One of the mistakes people make when discussing Elon Musk is treating every announcement as either inevitable genius or complete fantasy. The reality is considerably more complicated. Over the past two decades he has accomplished things that many experts genuinely believed couldn’t be done, from turning SpaceX into the world’s dominant launch provider to proving that electric vehicles could become a mass-market product. He’s earned respect for some of his business decisions. However the same cannot be said for all of his engineering decisions.. He’s also earned a reputation for timelines that rarely survive first contact with 10 seconds of thought.
That is precisely why announcements like Terafab deserve serious attention instead of immediate dismissal.
Unfortunately, they also deserve something that is often in short supply whenever Elon unveils a new vision; perspective and objectivity.
Despite my well known criticism of Elon this is not a “bash Elon” article. This is an analysis of a bold engineering claim (regardless of who made it). So with that.. let’s dive in…
A Hard look at Physics, Reality & Engineering
I’ve spent the better part of thirty years designing factories, building automation systems and watching large manufacturing projects evolve. One lesson gets reinforced every single time. There are always two completely different conversations taking place. Investors talk about markets, valuation and future demand. Engineers quietly start asking where the electrical service enters the property and whether anyone has calculated how many cubic yards of concrete this thing actually requires.
Those are very different conversations.
One determines whether the project sounds exciting. The other determines whether it ever exists.
Terafab immediately caught my attention because, if the publicly discussed numbers are even close to accurate, it would represent one of the largest manufacturing projects ever attempted. We’re talking about a facility approaching one hundred million square feet with ambitions that extend far beyond simply fabricating semiconductors. The vision appears to include logic, memory, advanced packaging, testing and, eventually, perhaps even developing a replacement for the EUV lithography systems that currently define the leading edge of semiconductor manufacturing.
It’s an extraordinary vision
It’s also one that becomes significantly more complicated the moment you stop looking at the rendering and start thinking like the people who would actually have to build it.
The first mistake almost everyone makes is assuming they’re looking at a factory.
They’re looking at the center of an entirely new industrial ecosystem.
Once a project reaches something on the order of one hundred million square feet, the building itself almost stops being the interesting part. The supporting infrastructure becomes the real engineering challenge because none of it exists simply to make the rendering look impressive. It exists because physics eventually demands it.
Before the first piece of semiconductor equipment ever arrives, someone has to acquire and prepare thousands of acres of land, complete years of environmental studies, relocate utilities, construct miles of new highways capable of supporting continuous heavy construction traffic, build rail infrastructure to deliver structural steel and process equipment, install utility corridors carrying electrical transmission, fiber, natural gas, water and wastewater systems, construct multiple high-voltage substations capable of supplying several gigawatts of continuous electrical demand, develop ultra-pure water treatment facilities, chemical storage and distribution systems, emergency response infrastructure and eventually housing for the tens of thousands of engineers, contractors, technicians and support personnel required to build and operate the site.
Read that sentence again. I’ll wait.
None of those projects actually manufacture semiconductors.
They simply create the conditions under which semiconductor manufacturing might eventually begin.
Terrafab-ville? Terrafab-burg?
The utility requirements alone deserve their own engineering organization. Estimates for a fully built-out Terafab suggest continuous electrical demand somewhere in the neighborhood of 3-5 gigawatts. To put that into perspective, that’s the output of 3-5 nuclear plants (which take 15-20 years to build) and enough electricity to power millions of homes, and it fundamentally changes the discussion from “Which utility serves the site?” to “Do we need to build additional generation capacity?”
People will say, but we can use solar. OK. That’s 3-5GW is 20-50 square miles of solar panels. For scale the gigantic Terrafab building is 3.5 square miles.
Multiple high-voltage transmission corridors, dedicated substations, redundant feeds and enormous backup systems cease being optional design features and instead become the minimum requirements for keeping production running.
Water presents a similar challenge, although it receives considerably less attention. Modern semiconductor fabrication consumes astonishing quantities of ultra-pure water that must be purified, distributed, recovered, recycled and treated continuously. Building that infrastructure is a major industrial project in its own right, and every gallon has to meet purity standards that would make pharmaceutical facilities seem relaxed by comparison. And we building in Texas. Not exactly a state known for it’s overabundance of water.
The logistics become equally fascinating once you begin thinking beyond the finished building. Tens of thousands of construction workers don’t simply appear every morning. They have to live somewhere. They have to drive somewhere. Their materials have to arrive somewhere. Hundreds of thousands of truckloads of concrete, structural steel, piping, HVAC equipment, electrical gear and process equipment will move through the site over many years. Eventually those workers are replaced by process engineers, maintenance technicians, software developers, facilities personnel and manufacturing operators who also require housing, transportation, schools, hospitals and all the other infrastructure that follows large populations wherever they go.
Oh … good news! Your assigned parking spot is 3 miles from the building.
Again, none of this is criticism.
It’s simply scale and reality. And it’s not as simple as “We’ll figure it out”.
Every project reaches a point where it stops being what people call it and starts becoming something else entirely.
Terafab may be described as a semiconductor factory, but at full scale it begins behaving much more like a new city whose primary industry happens to be semiconductor manufacturing.
Ironically, everything I’ve described so far is probably the easy part.
Given enough money and enough time, humanity already knows how to pour concrete, erect structural steel, build substations and install utility infrastructure. These are incredibly difficult engineering problems, but they are fundamentally solved problems. We know how to do them because we’ve been doing them for generations. Construction schedules slip, budgets explode and politicians inevitably hold press conferences in front of excavators wearing spotless hard hats, but eventually the roads get built, the concrete cures and the lights come on. None of those disciplines are waiting for a scientific breakthrough.
Lithography is a completely different conversation.
One of the things I’ve noticed over the years is that software people and manufacturing people instinctively look at completely different parts of the same project. Software engineers see the application. Manufacturing engineers see the factory. Civil engineers see the roads. Utility engineers see the substations. Financial analysts see the addressable market. Everyone is looking at the same rendering while mentally replacing it with the part they understand best. That’s why conversations around projects like Terafab become so strange. Part of the room is debating AI model sizes while another part is wondering where the nearest concrete batch plant is located.
Once the building exists, the conversation shifts from conventional engineering to one of the most exclusive clubs on Earth. Most people have never heard of ASML, yet it is difficult to overstate their importance to modern civilization. Without this Dutch giant, there are no leading-edge processors, no state-of-the-art GPUs, no latest-generation smartphones and certainly no AI revolution. They manufacture the EUV lithography systems that define the bleeding edge of semiconductor production, and after spending a little time learning what goes into one of these machines, you begin to understand why nobody else has managed to compete successfully.
Calling an EUV scanner a machine almost feels disrespectful. It is really an entire manufacturing ecosystem compressed into a single product. Each system contains well over one hundred thousand individual components sourced from hundreds of specialized suppliers. The optics are manufactured by Zeiss to tolerances measured in fractions of the wavelength of visible light. The light source generates plasma by striking microscopic droplets of molten tin with high-powered lasers tens of thousands of times every second. The wafer stage accelerates at astonishing rates while maintaining positioning accuracy measured in nanometers. Every subsystem sounds impossible until you discover that somebody actually solved it twenty years ago.
ASML doesn’t build these machines quickly. They build them carefully.
Final assembly takes months. Installation at the customer’s site takes months. Qualification takes months. The company currently manufactures these systems in quantities measured in dozens per year, not hundreds, and those dozens are already spoken for by Intel, TSMC, Samsung, Micron, and every other company attempting to remain competitive at the leading edge. Terafab doesn’t get to skip the line simply because its building is larger than everyone else’s.
Whenever I bring this up, someone inevitably says, “Fine, then Elon will just build his own.”
I understand the optimism.
I also think that statement dramatically underestimates what ASML actually represents.
Commercial EUV lithography wasn’t invented by one brilliant engineer working in a garage. It emerged from roughly three decades of collaboration involving ASML, Zeiss, Cymer, Intel, Samsung, TSMC, IMEC and hundreds of suppliers solving thousands of extraordinarily difficult problems one painful lesson at a time. The machine is impressive, but the accumulated knowledge required to build the machine is even more valuable. If you gave me unlimited money tomorrow, I could hire outstanding engineers, lease buildings and begin designing an alternative. What I couldn’t buy is thirty years of institutional experience watching prototypes fail in very specific ways until somebody finally understood why.
Could a company with effectively unlimited capital eventually produce a competing EUV platform?
I think so.
Would I expect it to happen in less than 10 years?
Hell no.
Even with the ability to recruit exceptional talent and remove every bureaucratic obstacle imaginable, I would still expect something on the order of 10-20 years before a truly competitive production-ready system existed. Engineering is wonderfully indifferent to optimism. It rewards persistence, experimentation and accumulated knowledge far more than ambition.
Then there is the discussion around Free Electron Laser technology, or FEL, which has become increasingly interesting because it represents an attempt to leapfrog today’s laser-produced plasma approach rather than merely copy it. On paper the idea is compelling. Instead of solving yesterday’s problem, solve tomorrow’s. If it works, you don’t catch ASML. You potentially pass them.
There is just one small complication.
Nobody has ever commercialized an FEL-based semiconductor lithography system.
Research accelerators already exist. National laboratories operate them every day. Turning one into a machine that fits inside a semiconductor fab, operates around the clock, achieves world-class uptime, produces hundreds of wafers per hour and remains economically competitive is an entirely different engineering challenge. It isn’t merely replacing one subsystem with another. It’s replacing the the most sophisticated manufacturing machine ever built with a technology that has never been proven in commercial production.
That doesn’t make it impossible, it simply means you’ve replaced a 30-year engineering program with another 30-year engineering program.
The biggest constraint, however, may not be technology at all.
It may simply be people.
One of the unfortunate realities of highly specialized engineering is that expertise doesn’t scale particularly well. There are probably only a few thousand engineers on the planet who truly understand advanced lithography at the level required to design the next generation of semiconductor manufacturing equipment, and many of them already work for ASML, Zeiss, Intel, Samsung, TSMC, Applied Materials etc... Their knowledge isn’t sitting in a database waiting to be downloaded into the next generation of AI. It exists because they’ve spent decades solving obscure problems that most of us will never even encounter.
Silicon Valley has become accustomed to believing that difficult problems can be solved by hiring more smart people. Manufacturing doesn’t always work that way. You can hire intelligence. You cannot hire 20 years of accumulated experience overnight.
Which leads me to the slightly ridiculous question that crossed my mind while reading about Terafab.
Would it actually be easier to buy ASML?
I’m only half joking.
If the long-term objective is to control advanced lithography, acquiring the one company that already solved the hardest engineering problem in semiconductor manufacturing certainly sounds more efficient than recreating thirty years of history. Of course, convincing the Dutch government and Europe’s technology sector to part with one of their most strategically important companies probably belongs in the same category as colonizing Mars. Interesting to think about, but not something I’d put on my short term to-do list.
None of this should be interpreted as rooting against the project. Quite the opposite.
A robust, vertically integrated semiconductor ecosystem in the United States would be strategically valuable, economically important and technologically exciting. The supply chain disruptions of the past several years demonstrated how fragile global manufacturing can become when too much capability is concentrated in too few places. Building more domestic capacity is a goal worth pursuing.
Aspirational Dreams != Reality
The challenge is separating technical possibility from commercial timelines.
History has taught us that Elon Musk excels at convincing people to attempt things that everyone else considers impossible. History has also taught us that his timelines frequently assume engineering problems will cooperate simply because they have a worthy objective. The new Roadster remains elusive after nearly a decade. The Tesla Semi took years longer than originally promised and is still not in mass scale production. Level 4 Full Self Driving has been “just around the corner” for longer than a decade. There are not “a million” RoboTaxis on the roads (and that was supposed to be back in 2020). Those examples don’t diminish his accomplishments. They simply remind us that vision and execution exist on different clocks and that any statement or prediciton needs to be measured..
There is another reality that investors should probably keep in mind as well. Elon has always understood that capital follows vision. Extraordinary projects require extraordinary amounts of money, and extraordinary amounts of money usually require investors who believe tomorrow will be dramatically better than today. Whether you’re raising money for reusable rockets, electric vehicles or the largest semiconductor manufacturing complex ever conceived, the incentive is always the same: keep the conversation focused on where you’re going rather than where you are. That isn’t unique to Elon. It is the nature of capital-intensive technology businesses. He just happens to be better at it than almost anyone in history.
The irony is that if someone announced plans to build a 100M square-foot warehouse, nobody would blink. If someone announced plans to build the world’s largest semiconductor fab, engineers would nod respectfully. If someone announced plans to replace ASML, people would understand they were discussing a decade or two of research and development. Terafab isn’t one ambitious engineering project. It’s half a dozen of the most ambitious engineering projects in the world happening simultaneously on roughly the same timeline.
That’s the part I struggle with. Not because I doubt that any individual piece can eventually be accomplished, but because experience tells me that difficult projects don’t add together. They multiply one another’s risks. It’s like efficiency. 95% * 95% * 95% * 95% starts to get to be a low number very quickly (81% in fact).
I genuinely hope Terafab succeeds because manufacturing occasionally needs people willing to attempt the absurd. Many of the technologies we now take for granted began life as ideas that sounded unreasonable to everyone except the people building them. But I’ve also learned that engineering has an uncanny ability to humble even the brightest people in the room.
I’ll happily applaud the first profitable production wafers that roll out of Terafab.
I just suspect I (and Elon) won’t be around when that happens.
The Automation Navigator is written by Sean Dotson PE, founder of Automation AMA. Over 20 years, Sean built RND Automation from the ground up into a successful industrial automation company and ultimately through a private equity exit.
Today, he works with manufacturers and business owners who need an experienced sounding board, a second set of eyes, or hands-on help improving and changing their business.
That may include operations, growth, technology, leadership, sales, acquisitions, or preparing for an eventual exit. The focus is practical: identify what is holding the business back, challenge assumptions, and turn ideas into measurable results.
Learn more at www.automationAMA.com or reach out directly.
