ANALYSIS: Elon Musk Wants to Make Chips in Seven Days — and No One Should Believe Him

A List of Unfulfilled Promises

It would be dishonest to ignore the track record. In 2019, Musk promised a million Tesla robotaxis on the roads by 2020. It’s now 2025, and the first commercial service still doesn’t exist on a large scale. In 2017, he announced the Tesla Semi for 2019—the first significant deliveries didn’t begin until 2022. The Cybertruck, announced in 2019, was delivered four years late and with downgraded specifications.

The Hyperloop, announced in 2013 as a transportation revolution, has become a tunnel in Las Vegas where Teslas travel at 50 km/h. The Optimus robot, presented in 2022 as the future of physical labor, was shown stumbling on stage. Neuralink promised brain implants for healthy humans—and is still struggling with its first clinical trials.

That doesn’t mean Musk always fails. SpaceX is a spectacular success. Starlink works. Tesla has transformed the automotive industry. But there is a systematic and well-documented gap between Musk’s timeline promises and the reality of execution—a gap that, in the semiconductor sector, could amount to tens of billions of dollars.

The Difference Between Rockets and Chips

And yet, some might say: he succeeded with SpaceX, so why not with chips? The answer lies in one word: ASML. SpaceX was able to reinvent the rocket because launch vehicle technology wasn’t locked down by an industrial monopoly. The materials were available. Engineers could be trained. The physics were understood.

Advanced semiconductors, on the other hand, depend on a supply chain of unprecedented complexity. A single supplier—ASML—manufactures EUV lithography machines. A single country—Japan—produces some of the critical photoresists. A single ecosystem—Taiwan’s—possesses the know-how accumulated over three decades of large-scale production. Musk cannot “disrupt” the physics of semiconductors the way he disrupted the cost of reaching orbit.

One hundred thousand GPUs, and it’s still not enough

To understand Terafab, you have to understand xAI’s insatiable appetite. Musk’s artificial intelligence company has built what is described as the world’s largest computing cluster in Memphis, Tennessee—more than 100,000 Nvidia H100 GPUs, installed in just a few months in a converted warehouse. The power consumption is so massive that it required special agreements with the local power grid.

But 100,000 GPUs aren’t enough to train next-generation models. OpenAI, Google DeepMind, Meta AI—all are building even larger clusters. The race for AGI (artificial general intelligence) is a race for raw computing power, and that power depends on chips that Musk currently buys from Nvidia—his direct competitor in the AI race.

Buying his weapons from his rival. That’s the situation Musk wants to resolve. And that’s the only logic that makes Terafab understandable—not as a rational industrial project, but as a strategic survival move in the AI war.

Vertical Integration as a Doctrine

This is the Musk doctrine, applied to every company. Tesla manufactures its own batteries. SpaceX manufactures its own engines. Neuralink designs its own implants. xAI is expected to manufacture its own chips. The logic looks appealing on a whiteboard. In a semiconductor cleanroom, however, it collides with the reality of yields, defects per billion transistors, and the years of learning required to achieve mass production.

Samsung, with decades of experience, still struggles to compete with TSMC on the most advanced nodes. Intel, the inventor of the microprocessor, lost its manufacturing crown for an entire decade. And Musk thinks he can do better. Starting from scratch. On a Saturday.

The semiconductor war is already here

Terafab isn’t emerging from a vacuum. It’s emerging into a world where semiconductors have become the most powerful geopolitical weapon since oil. The United States has imposed the most severe trade restrictions on China since the Cold War—banning the export of advanced chips, lithography machines, and even technical expertise. ASML can no longer sell its EUV machines to China. Nvidia can no longer sell its AI training GPUs to Chinese companies.

In this context, any announcement regarding U.S. manufacturing capacity becomes as much a political act as an industrial one. And Musk, who maintains close ties with the Trump administration—to the point of having headed the DOGE department tasked with reducing federal spending—knows exactly which button to push. Terafab isn’t just a factory. It’s a bargaining chip to secure grants, military contracts, and a position of power within the U.S. national security framework.

The CHIPS Act and Public Funding

The question no one is asking loudly enough: how much public money does Musk hope to secure? The CHIPS Act has already distributed billions to Intel ($8.5 billion), TSMC ($6.6 billion), and Samsung ($6.4 billion). Funds remain. And the current administration, which has shown a clear willingness to favor politically aligned companies, might see Terafab as an ideal vehicle.

But there’s a paradox that Musk can’t resolve with rhetoric alone. CHIPS Act subsidies are contingent on verifiable production commitments, binding timelines, and clawback clauses—mechanisms to recover funds if targets aren’t met. Promising a launch on a Saturday is easy. Sticking to a five-year timeline under federal oversight is an entirely different matter.

The Industry’s Telling Silence

The most revealing aspect of the Terafab announcement isn’t what Musk said—it’s what the industry didn’t say. No reaction from TSMC. No comment from Intel. No analysis from Samsung Foundry. The silence of the semiconductor giants in the face of Musk’s announcement isn’t politeness—it’s polite professional disdain. These companies know what it takes to manufacture a chip. And they know it’s not something you just “launch” on a Saturday.

Pat Gelsinger, Intel’s former CEO, used to say that building a fab is “the most complex construction project on Earth.” Morris Chang, TSMC’s founder, spent thirty years building Taiwan’s ecosystem. Thirty years of training engineers, fine-tuning yields, and perfecting chemical processes at the atomic level. And yet, even TSMC admits that its Arizona fab has experienced significant delays due to a shortage of skilled American workers.

The talent problem—the real bottleneck

Building the facility, buying the machines—that’s the easy part. The impossible part is finding the engineers. The United States trains about 30,000 semiconductor engineers per year. The industry needs more than 60,000. The shortage is structural, deep-rooted, and cannot be solved with a tweet.

TSMC had to bring hundreds of Taiwanese engineers to Arizona to run its fab—causing well-documented cultural and labor union tensions. Intel has been hiring heavily, but training times for a process engineer still take two to three years. Musk can promise astronomical salaries. He cannot speed up the physics of human learning.

Why a Saturday Changes Everything in Storytelling

The choice of Saturday is no accident. It’s a statement. Normal companies launch projects on Tuesdays, in a conference room, with a PowerPoint presentation. Musk launches on a Saturday, because Saturday says: I’m not a normal company. I don’t follow your schedule. I don’t follow your conventions. Saturday is the day for revolutions—or publicity stunts.

It’s the same logic behind the Cybertruck unveiled at night, the Starship launch watched live by millions, and the Twitter announcements at 2 a.m. Musk has understood something that most CEOs haven’t: the spectacle IS the product. As long as people are watching, investors believe. As long as investors believe, the money flows. As long as the money keeps flowing, the project may eventually—perhaps—one day—come to fruition.

The Cult of Personality as an Industrial Strategy

There is something deeply unsettling about the fact that a single man can announce a project worth tens of billions of dollars in a social media post—without a public business plan, without confirmed partners, without a selected site—and have the whole world take it seriously. This isn’t engineering. It’s industrial mythology.

And the mythology works—until it doesn’t. WeWork was a mythology. Theranos was a mythology. FTX was a mythology. All these companies shared one thing in common: a charismatic founder whose vision was so grand that it made technical questions seem out of place. Musk isn’t Elizabeth Holmes—he has real successes to his credit. But the mechanism for suspending disbelief is exactly the same.

The Actual Cost of a State-of-the-Art Fab in 2025

Let’s talk numbers, since Musk doesn’t. A fab capable of producing chips at advanced nodes (3 nm or smaller) costs between $20 billion and $40 billion. That covers the cost of the building, the cleanrooms, air filtration systems capable of maintaining fewer than 10 particles per cubic meter, thousands of machines, ultra-pure chemical delivery systems, and electrical infrastructure capable of powering a small town.

If Terafab aims for the “tera” scale—that is, beyond the gigafab—costs could exceed $100 billion over the course of the project. For reference, that’s more than Luxembourg’s annual GDP. It’s more than SpaceX has spent in its twenty-year history. And that’s money that must be spent before a single chip is produced.

The realistic timeline—not Musk’s

Here’s what industrial reality looks like—not the post on X:

Years 1–2: Site selection, environmental studies, building permits, negotiations with local authorities. Years 2–4: Building construction, cleanroom installation, equipment ordering (ASML lead time: 18 to 24 months). Years 4–5: Machine installation, calibration, first test wafers. Years 5–7: Yield ramp-up, process optimization, first marketable products.

Seven years. Not seven days. And even that’s the optimistic timeline—the one that assumes no construction delays, no supply issues, no recruitment difficulties, and no technical surprises. In the history of fabs, this timeline has never been met. Never.

Designing a chip is not the same as manufacturing it

There’s a possibility that the industry is hinting at without saying so outright. Terafab might not be a fab in the traditional sense. Musk could be aiming for a fabless model—designing custom chips for xAI and Tesla, but having them manufactured by TSMC or Samsung. That’s the model used by Apple, Nvidia, Qualcomm, and AMD. None of these companies manufactures its own chips. They all have them fabricated by TSMC.

If that’s the case, the name “Terafab” is deliberately misleading. But it would be consistent with Musk’s strategy: promise the moon, deliver a satellite. Promise a factory, deliver a design office. The result may be excellent—Tesla’s custom chips for Autopilot are remarkable—but that’s not what the word “fab” implies.

The Advanced Packaging Hypothesis

Another possibility: Terafab could focus on advanced packaging—the assembly of chips from different manufacturers into a single module. This is the technology behind Apple’s processors (M-series) and AMD’s accelerators (MI300). Packaging is less complex than fabrication, requires more modest investments, and can be up and running in two to three years.

If Terafab is a packaging plant, the seven-day timeline for a “launch”—understood as the announcement of a plan—becomes almost reasonable. But once again, Musk doesn’t clarify. Ambiguity is the product.

Six companies, one mind

Even for a man who sleeps four hours a night, there are limits. Tesla. SpaceX. xAI. Neuralink. The Boring Company. X (formerly Twitter). And now Terafab. Seven companies, each demanding the attention of a full-time CEO. Seven fronts simultaneously in industries that have nothing in common—automotive, aerospace, artificial intelligence, neuroscience, infrastructure, social media, and semiconductors.

The myth of the multitasking genius has its limits. Steve Jobs ran only Apple and Pixar—and he was pushed out of Apple the first time because he couldn’t manage everything. Jeff Bezos stepped back from Amazon’s day-to-day management to focus on Blue Origin. Musk, on the other hand, adds companies to his portfolio the way others add tabs to their browsers.

And yet. Tesla’s results are stagnating. Sales are plummeting in Europe. X is losing advertisers. Neuralink is moving slowly. The DOGE department, which he ran with the subtlety of a bulldozer, has sparked a series of political crises. Adding semiconductor manufacturing to this list isn’t ambition—it’s the elevation of scattered focus to the status of a doctrine.

The Risk of the “Man-Institution”

There is a systemic risk that financial markets refuse to assess: what happens if Musk slows down? Not if he fails—if he simply slows down. If a health issue, a legal dispute, or a personal crisis reduces his work capacity by 20%, seven companies are affected simultaneously. Tesla lost 40% of its value in 2024 partly because investors doubted the attention Musk was paying to the company. Terafab adds another layer of risk to an already dizzying house of cards.

A country that prefers heroes to systems

Terafab isn’t just one of Musk’s projects. It’s a symptom. A symptom of a country that, faced with an existential industrial challenge—dependence on Asian semiconductors—prefers to bet on a man rather than a system. The CHIPS Act is a government program: methodical, slow, bureaucratic. Terafab is a tweet: spectacular, instantaneous, exciting.

The America of 2025 chooses the tweet. It chooses the billionaire entrepreneur who promises to solve in seven days what the government has failed to solve in seven years. It chooses mythology over methodology. And this choice says something profound about the state of the American social contract—about the collapse of trust in institutions, about the fascination with individual power, about the desperate need to believe that one man alone can save an entire system.

Technological sovereignty isn’t something you tweet about

Taiwan did not become the global hub for semiconductors thanks to a visionary entrepreneur. It became the global hub thanks to forty years of consistent industrial policy, massive investment in education, patient public-private partnerships, and an engineering culture that values precision over spectacle. Morris Chang never tweeted. He built.

If the United States truly wants to bring semiconductor manufacturing back home, it doesn’t need a Terafab. It needs to train 100,000 engineers, reform immigration policy to attract global talent, invest in basic research, and accept that technological sovereignty is built over decades, not news cycles. But that doesn’t fit into a 280-character tweet.

The SpaceX Factor—Why We Can’t Completely Ignore the Announcement

Here’s the paradox that makes Musk impossible to categorize: he’s already done the impossible. In 2002, when he founded SpaceX, the entire space industry laughed. Literally. Engineers at Lockheed Martin and Boeing organized betting pools on the date the company would go bankrupt. The first three Falcon 1 rockets exploded. The fourth reached orbit. And today, SpaceX launches more rockets than all the world’s governments combined.

This track record makes any criticism of Musk uncomfortable. Because he has already proven that the established industry could be wrong. Because he has already shown that sheer will, combined with virtually unlimited capital, could compress timelines that everyone thought were uncompressible. The question isn’t whether Musk is capable of surprising us—it’s whether semiconductors are a field where such a surprise is physically possible.

The Limits of Disruption Through Willpower

The answer, most likely, is no. Not on the scale he promises. Not within the timeframes he announces. SpaceX succeeded because Musk could iterate quickly—build, test, blow up, rebuild. The cost of a rocket failure is one rocket. The cost of a fab failure is billions of dollars and years lost, with no possibility of rapid iteration.

And yet. And yet. If Musk actually invests tens of billions, if he recruits the best engineers, if he accepts a realistic timeline of seven to ten years rather than seven days—then yes, Terafab could become something. Not what he’s promising today. Not by Saturday. But something. The difference between Musk and most charlatans is that he sometimes ends up delivering—just not what he promised, and never when he promised it.

What We Know, What We Don’t Know, and What We Refuse to See

Let’s summarize what we know for certain. One: Musk announced a chip manufacturing project called Terafab. Two: he said it would launch on a Saturday, seven days from now. Three: no technical, financial, or logistical details have been disclosed. Four: semiconductor manufacturing is the most difficult, most expensive, and slowest-to-get-off-the-ground industry in the world.

What we don’t know: the exact meaning of the word “launch.” The intended scale. The partners. The funding. The location. The technology. In other words—everything that matters.

Silicon will be the sole judge

In the semiconductor industry, there’s no room for bluffing. A wafer either works or it doesn’t. A yield is either 95% or 30%—and the difference between the two is the difference between profitability and bankruptcy. Atoms aren’t impressed by tweets. Physics doesn’t negotiate with billionaires. Lithography doesn’t speed up just because a rich man wants it badly enough.

Musk knows this. Somewhere, beneath the layers of theatrics and calculated provocation, he knows it. And maybe that’s why he’s launching on a Saturday—because on Saturday, no one asks the tough questions. By Monday, the analysts will have questions. On Tuesday, the engineers will have doubts. On Wednesday, reality will have begun to chip away at it. But on Saturday, for a few glorious hours, Terafab will be the most exciting thing in the world.

And maybe that’s all Musk wanted.

Signed, Jacques PJ Provost