Elon Musk announced Terafab on Saturday, March 21, 2026, at the decommissioned Seaholm Power Plant in downtown Austin, Texas. In a roughly half-hour presentation, he unveiled a $20–25 billion semiconductor project targeting one terawatt of AI compute per year — and framed it as a step toward humanity becoming a galactic civilization.
👉 Five months on: the announcement's biggest gap — the location — was finally closed on August 6, 2026, when Grimes County was confirmed with a $16.8 billion first phase. Most of the other gaps are still open.
- When: March 21, 2026, around 8 p.m. local time, livestreamed by SpaceX on X. Texas Governor Greg Abbott sat in the audience.
- The headline claim: every advanced fab on Earth combined produces roughly 2% of what Musk says his companies will need.
- The plan: two fabs under one campus, each producing a single chip design — edge inference silicon for vehicles and Optimus, and space-hardened chips for orbital AI.
- The number nobody expected: about 80% of output was earmarked for orbital infrastructure, only 20% for ground-based use.
- What was missing: no construction timeline, no completion date, and no confirmed production site.
Terafab Announcement Date, Time and Venue
The event took place on the night of Saturday, March 21, 2026, inside the Seaholm Historic Power Plant — a decommissioned power station in downtown Austin that Musk chose as a deliberate backdrop. Light beams were fired into the sky above the building, drawing local news coverage before the presentation even started.
It was not a surprise so much as a countdown. On March 14, Musk said publicly that the Terafab project would launch in seven days. Seven days later, it did.
The signal ran back further still. Tesla's January 28, 2026 earnings call was where Musk told investors the company would hit a chip supply constraint within three to four years unless it built its own fabrication capability.
✔ For the record: the Terafab announcement date is March 21, 2026. Some coverage says March 22 — that reflects publication timing, not the event.
What Musk Actually Announced: 7 Key Reveals
- The name and the scale. Musk described Terafab as the most epic chip building exercise in history by far, targeting one terawatt of annual compute output.
- The supply gap rationale. Existing global fab output covers roughly 2% of projected Tesla and SpaceX demand. His framing was blunt: they either build it or they do not get the chips.
- Vertical integration under one roof. Design, lithography, fabrication, memory, advanced packaging and testing in a single facility, enabling a rapid mask-revision loop no existing fab offers.
- Two fabs, two chips. Musk clarified the next day that Terafab would technically be two fabs, each producing only one chip design.
- The process node. A 2-nanometer target — the most advanced node then entering commercial production anywhere.
- The capacity ramp. An initial 100,000 wafer starts per month, scaling toward one million at full build-out, and 100–200 billion custom AI and memory chips per year.
- The cost. $20–25 billion — and Tesla's CFO confirmed at the event that this was not yet inside Tesla's 2026 capital expenditure plan, which already exceeded $20 billion on its own.
⚠ Context that ages the March numbers: the $20–25B figure has since been superseded. SpaceX filings in May referenced $55 billion initial and $119 billion total for the prototype fab phases, and the August announcement was framed as a $16.8 billion first phase of a larger complex.
The Two Chip Families Unveiled
1. Edge-inference silicon (terrestrial)
Chips designed to run AI models directly on a device rather than in a data center. Target hardware: Tesla vehicles running Full Self-Driving, Cybercab robotaxis, and Optimus humanoid robots. Musk's AI5 generation sits in this lineage.
2. Space-hardened compute (non-terrestrial)
A separate chip built for orbit — radiation-hardened, with specialised thermal management to survive an environment with no air and no water cooling. Its destination is SpaceX's planned orbital data centers.
Splitting them was a deliberate manufacturing choice, not marketing. One fab, one design, no line changeovers — the way you maximise yield when you control both the chip and the customer.
The AI Sat Mini: The Reveal Most Recaps Skipped
Terafab was not presented alone. Musk also showed conceptual imagery of an orbital AI compute satellite — referenced in coverage as the AI Mini Sat — designed as the demand engine that justifies a terawatt of chips.
Published details of the concept describe a deployed length beyond 170 metres, solar arrays supplying around 100 kilowatts to onboard AI processors, and roughly 100 square metres of radiators for heat rejection. Musk indicated later versions would scale toward the megawatt range.
The scale of the intended constellation is the part that reframes everything: SpaceX has sought regulatory clearance for up to one million data-centre satellites in low Earth orbit.
💡 The connection to make: a million orbital compute satellites is where the terawatt number comes from. Terafab's capacity target was reverse-engineered from a space ambition, not from the chip market.
The 80/20 Split: Terafab Is Mostly Not for Tesla
This is the single most under-reported detail from the presentation. Roughly 80% of Terafab's compute output was earmarked for orbital AI infrastructure; only about 20% for terrestrial applications including Tesla vehicles and robots.
For anyone reading Terafab as a Tesla catalyst, that ratio matters enormously. The project is majority-aimed at SpaceX's space-based data centre ambitions, with Tesla's vehicle and Optimus silicon as the minority share.
It also explains why SpaceX ended up leading the full-scale build. Musk laid out the division in May: Tesla focused on the prototype fab, SpaceX leading the initial full-scale complex.
Every Number From the Presentation, in One Table
| Metric | Announced March 21, 2026 | Status as of August 2026 |
|---|---|---|
| Cost | $20–25 billion | $16.8B phase one confirmed; filings reference up to $119B |
| Compute target | 1 terawatt per year | Unchanged as stated goal; no capacity schedule published |
| Process node | 2 nm | Intel 14A named for the full-scale facility |
| Initial capacity | 100,000 wafer starts / month | Revised by Musk to a few thousand for the prototype fab |
| Full capacity | 1,000,000 wafer starts / month | Long-term goal; no timeline |
| Chip output | 100–200 billion chips / year | Unchanged as stated goal |
| Output allocation | ~80% orbital, ~20% terrestrial | Not restated in the August announcement |
| Location | Austin (Giga Texas North Campus) | Austin research fab + Grimes County production complex |
| Timeline | None given | Still none — no groundbreaking or first-wafer date |
How Markets and Analysts Reacted
The reaction was genuinely split, and it stayed split.
- Tesla stock wobbled — shares slid in Monday premarket trading before recovering to close higher on the session. The market could not decide whether Terafab was a moat or a cash drain.
- Wedbush reiterated a Buy with a Street-high $600 target, with Dan Ives framing Terafab as an opening move toward an eventual Tesla-SpaceX combination.
- Barclays held Equalweight at $360, cautioning that the project could demand spending many multiples above even a $50 billion bull case.
- Electrek published one of the harshest takes, arguing the announcement read as desperation rather than strength.
- Nvidia's Jensen Huang had already warned, before the event, that building advanced chip manufacturing is extremely hard — a line that followed the project into every subsequent analysis.
Tom's Hardware's later deep-dive added the structural objection: hitting one terawatt would require over 150 fab modules plus matching packaging capacity, and it would likely be the 2030s before Terafab could produce chips on its own process technology.
✔ Balanced read: the March presentation was a vision document with a price tag attached. Nothing announced that night was falsifiable within a year — which is exactly why the August site confirmation mattered more than the launch event did.
Announced in March vs. Delivered by August 2026
- April 7 — Intel joined. Intel signed on as a manufacturing partner, adding the fab expertise neither Tesla nor SpaceX had. Not announced in March.
- April — ground broken in Austin. Tesla started construction on the research fab at Giga Texas North Campus, the precursor facility.
- May 6 — the numbers grew. SpaceX filings referenced $55B initial and $119B total; Musk also scaled the prototype fab's stated output down sharply.
- June 3 — Grimes County moved. Commissioners approved a reinvestment zone for the Gibbons Creek Reservoir site.
- June 12 — SpaceX went public, turning the project's lead developer into a listed company with disclosure obligations.
- August 6 — the site was confirmed. Grimes County, $16.8B phase one, 3,000 permanent jobs, a $30M Texas Enterprise Fund grant, and the first rendering of a 2.5-mile-long building.
⚠ Still undelivered from the March pitch: a construction timeline, a first-wafer date, a finalized Tesla–SpaceX intellectual-property split, and any published path to the terawatt figure. SpaceX's own filings have described the Tesla arrangement as a framework rather than a binding commitment.
Checklist: What the Announcement Did Not Include
- No groundbreaking date for the production facility.
- No completion or first-production date.
- No confirmed manufacturing partner — Intel came 17 days later.
- No disclosed ownership percentages or funding split between Tesla and SpaceX.
- No production site beyond Austin; the Grimes County complex was not part of the March plan.
- No answer on where the workforce or the lithography equipment would come from.
👉 How to read any Musk announcement: separate the specification from the schedule. Terafab's specification has stayed remarkably consistent since March. Its schedule has never existed.
FAQ
What date was Terafab announced?
March 21, 2026, at an evening event at the Seaholm Power Plant in Austin, Texas, livestreamed by SpaceX on X.
How long was the Terafab presentation?
Roughly half an hour. Musk used it to lay out the demand rationale, the two chip families, the capacity targets and the cost range.
Where can I watch the Terafab announcement?
SpaceX broadcast it live on X, and the stream remains linked from the project's reference page. Full recaps were published by several outlets the following day.
What did Musk say Terafab was for?
Chips for Tesla vehicles and Optimus robots, and space-hardened silicon for orbital AI data centers — framed as a step toward becoming a galactic civilization, with science-fiction authors cited directly.
Has anything changed since the announcement?
Yes — substantially. Intel joined in April, the cost framing changed twice, and the production site was confirmed in Grimes County in August 2026 rather than Austin.
Summary
The Terafab announcement on March 21, 2026 was a specification without a schedule: a terawatt of annual compute, two dedicated fabs, a 2nm target, and a $20–25 billion price tag — with no date attached to any of it.
Five months later the specification has largely held while the numbers and the geography moved underneath it. Intel supplies the process expertise, SpaceX leads a $16.8 billion first phase in Grimes County, and the terawatt target remains an ambition with no published path.
The presentation's most durable insight was not the chip plant. It was the 80/20 split — confirmation that Terafab exists primarily to feed an orbital compute constellation, with Tesla's silicon as the minority share.
Following how this feeds Optimus production and Tesla's AI roadmap? The announcement set the targets; the construction milestones will decide whether they hold — bookmark this page for updates.
NEVER MISS AN OPTIMUS UPDATE
We track every Tesla Optimus development — specs, deployment milestones, pricing and competitive moves — updated as news breaks.