In March 2026, Elon Musk announced a $25 billion investment to build the “Terafab” chip empire in Texas, integrating design to manufacturing in a one-stop shop, claiming that the final capacity will reach 70% of TSMC’s global total capacity. But from the 4680 batteries to this “most spectacular chip factory in history,” the biggest question from the outside world is only one: he said the same thing last time.
(Background recap: Breaking news! SpaceX acquires xAI with a valuation of $1.25 trillion, Musk is building a “space + AI” giant to prepare for IPO)
(Additional background: Tesla + xAI + SpaceX: Understanding Musk’s ultimate AI flywheel)
Table of Contents
Toggle
- What exactly is Terafab? Explained in simple terms
- Why is he doing this? Reason 1: Chips are simply not enough
- Reason 2: Tesla needs a new story
- Reason 3: He wants to send AI servers into space
- How difficult is this? Let the numbers speak
- Lessons from the 4680 battery
- If he really succeeds, what does it mean for tech and crypto industries?
- Genius or lunatic?
Imagine this: you’re on the road every day trying to hail a taxi, but drivers are always scarce, waiting is long, and prices are high. You can keep tolerating it — or decide to buy your own car factory.
That’s exactly what Musk did on March 21, 2026, except he didn’t buy a car factory — he’s building a chip factory himself. And not just any chip factory: he claims this is “the most epic chip manufacturing effort in history, bar none.”
“The most epic chip building exercise in history by far.”
— That’s his own assessment of the project. Humility has never been his style.
What exactly is Terafab? Explained in simple terms
Terafab is a semiconductor factory jointly built by Tesla, SpaceX, and xAI, with an investment of $20 to $25 billion, located in the northern part of Giga Texas in Austin.
If you’re unfamiliar with the semiconductor industry, think of this analogy: If TSMC is the world’s top outsourced chip manufacturer, what Musk is saying now is — I’m not hiring a manufacturer anymore, I’m building a bigger factory myself.
TSMC’s model is “foundry manufacturing”: they don’t design chips, only turn your design into actual chips and send them to you. This model has dominated the semiconductor industry for decades, with Apple, Nvidia, AMD as clients.
Terafab will take a completely different approach: from chip design, lithography (printing circuits onto silicon wafers), manufacturing, memory, packaging, to testing — all done in-house, all within a single building. Industry jargon calls this “vertical integration,” in plain language: from raw materials to finished product, everything is handled internally, no outsourcing.
Initial scale target: 100,000 wafers per month; ultimate goal: 1 million wafers — roughly 70% of TSMC’s current total global capacity. Annual chip output is projected to reach 100 to 200 billion units.
Why is he doing this? Reason 1: Chips are simply not enough
Musk’s business empire might be the most chip-hungry on Earth: Tesla’s autonomous driving needs computing power, Optimus humanoid robots need it, xAI’s large language models require it, SpaceX’s Starlink satellites also need it.
The problem is, ordering from giants like TSMC, Samsung, Micron, and waiting endlessly just isn’t enough. Musk himself said:
“We’re very grateful to our existing supply chain, to Samsung, TSMC, Micron and others, but there’s a maximum rate at which they’re comfortable expanding. That rate is much less than we would like… and we need the chips, so we’re going to build the Terafab.”
In translation: “We’re grateful to Samsung, TSMC, Micron, but their expansion speed is far below what we need. We need chips, so we’re building Terafab ourselves.”
He also threw out an even more exaggerated number: the total output of all wafer fabs on Earth is only about 2% of what all his projects require.
Whether this number is accurate or not, one thing is clear: Musk’s demand is so huge that any foundry alone would struggle to meet it.
Why is he doing this? Reason 2: Tesla needs a new story
By 2025, Tesla’s car sales declined for the second consecutive year. Europe’s market plummeted, China saw its first annual negative growth. Meanwhile, Musk faced political turmoil over DOGE, nine co-founders of xAI resigned one after another, and SpaceX was preparing for an IPO valued at $1.75 trillion.
In this context, Terafab is more than just a chip factory — it’s a “super narrative” linking the three companies. Tesla’s manufacturing prowess, SpaceX’s rocket tech, and xAI’s AI R&D could all share chips from this factory. It’s a glue to keep investors believing in Musk’s ecosystem.
Why is he also aiming to send AI servers into space?
This is where Musk’s cosmic ambitions come into play.
80% of Terafab’s capacity will produce D3 space chips — specifically for SpaceX’s orbital AI satellites. Musk believes space has two natural advantages: abundant solar energy (for continuous power) and a vacuum environment that’s easier to cool chips (heat dissipation being a major challenge).
He claims that within 2-3 years, space-based AI computation costs will be lower than ground data centers. The ultimate goal: produce 1 terawatt of computing power annually — equivalent to 10,000 large data centers running simultaneously.
He even threw out a far-fetched goal:
“I want us to live long enough to see the mass driver on the moon, because that’s going to be incredibly epic.”
The moon’s mass driver is a sci-fi device that uses electromagnetic force to accelerate objects off the lunar surface. When Musk said this, the audience chuckled. The question is: is he joking or serious? No one really knows.
How hard is this? Let the numbers speak
Musk has the ability to turn the impossible into possible — SpaceX’s rocket reusability is the best example. But semiconductor manufacturing is a different dimension of challenge.
First, scale: TSMC has built six fabs in Arizona, with a total investment of $165 billion, and those factories are not expected to produce 2nm process until 2029 — the process that Terafab claims to adopt. A single 2nm fab costs about $28 billion, with a lead time of at least 38 months.
Second, equipment: only one company in the world makes the most advanced EUV lithography machines — Dutch firm ASML. TSMC, Samsung, Intel are all waiting in line. Tesla has no priority and no orders.
Third, site selection: Terafab is planned right next to Giga Texas’s stamping plant. Lithography equipment is extremely sensitive to vibration — factory machinery, trucks nearby, even distant roads — all can cause failures in the delicate process.
Finally, talent: TSMC’s decades of process knowledge is the result of generations of engineers’ experience. It’s not something that can be quickly replaced by a few job openings. Nvidia founder Jensen Huang once said that wafer foundry is “a combination of engineering, science, and art” — an extremely difficult field. Even Intel, with vast resources, has suffered years of losses and setbacks in this area.
Lessons from the 4680 battery
In 2020, Musk promised a revolutionary 4680 battery at Tesla Battery Day, claiming it would drastically change EV costs. Five years later, actual output is only about 2% of that initial target.
Tech media Electrek commented on Terafab: “We’ve seen this movie before with battery cells, and we know how it ends.”
Another notable detail: the $20-25 billion needed for Terafab is not included in Tesla’s announced $20 billion capital expenditure plan for 2026. Where this money will come from remains unclear.
By the way, Musk also said at the launch event that he wants to “reinvent the entire cleanroom industry” — building a wafer fab where engineers can “eat cheeseburgers and smoke cigars.” It sounds relaxed and easy, but the reality is that existing cleanroom standards are the result of decades of engineering experiments. It’s not laziness or lack of effort, but physical limits of environmental control.
If he really succeeds, what does it mean for tech and crypto industries?
The most optimistic scenario: if Terafab truly operates, it would be the biggest structural shock to the semiconductor industry in decades.
Currently, global AI compute power is highly concentrated among a few companies — TSMC controls the most advanced processes, Nvidia dominates GPUs, and a handful of cloud giants control data centers. This centralization makes compute a scarce resource, a major bottleneck for AI development.
If Musk’s “space AI compute” vision comes true, it could turn scarcity into surplus, lowering inference costs. This would directly benefit the crypto industry: cheaper compute means lower costs for AI model calls, benefiting DeFi protocols, on-chain oracles, decentralized AI inference networks like Bittensor.
More fundamentally, chip autonomy is a sensitive geopolitical issue. Tensions in the Taiwan Strait have everyone pondering: “What if TSMC is gone — what happens to AI?” If Terafab succeeds, it at least partially answers this question.
Of course, the phrase “if he succeeds” carries a lot of skepticism.
Genius or lunatic?
Musk is someone difficult to evaluate with ordinary logic. He said he wanted humans to become a multi-planetary species, then made rocket reusability routine; he said he’d buy Twitter, then actually did; he said he’d build hyperloop trains, and… well, that one didn’t happen.
Terafab is at least 38 months from production, and the ultimate goal of 1 million wafers per month might be 10 years away or forever. Challenges include ASML’s equipment backlog, TSMC’s decades-long technological moat, and talent shortages.
But one thing is certain: he has turned a topic that used to circulate only among TSMC, Samsung, and Intel into something everyone is talking about.
And that attention, in itself, is a resource.
