Musk's Terafab: One Man Challenges the AI Chip Dominance

March 21, Austin. Elon Musk held a press conference in an abandoned power plant, announcing something no one in the AI industry dared to do: making their own chips.

Tesla, SpaceX, and xAI are jointly building Terafab—a $250 billion 2-nanometer chip fabrication plant. It integrates chip design, lithography, manufacturing, packaging, and testing all under one roof. The first product, Tesla AI5, boasts single-chip performance close to NVIDIA's H100, with claimed inference costs 10 times cheaper. Sample chips are planned for late 2026, with mass production in 2027. Target capacity: 1 terawatt of AI computing power annually—50 times the current global total AI computing power.

80% of these chips will be sent into space, running AI on satellites and distributed back to Earth via Starlink. Only 20% will remain on Earth. Because the electricity required for 1 terawatt of computing power is simply beyond what Earth's power grid can supply—space offers 5 times the solar radiation, higher vacuum cooling efficiency, and no grid bottlenecks.

The news sent shockwaves through the industry.

He's challenging another giant no one dares to touch.

In the AI chip market, NVIDIA holds a near-absolute monopoly.

Global AI infrastructure capital expenditure this year is estimated at $400-450 billion, with chip procurement accounting for $250-300 billion. The vast majority of this money flows to one company—NVIDIA. It has a backlog of $1 trillion in orders, a market cap exceeding $3 trillion, and gross margins above 55%. An H100 sells for $30,000, with demand far outstripping supply.

The entire AI industry is being choked by one company, and everyone knows it. Google has TPUs but uses them internally, AMD is catching up but the gap remains large, and Intel's foundry ambitions are a mess. No one has truly stepped up to say: I'll take you on head-to-head.

Then Musk stepped forward. Designing his own, manufacturing his own, even building his own factory.

What is he really thinking?

The surface reason is easy to understand—chip shortage. Tesla's autonomous driving, robotaxi, Optimus robot, and xAI's Grok are all computing power-hungry monsters. Queue up to buy H100s from NVIDIA? Even with large volumes, you might not get a spot. Use TSMC as a foundry? Apple is ahead in line.

But spending $250 billion just to solve the supply chain issue seems too heavy-handed. He sees something much bigger than "buying chips."

Lay out the cards in his hand: Tesla has millions of vehicles and robots on the ground. xAI has the Grok large model. SpaceX has rockets to send things into space. Starlink has a global satellite network to beam data back to Earth from the sky. Now Terafab adds the final piece—chip manufacturing.

From making chips, to running AI models, to launching into space, to global distribution. An entire chain, all in one person's hands.

The last person to do something like this was Rockefeller—controlling oil from extraction, refining, transportation, to retail, dominating the entire chain and becoming the most powerful person of that era. Only this time, the resource has shifted from oil to computing power.

Can it be done? Opinions are divided.

This matter is highly controversial within the industry. Optimists point out that Musk has done the "impossible" too many times—SpaceX, Starlink, Tesla—each time doubted but ultimately achieved. Wedbush directly gave Tesla a $600 price target.

The skeptics also have solid reasons. Jensen Huang responded with a subtle remark: "Building your own chip fab is an extremely difficult challenge." TSMC's process technology has accumulated over decades, with know-how across over 2000 steps that money can't simply buy. Bernstein did the math: reaching the 1-terawatt target could ultimately cost $5 trillion. Furthermore, every advanced process chip fab in the last decade—TSMC Arizona, Samsung Taylor, Intel Ohio—has gone over budget and been delayed. Without exception.

Whether it can be done, no one knows now. But a more worthwhile question is—

If it succeeds: A disruptive landscape.

First, a crack will appear in NVIDIA's monopoly.

The feeling of the entire AI industry being choked by NVIDIA is something everyone in the field has experienced. If a competitive product emerges with similar performance and 10x cheaper inference costs, even if only for Musk's own companies, it means NVIDIA loses a super client. Competitive pressure would force NVIDIA to lower prices or accelerate innovation. The price of AI computing power globally could drop as a result.

Second, AI computing power moves from Earth to space.

If the plan to send 80% of chips into space is realized, the physical foundation of AI computing changes. Previously, everyone assumed computing power resided in data centers, constrained by power grids, cooling, and land. Space deployment shatters this ceiling. Thousands of AI satellites operating in orbit, selling computing power to the world via Starlink—the potential of this business model is immense.

Third, the power structure of the AI industry will be reshuffled.

The current power distribution in AI is: NVIDIA makes chips, TSMC manufactures, Meta/Google/OpenAI build models, AWS/Azure provide cloud services. Each layer has different players. But if Musk integrates chip manufacturing + AI models + space deployment + global distribution, he alone spans four layers. This vertically integrated power would make all other players uncomfortable.

Finally, geopolitics. Currently, over 90% of the world's advanced chips are produced by TSMC, with fabs in Taiwan. Any incident in the Taiwan Strait could halt global AI. Terafab being built on US soil—for Washington, $250 billion for domestic advanced chip capacity is a bargain any way you calculate it.

My Take

Completed on time? Most likely delayed. Final cost? Most likely far exceeding $250 billion.

But the direction is right.

Musk is currently the only person in the world simultaneously holding chip manufacturing, AI models, rocket launches, global satellite communications, and millions of hardware terminals. Each of these five puzzle pieces is a trillion-dollar business; others simply don't have the conditions to replicate it.

The 20th-century power map was drawn by oil. This century is being redrawn by computing power.

Musk has staked out a huge plot of land on this new map. Whether anything grows there depends on the coming years. But the location he chose is indeed prime.