Terafab — Musk Is Building the Largest Factory on Earth to End His Chip Dependency
On August 6, 2026, Tesla and SpaceX jointly confirmed what Elon Musk had been hinting at for months: they are building a semiconductor factory in Grimes County, Texas, called Terafab, with an initial investment of $16.8 billion. The facility will manufacture, package, and test advanced logic and memory chips entirely under one roof — a vertically integrated chip campus designed not for today's compute needs, but for a future where Musk's companies require more computing power than the entire global semiconductor industry currently produces. Musk described Terafab on X as "the largest and most valuable building on Earth by far." At over 100 million square feet of planned manufacturing space — more than ten times the floor area of TSMC's biggest gigafab — the claim is not obviously exaggerated.
The $16.8 billion figure is phase one. SpaceX's own IPO filing, submitted in May, described a multi-phase plan that could reach $55 billion in initial scope and $119 billion if all expansion phases complete. Intel has confirmed it is contributing to the project, helping Tesla and SpaceX "refactor" silicon fabrication technology for the scale Terafab is targeting. Texas Governor Greg Abbott announced a $30 million Texas Enterprise Fund grant to SpaceX alongside the confirmation. The factory is expected to employ at least 3,000 people from Grimes and neighbouring Brazos County. Hundreds of those county residents had already gathered at a public meeting the night before to raise concerns about tax incentives and a perceived lack of transparency. The era of building at this scale without community friction is apparently over everywhere — including Texas.
Why Musk Is Building His Own Chips — The Logic Behind Terafab
The semiconductor industry has a structural problem that no single company has yet solved: the world's most advanced chip manufacturing is concentrated in a handful of facilities operated by TSMC in Taiwan, Samsung in South Korea and Texas, and Intel in the United States and Europe. That concentration creates geopolitical risk, supply bottlenecks, and queue dynamics where the most powerful buyers — Nvidia, Apple, Google, Amazon — command priority access while everyone else waits.
Musk's companies sit in an unusual position. Tesla already designs its own AI chips — the AI5 in current production at Samsung's Taylor, Texas fab, and the next-generation AI6 covered by a $16.5 billion deal with Samsung signed last year. SpaceX is building orbital data centres that will require chips capable of operating in the radiation environment of low Earth orbit. Tesla's Optimus robots need dense, power-efficient AI compute at manufacturing scale. Cybercab robotaxis need automotive-grade AI inference chips. And xAI's Colossus supercluster — now absorbed into SpaceX — is training Grok models that are scaling toward trillions of parameters and require hundreds of thousands of GPUs. Musk's own arithmetic, presented in SpaceX filings, is that his companies will collectively need one terawatt of compute annually — a figure that exceeds current and projected global chip production capacity. Terafab is his answer to that gap.
What Terafab Actually Is — Inside the Plan
The defining characteristic of Terafab is vertical integration. Most chip factories are specialised: a wafer fab does lithography and etching; a packaging facility does assembly and interconnect; a test facility validates finished chips. Different chips go through different facilities, often in different countries, connected by a global logistics network. Terafab is designed to do everything on a single campus — logic fabrication, memory fabrication, advanced packaging, and full testing — eliminating the logistics dependencies and supply chain fragility of the distributed model.
Intel is contributing to the project — the company said it would help Musk's companies "refactor" silicon fab technology for Terafab's target scale. Intel's involvement is significant: the company has been investing heavily in its own next-generation fabrication processes (Intel 18A and Intel 14A) and has the process engineering expertise that Tesla and SpaceX lack as chip designers rather than manufacturers. The exact nature of Intel's contribution — whether it involves licensing process technology, supplying equipment, or co-investing in the facility — has not been publicly detailed. Intel has been characteristically indirect about its exact role.
| Aspect | Terafab Plan | Context |
|---|---|---|
| Location | Grimes County, Texas (outside Houston) | Near Samsung's Taylor fab — Texas semiconductor corridor |
| Phase 1 Cost | $16.8 billion | SpaceX filings suggest $55B initial → $119B total |
| Footprint | 100M+ sq ft planned | 10× TSMC's largest fab — no building this size has ever been built |
| What It Makes | Logic + memory + packaging + testing | Fully vertically integrated — first of its kind at this scale |
| Customers | Tesla Optimus, Cybercab, SpaceX orbital data centres | Internal only initially — no third-party foundry service announced |
| Partner | Intel contributing process technology | Exact scope undisclosed — Intel 18A process rumoured |
| Timeline | No completion date given | Described as a "general framework" — no binding commitments |
The Honest Assessment — What Terafab Is Not
Electrek's coverage cuts through the scale rhetoric with useful precision: "None of this is about Tesla's near-term supply, which is already contracted out." Tesla's AI5 chip is in production at Samsung's Taylor, Texas fab right now. The AI6 chip is covered by the $16.5 billion Samsung deal. TSMC covers the rest of Tesla's silicon needs. Terafab does not change any of that. Today's chips, and tomorrow's chips, will continue coming from existing supplier relationships for years.
Terafab is a bet on a future that does not yet exist — a future where Musk's vision of millions of Optimus robots doing manufacturing work, millions of Cybercab robotaxis doing all urban transportation, and a global mesh of SpaceX orbital data centres doing AI inference from space has all materialised simultaneously and requires compute volumes that no external supplier can provide. Whether that vision is achievable, and on what timeline, is the essential risk factor for the entire Terafab premise. The $16.8 billion phase one investment is real and committed. The $119 billion total is conditional on a future that Musk is uniquely confident about and everyone else is still modelling.
How Terafab Connects to the Musk Empire's Larger Architecture
Terafab cannot be understood in isolation. It is one node in a vertical integration strategy that Musk has been assembling across all of his companies simultaneously. The architecture looks like this: xAI trains Grok on Colossus, which runs on Nvidia GPUs. Grok powers Tesla's Optimus robots, which are manufactured in Tesla's Gigafactories. Tesla designs the AI chips (AI5, AI6) those robots use. Terafab will eventually manufacture those chips domestically, eliminating Samsung and TSMC from the critical path. SpaceX launches Starlink satellites that provide connectivity to Optimus robots in locations without ground infrastructure. SpaceX orbital data centres — powered by Terafab chips — run AI inference in space.
Every company in this diagram depends on the others. The compute Grok needs comes from chips Terafab will build. The robots that use Grok are manufactured in factories that Terafab chips will help design and optimise. The orbital data centres that run Grok are carried to orbit by Starship. It is the most ambitious vertical integration programme in the history of consumer technology — and Terafab is the missing link that closes the loop on domestic chip supply.
Why This Matters Beyond Musk
Even if Terafab only half-delivers on its ambitions, it matters to the entire semiconductor industry. The United States currently has no domestic facility capable of producing the most advanced logic chips at scale — TSMC's Arizona fab is producing them, but TSMC is a Taiwanese company building on American soil. A fully American-owned, American-operated advanced chip manufacturing facility — the thing the CHIPS Act was written to incentivise — has not materialised from the government programme the way its architects hoped. Terafab, backed by private capital rather than government mandates, may be the closest the US comes to that goal in this decade. The irony that the most serious attempt at domestic chip sovereignty is being led by Elon Musk rather than an established semiconductor company is not lost on anyone watching the industry closely.
Terafab is real, confirmed, and enormous. The $16.8 billion first-phase investment is committed. Grimes County, Texas will be home to what Musk calls the largest and most valuable building on Earth. Intel is involved. The State of Texas is subsidising it. And the facility, when built, will produce the logic and memory chips that Optimus robots, Cybercabs, and SpaceX orbital data centres need. The caveats are real too: no timeline, no binding commitments in the IPO filings, near-term chips still coming from Samsung and TSMC, and a community already pushing back. Terafab is a bet on a future that Musk believes in more than anyone else does. Whether that confidence is visionary or delusional is the question the next decade will answer — and the $119 billion total price tag means the answer matters enormously either way.