Musk's Terafab: A Person Challenging the AI Chip Dominance

On March 21, Austin. Elon Musk held a press conference at an abandoned power plant, announcing something that no one in the AI industry dared to do: to build chips himself.

Tesla, SpaceX, and xAI jointly constructed Terafab—a $25 billion 2-nanometer chip manufacturing plant. All processes from chip design, photolithography, manufacturing, packaging, to testing are completed under one roof. The first product, Tesla AI5, has performance close to the NVIDIA H100, with claimed inference costs being ten times cheaper, with plans to release samples by the end of 2026 and mass production in 2027. The target capacity: 1 terawatt of AI computing power per year—50 times the current total global AI computing power.

Eighty percent of the 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 power required for 1 terawatt of computing is simply beyond what the Earth's grid can supply—the solar radiation in space is five times that on the ground, and the vacuum has higher heat dissipation efficiency, eliminating grid bottlenecks.

Once the news broke, the industry exploded.

He is challenging a giant that no one dared to touch.

In the AI chip market, NVIDIA is basically an absolute monopoly.

Global capital expenditure for AI infrastructure is expected to be $400-450 billion this year, 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 value exceeding $3 trillion, and a gross margin of over 55%. The H100 sells for $30,000 each, with demand far exceeding supply.

The entire AI industry is being choked by one company, and everyone knows it. Google has TPUs but only uses them for itself, AMD is chasing but still far behind, and Intel's attempts at contract manufacturing have been disastrous. No one has truly stepped up to say: I will confront you head-on.

Then Musk stepped up. Designing it himself, manufacturing it himself, even building the factory himself.

What is he really thinking?

The surface reason is easy to understand—there aren't enough chips. Tesla's autonomous driving, robotaxi, Optimus robot, and xAI's Grok are all power-hungry monsters. Queueing up to buy H100s from NVIDIA? Even in large quantities, you might not get in line. Looking for TSMC to do contract manufacturing? Apple is ahead in line.

But $25 billion just to solve the supply chain is too heavy. What he sees is much bigger than "buying chips."

He lays his cards on the table: 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 transmit data from the sky back to the ground. Now Terafab fills in the last piece—chip manufacturing.

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

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

Can it be done? Opinions vary.

This matter has sparked significant controversy in the industry. Those optimistic say Musk has done too many "impossible" things—SpaceX, Starlink, Tesla, each time facing skepticism but ultimately succeeding. Wedbush directly set a target price of $600 for Tesla.

Skeptics have strong reasons too. Jensen Huang responded with a very subtle remark: "Building a chip factory is an extremely difficult challenge." TSMC's processes have accumulated over decades, with know-how involving over 2,000 steps, which cannot be bought with money. Bernstein calculated that to achieve the 1 terawatt target, the final cost could reach $5 trillion. Moreover, in the past decade, advanced process chip factories—TSMC Arizona, Samsung Taylor, Intel Ohio—have all exceeded budgets and faced delays. There are no exceptions.

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

If it succeeds: a disruptive pattern.

First, NVIDIA's monopoly will be breached.

Currently, the entire AI industry feels choked by NVIDIA, a sentiment everyone in the industry has experienced. If a substitute chip with similar performance and ten times cheaper inference costs appears on the market, even if it only serves Musk's own company, it means NVIDIA loses a super major client. Competitive pressure will force NVIDIA to lower prices or accelerate innovation. The overall price of AI computing power may decrease as a result.

Second, AI computing power moves from the ground to space.

The plan to send 80% of the chips into space, if realized, would change the physical foundation of AI computing. Previously, everyone assumed computing power was in data centers, constrained by the grid, heat dissipation, and land. Space deployment breaks this ceiling. Thousands of AI satellites operating in orbit, selling computing power to the world through Starlink—this business model has immense imaginative potential.

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

Currently, the power distribution in AI is: NVIDIA makes chips, TSMC does manufacturing, Meta/Google/OpenAI create models, and AWS/Azure provide cloud services. Each layer has different players. But if Musk integrates chip manufacturing, AI models, space deployment, and global distribution, he spans four layers alone. This kind of vertical integration power will make all players uncomfortable.

Finally, geopolitical implications. Currently, over 90% of advanced chips globally are produced by TSMC, with factories in Taiwan. If something happens in the Taiwan Strait, global AI will come to a halt. Building Terafab on U.S. soil, for Washington, means that $25 billion buys domestic advanced chip capacity, which is a deal that makes sense in any calculation.

My view:

Will it be completed on time? Most likely delayed. How much will it ultimately cost? Most likely far exceeding $25 billion.

But the direction is correct.

Musk is currently the only person in the world holding chip manufacturing, AI models, rocket launches, global satellite communication, and millions of hardware terminals simultaneously. Each of these five pieces is a trillion-dollar business, and others lack the conditions to replicate it.

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

Musk has circled a large piece of land on this new map. Whether it can yield results will depend on the next few years. But the chosen position is indeed good.