morning: on our base case the Cybercab ends 2026 with about six cars built for every one carrying a passenger. Building them was never the constraint.
Why Terafab is the hard case
Terafab is neither a shed nor a derivative.
A vehicle plant is a very large building with a very good line in it, and both companies are exceptional at that. A leading-edge fab is a different physical object: cleanroom class, vibration and contamination budgets, water and power at multi-gigawatt scale, and a tool set — lithography above all — with lead times in years and one credible supplier for the critical step. Its learning curve is yield, and yield does not respond to sleeping on the factory floor. Intel's involvement is the tell: it is the part of this that Tesla is not claiming it can teach itself.
Note where Terafab sits on the ladder. The precursor fab at Giga Texas is at stage 2 — footings and grade beams. The Grimes County campus, on the old Gibbons Creek site with its reservoir doing the water math, is at stage 1 going on 2: thousands of acres controlled, incentive and school-district agreements active, clearing visible in places, civil works expected within months. Every square foot of that 100 million is stage 1 or 2. That is not a criticism — it is a date-check on anyone treating the number as capacity.
The money sits on two cash-flow statements already strained by their own build-outs. Tesla guides above $25 billion of capex this year; its Q2 2026 Rule of 40 score was 21.65 with the free-cash-flow half already negative on a $5,789M capex quarter. SpaceX spent $18,369M of capex in a single quarter and does not publish the quarterly cash line that would let anyone score it at all. A $16.8B first phase is affordable across those two. The path to $119B is a financing structure nobody has described, and that absence is the most substantive open question about the project.
Almost everything in the "known and physical" column above is visible because independent observers fly drones over it — @JoeTegtmeyer at Giga Texas and Starbase, NASASpaceflight and LabPadre at the pads. Weight that source correctly: excellent for "did concrete get poured this week", useless for "what will 2031 output be."
What to watch: the next stage transition, per building
- Giga Texas precursor fab, 2 → 3 → 4. Steel proves construction, which is not in doubt. Tool move-in is the first evidence that either company can execute stage 4 in semiconductors.
- Fremont Optimus, 4 → 5. The nearest-term new-product ramp, on a machine with roughly 10,000 unique parts. A slip here is a slip in exactly the skill Terafab needs.
- Gigabay, 3 → 4. Fit-out and the first vehicle into a work cell is what converts Starbase's building rate into a flight rate.
- Semi's run rate against "many thousands" in 2026. The plant is done; this is pure ramp, and the cleanest scoreboard of the two skills separated.
- A financing structure for Terafab beyond phase one, and whether it lands on Tesla's balance sheet, SpaceX's, or a vehicle outside both.
The past twelve months show companies that can put up a 50 GWh factory in sixteen months, frame a 700,000 sq ft integration hall in ten, and start a purpose-built robotaxi line in a quarter of the time a new vehicle used to take. That is the argument at full strength and it deserves to be stated that way. Terafab asks whether the machine still works when the constraint stops being steel and starts being physics — and on that question, the evidence does not exist yet in either direction.
