On August 6, a night-time render of Tesla and SpaceX’s under-construction Terafab in Grimes County, Texas, started making the rounds on X. The image showed a ring of structures with what looked unmistakably like a particle accelerator running underneath the facility. Joshua Steinman posted it with a simple question: is there a particle accelerator under there? Beff Jezos, founder of Extropic and a former Google quantum computing engineer, took it a step further, guessing that Musk might be planning to build a free electron laser based EUV machine for the fab.
Musk’s reply came a day later, and it was short even by his standards: “FEL FTW.”
Free Electron Laser, For The Win.
Two words that confirmed months of speculation and set off a wave of excitement among chip industry watchers, because if Musk really is building an FEL-based light source into Terafab, he’s attempting something that has eluded the entire semiconductor industry for over a decade, including ASML, the one company that has a near-total monopoly on the machines that make advanced chips possible.
To understand why this matters, you have to understand what actually happens inside a chip factory, and why the light source at the heart of it has become the biggest bottleneck in modern semiconductor manufacturing.
Why chips need light in the first place
Making a modern processor involves projecting incredibly fine patterns onto silicon wafers, in the same way a photograph is exposed onto film. The patterns define where transistors, wires, and other structures will eventually be etched. The smaller and sharper the patterns, the more transistors you can pack onto a chip, and the faster and more efficient that chip becomes.
The catch is that you can only draw features as small as the wavelength of light allows. Visible light, at around 400 to 700 nanometers, was fine for chips a generation or two ago. But to get down to the tiny features found in today’s most advanced processors, the industry needed something with a much shorter wavelength: extreme ultraviolet light, or EUV, at 13.5 nanometers.
Building a machine that can reliably generate, focus, and project EUV light turned out to be one of the hardest engineering problems in modern industry. ASML spent decades and tens of billions of dollars solving it, and today it is the only company in the world that sells EUV lithography machines. Every advanced chip made by TSMC, Samsung, or Intel passes through an ASML EUV scanner at some point. That single point of dependency is a big reason the entire global electronics industry watches ASML’s roadmap so closely.
The problem with how EUV light is made today
Here’s where it gets interesting, and where Terafab’s synchrotron ring comes in.
ASML’s current method for generating EUV light is genuinely strange when you first hear about it. Inside the scanner, a high-powered carbon dioxide laser fires at tiny droplets of molten tin, roughly 50,000 times a second. Each droplet gets hit twice: once to flatten it into a disc, and once to vaporize it into a plasma so hot it emits EUV light as a byproduct. That light gets collected by a curved mirror and sent on its way toward the wafer.
It works, and it’s the reason the entire EUV era of chipmaking exists at all. But it’s also inefficient and messy. Turning tin droplets into plasma wastes enormous amounts of electricity relative to how much usable EUV light comes out the other end, and the process leaves behind tin debris that coats and degrades the delicate mirrors inside the machine, shortening their lifespan and adding to maintenance costs. Because so few of the initial photons survive the journey through the mirror system and optics, only a small fraction of the light generated ever reaches the wafer.
For years, EUV was going to be the technology that carried Moore’s Law forward indefinitely. Instead, it’s run into a power ceiling. Chipmakers want brighter, cleaner EUV sources so they can run their scanners faster and cut costs per wafer, but the tin plasma approach is running up against physical limits on how much power it can squeeze out reliably.
That’s the opening a free electron laser is built to fill.
What a free electron laser actually is
A free electron laser works on a completely different principle than the laser pointer on your desk, or even the CO2 laser inside ASML’s tin-vaporizing setup. Conventional lasers rely on a gain medium, a material like a crystal or gas whose atoms get excited and release photons of a specific wavelength. An FEL skips the gain medium entirely. Instead, it takes electrons that have been stripped free of any atom, accelerates them to close to the speed of light using a particle accelerator, and then sends them through a device called an undulator, a long channel lined with alternating magnets.
As the electrons pass through those alternating magnetic fields, they get pushed side to side in a wave-like path, and that wiggling motion forces them to emit radiation. Do this with enough electrons moving coherently together, and the emitted light builds on itself until it comes out as an intense, tightly focused beam. Because there’s no gain medium involved, an FEL can be tuned to produce light at almost any wavelength simply by adjusting the electron energy and the undulator’s magnetic spacing, including exactly 13.5 nanometers.
The appeal for chipmaking is fairly direct. Free electron lasers can, in principle, generate EUV light with far higher power and much better wall-plug efficiency than tin plasma sources, since almost none of the input energy is lost heating up a target to plasma temperatures. There’s also no tin spray to contaminate the optics, which means fewer maintenance shutdowns and longer-lived equipment. Perhaps most importantly for a facility the scale of Terafab, one large, centralized FEL and its accelerator ring could theoretically supply EUV light to dozens of individual lithography bays at once, rather than every single scanner needing its own self-contained tin-droplet light source, the way ASML machines work today.
Researchers at Japan’s KEK laboratory have already demonstrated a working prototype along these lines, and it points to just how large a leap in efficiency is possible. Their setup uses about 7 megawatts of electricity to produce 10 kilowatts of usable EUV power. Scaled up, that ratio comes out meaningfully better than what current tin plasma systems manage, which burn through roughly four to five times as much electricity for the same EUV output. In a facility running around the clock, electricity is one of the largest ongoing costs, so a jump in efficiency like that changes the economics of the whole operation.
None of this is easy or cheap to build, which is exactly why nobody in the industry has done it commercially yet. Estimates for a single FEL-based light source run into the hundreds of millions of dollars, on top of the accelerator infrastructure itself, the shielding required around a machine that’s effectively a compact synchrotron, and the specialized undulators. A handful of startups, including one called xLight, have been trying to bring the idea to market by focusing purely on the light source and partnering with existing lithography makers rather than building an entire scanner from scratch. So far, none of them has shipped a working commercial system.
Why Musk confirming this is a bigger deal than it looks
If the render posted on X is genuinely a synchrotron ring built into the base of Terafab, and Musk’s reply suggests it is, then Tesla and SpaceX are not just building a chip fab. They’re attempting to build their own next-generation EUV light source in-house, something that would put them in direct competition with capabilities ASML has spent decades and enormous sums perfecting, and that no other company has managed to bring to production.
It also fits the pattern that’s defined Musk’s approach to hard manufacturing problems before, from rockets to batteries: rather than buying critical infrastructure from an outside supplier with pricing power and long lead times, build it yourself, even if it means reinventing physics-heavy equipment that has taken an entire industry decades to get right.
Whether Terafab’s FEL, if it exists, actually reaches the kind of reliability and output ASML’s tin plasma sources have after years of refinement is a separate question, and one that will take actual chips rolling off the line to answer. High-volume manufacturing has a habit of exposing problems that never show up in a lab prototype, especially with equipment as unforgiving as a particle accelerator running continuously at industrial scale.
But for a technology that researchers have been chasing since the early 2010s without a single commercial deployment to show for it, two words from Musk have put free electron lasers back at the center of the conversation about what comes after tin plasma in chipmaking.