Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.

Terafab: Musk’s Plan to Build the World’s Largest Chip Factory

One man already controls electric cars, rockets, satellites, social media, a humanoid robot company, and the world’s largest AI cluster. Now he wants to make the chips too.

The Announcement Nobody Should Have Been Surprised By

On March 23, 2026, Engadget broke a story that should have been front-page news everywhere but somehow competed for oxygen with the latest TikTok controversy. Elon Musk announced Terafab, a joint venture between Tesla, SpaceX, and xAI, to build what he called the largest chip manufacturing facility ever constructed. Location: Austin, Texas. Target output: one terawatt of annual compute production.

Let that sink in for a second. Not gigawatt. Terawatt.

For context, the Colossus GPU cluster in Memphis currently consumes about 250 megawatts. Terafab is proposing to produce chips at a scale four thousand times that facility’s power envelope. Whether you find that inspiring or terrifying probably says more about your relationship with concentrated power than your understanding of semiconductor physics.

The stated purpose is straightforward: produce custom silicon for Tesla vehicles, SpaceX spacecraft, xAI’s training clusters, and the Optimus humanoid robots. Stop buying from TSMC. Stop depending on Samsung. Bring the entire computational supply chain under one roof, controlled by one corporate ecosystem, answerable to one man.

Vertical integration, Musk-style.

What We Know (And What We Don’t)

Here’s what was conspicuously absent from the announcement: a production timeline, a confirmed chip architecture, and an investment figure.

That’s not a small omission. That’s the whole business case.

TSMC spent more than $30 billion on its Arizona fabs, and those facilities still aren’t fully operational. Intel has burned through tens of billions trying to modernize its foundries. Samsung’s Texas expansion has been plagued by delays. Building a cutting-edge semiconductor fab is one of the hardest industrial undertakings humans have attempted, right up there with nuclear power plants and space stations. The difference is that when a nuclear plant goes over budget, nobody tweets about it at 2 AM with a skull emoji.

So what do we actually know?

The joint venture structure. Tesla, SpaceX, and xAI are pooling resources. This matters because each company brings something to the table. Tesla brings manufacturing scale and robotics expertise. SpaceX brings radiation-hardened chip design requirements and a culture of rapid iteration. xAI brings massive compute demand and AI training workloads that would justify the capital expenditure. Together, they represent a captive customer base large enough to make the economics work without needing to sell chips externally.

The location. Austin, Texas, is already home to Tesla’s Gigafactory and has become a de facto Musk company hub. Texas offers favorable tax treatment, available land, relatively fast permitting, and a governor who would probably approve a rocket launch pad in a school zone if Musk asked nicely enough.

The ambition. One terawatt of annual compute production. To put that number in perspective: the entire global semiconductor industry currently produces chips consuming roughly 300-400 gigawatts at peak load. Musk is proposing a single facility that would, in theory, produce more compute capacity per year than the entire world currently uses.

This is either the most ambitious industrial project since the Manhattan Project or the most expensive vaporware announcement in history. Possibly both.

The Bell Labs Comparison (And Why It Falls Short)

When people reach for a historical parallel to what Musk is attempting, they usually land on Bell Labs. And the comparison is instructive, though not in the way most people think.

Bell Labs, at its peak in the mid-20th century, was the most productive research institution in human history. It invented the transistor, the laser, information theory, Unix, C programming language, and the cosmic microwave background radiation discovery (that last one accidentally, which is the most Bell Labs thing imaginable). It did all this because AT&T’s telephone monopoly generated so much profit that the company could afford to let brilliant people tinker with whatever interested them.

But Bell Labs made components. It invented the transistor; it didn’t build all the world’s phones, computers, satellites, and cars that used transistors. The knowledge diffused. Other companies took the inventions and ran with them. Competition happened.

Terafab proposes something different. This isn’t a research lab whose innovations spread across an industry. This is a vertically integrated manufacturing facility whose output stays inside a closed ecosystem. Tesla chips go in Tesla cars. SpaceX chips go in SpaceX rockets. xAI chips train xAI models. The chips power the Optimus robots that work in Tesla factories that build more Tesla cars.

It’s a closed loop. And closed loops, historically, produce one of two outcomes: breathtaking efficiency or catastrophic fragility. Usually both, sequentially.

The Concentration Alarm

Here’s where the conversation gets uncomfortable.

Before Terafab, Musk already controlled the leading electric vehicle company, the only operational orbital launch provider with reusable rockets, a satellite internet constellation covering most of the planet, a social media platform, the world’s largest AI training cluster, and a humanoid robot program with ambitions to deploy millions of units.

Now add chip manufacturing.

The Iron Law of Oligarchy describes how power concentrates in organizations over time, almost as a law of social physics. What Musk is building isn’t an organization. It’s an ecosystem. And the concentration pattern isn’t slow-motion institutional drift. It’s deliberate architectural design.

Consider what Terafab means in practice. One person’s companies would design the chips, manufacture the chips, install the chips in vehicles and robots and spacecraft, train AI models on those chips, and deploy the resulting intelligence across a fleet of machines operating in the physical world. The algorithms baked into those chips would decide how Teslas drive, how Optimus robots move, how SpaceX vehicles navigate, and how xAI models think.

Who audits those algorithms? Who checks the silicon for backdoors? Who verifies that the chips in 50 million Tesla vehicles don’t have capabilities that haven’t been disclosed? The answer, right now, is nobody. There is no regulatory framework for vertically integrated chip-to-deployment pipelines. The governance gap is total.

Compute is becoming the new oil. And Musk is trying to own the refinery, the pipeline, and the gas stations.

The Energy Math That Nobody Wants to Do

Let’s talk about power, because the numbers here are genuinely alarming.

A cutting-edge semiconductor fab consumes enormous amounts of electricity. TSMC’s facilities in Taiwan use approximately 5% of the country’s total power output. A single EUV lithography machine draws about one megawatt. A modern fab might have dozens of them, plus the clean rooms, vacuum systems, chemical processing, and cooling infrastructure.

But forget the fab itself. Think about what the chips will do once they’re built.

Data centers consumed 183 terawatt-hours in 2024, roughly 4% of total US electricity generation. By 2030, projections suggest data centers could consume 8-10% of global electricity. And that’s with current chip production rates. Terafab proposes to dramatically increase the supply of compute, which, if demand follows supply (and it always does in computing), means dramatically increasing electricity consumption.

Where does the power come from? The US grid is already strained. New natural gas plants face environmental opposition. Nuclear takes a decade to permit and build. Renewables are intermittent. Fusion remains on a timeline that’s measured in “always twenty years away” units.

Musk knows this, of course. Tesla Energy sells solar panels and Megapack batteries. SpaceX has explored space-based solar concepts. There’s likely a plan to make Terafab partly self-powered, or at least to locate it near sufficient generation capacity. But a terawatt-scale chip operation would need power infrastructure that dwarfs anything currently connected to the Texas grid.

ERCOT, the Texas grid operator, has already experienced catastrophic failures (see: February 2021). Adding a facility that could consume more power than several large cities is not a minor grid planning exercise. It’s a fundamental infrastructure challenge that connects directly to the electron gap between what civilization needs and what it can currently generate.

Musk’s Track Record: Always Late, Directionally Correct

The easiest criticism of Terafab is that it won’t happen. And honestly, that’s a reasonable bet. Musk’s track record with timelines is, to put it charitably, aspirational.

The Cybertruck was announced in 2019 for a 2021 delivery date. It shipped in late 2023. Full Self-Driving has been “next year” since 2016. Starship took years longer than projected. The Boring Company’s tunnel network remains mostly theoretical. Neuralink’s human trials happened years behind schedule.

But here’s the thing: the Cybertruck exists. Starship works. FSD, while not truly autonomous, is the most advanced driver-assistance system on public roads. Musk is consistently wrong about when and consistently right about whether.

If Terafab follows the pattern, we’re looking at a facility that’s announced for 2028, breaks ground in 2029, starts producing test wafers in 2032, and achieves meaningful output around 2034. That’s still extraordinarily fast by semiconductor fab standards. And even a facility producing a tenth of the claimed capacity would be transformative.

The question isn’t whether Musk can build a chip fab. He can. The question is whether anyone should be comfortable with him doing so, given what he already controls.

The Governance Gap

This is the part where the Unscarcity framework matters.

The technology itself is neutral. Custom chips designed for specific applications are, in principle, a good thing. Vertical integration can produce better products, faster iteration, and lower costs. The semiconductor industry’s current dependence on TSMC represents a genuine geopolitical vulnerability that many governments are trying to address. If Terafab reduces that dependence, it solves a real problem.

But technology doesn’t exist in a vacuum. It exists inside power structures. And the power structure around Terafab has no democratic checks.

There’s no congressional oversight of the chip architectures. No independent audit of the AI models baked into silicon. No public review of the supply chain decisions. No mechanism for the millions of people who drive Teslas, use Starlink, or interact with xAI products to have any say in how the computational substrate underneath those products is designed.

The Foundation chapter of Unscarcity argues that post-scarcity technology requires post-scarcity governance. We don’t have that. We have 19th-century institutional frameworks trying to regulate 21st-century power concentrations, and the gap between what governance can do and what it needs to do grows wider every quarter.

The labor cliff is coming. Robots will need chips. Those robots will reshape the economy. The chips in those robots will encode decisions about what the robots can and cannot do. And right now, one person is positioning himself to make all those decisions, with no structural accountability beyond the market price of Tesla stock.

Whether this leads to one man’s empire or civilization’s liberation depends on governance decisions that nobody in power seems interested in making.

Terafab: Musk's Plan to Build the World's Largest Chip Factory