Inversion Semiconductor Is Building a Chip Fabrication Machine That Sounds Impossible

The Macro: The Chip Bottleneck Nobody Talks About

Everyone talks about chip design. Who has the best architecture, the most efficient transistor layouts, the cleverest instruction sets. Far fewer people talk about chip fabrication, which is where the actual bottleneck sits. You can design the most brilliant chip in history and still wait months to get it manufactured because there are exactly a handful of facilities in the world that can print transistors at the leading edge.

The reason for this concentration is a single machine. ASML’s extreme ultraviolet (EUV) lithography system costs roughly $380 million per unit, weighs over 150 tons, and requires multiple 747 cargo flights to deliver. There are about 200 of them in existence. TSMC, Samsung, and Intel own most of them. If you are not one of those three companies, you are probably not fabricating chips at the leading edge. This is the most consequential supply chain bottleneck in technology, and it has massive geopolitical implications. The entire Western semiconductor strategy depends on machines built by a single Dutch company.

The physics behind EUV lithography is well understood. You generate extreme ultraviolet light by hitting molten tin droplets with a high-powered laser, then focus that light through a series of mirrors to etch patterns onto silicon wafers. The machines work. They are also absurdly large, expensive, and slow. A new EUV system takes about 18 months to build and install. The question nobody has seriously asked until recently: is there a fundamentally different way to do this?

The Micro: Particle Accelerators, But Make Them Small

Inversion Semiconductor is developing a lithography machine that uses a completely different approach. Instead of EUV light, they’re working with particle accelerator technology shrunk by a factor of 1000. The claim is that this approach can scale transistors to physical limits while being dramatically faster to manufacture. 15x faster than current methods, according to their pitch.

Rohan Karthik is the CEO. He has a Master’s in Mechanical Engineering from Imperial College London and a Royal Academy of Engineering award. Before Inversion, he automated chip design at Arm and led the Karman Space Programme, building what he describes as record-breaking high power rockets. Daniel Vega is the CTO with a Master’s in Applied Physics from UCL, focused on particle physics applications. He built novel machine learning models for stability optimization of tabletop particle accelerators at CERN and developed particle accelerators for cancer treatment at Lumitron Technologies.

Those backgrounds matter. This is not a software team pivoting into hardware. These are people who have actually worked on particle accelerators and semiconductor design at institutions where that kind of work happens. A five-person team in San Francisco, part of YC’s Winter 2025 batch, taking on what is arguably the most capital-intensive and technically demanding problem in all of technology.

The reshoring angle is significant. Western governments are spending hundreds of billions on semiconductor independence. The CHIPS Act alone allocated $52 billion. But building new fabs using existing ASML machines doesn’t fundamentally change the dependency. You’re still relying on one company’s machines, one company’s maintenance teams, one company’s upgrade cycle. A genuinely different lithography technology could change the equation entirely.

I want to be clear about the ambition level here. They are trying to replace the single most complex machine ever built by human beings. The ASML EUV system is often cited as the most sophisticated manufacturing tool in existence. That’s not the kind of thing a five-person startup typically takes on. But someone has to, and the people who do it will need exactly the kind of backgrounds Karthik and Vega bring.

The Verdict

This is either one of the most important startups in the current YC batch or a physics experiment that will burn through funding before producing a working prototype. There is very little middle ground. The technical challenge is so extreme that incremental progress is still meaningful. If they can demonstrate a working proof-of-concept that prints features at relevant scales, that alone would be a significant event in semiconductor manufacturing.

The funding question is the obvious one. Lithography machines are not software. You cannot iterate in a garage. The capital requirements for this kind of hardware development are enormous, and the timeline from prototype to production is measured in years, not months. YC seed funding gets you started. It does not get you to a machine that can compete with ASML.

The geopolitical tailwind is real, though. Governments that are currently writing very large checks to reshore semiconductor manufacturing would be extremely interested in a lithography technology that doesn’t depend on ASML. That’s not just a commercial opportunity. That’s a national security opportunity, and those come with different funding mechanisms and different levels of patience.

At 30 days, the metric is whether they can demonstrate the core physics at lab scale. At 60 days, whether serious semiconductor industry people are paying attention. At 90 days, the question is funding. A company like this needs to move from YC to deep-tech investors or government grants quickly, because the burn rate on particle accelerator R&D is not compatible with standard seed stage economics. I have no idea if they’ll succeed. I think the attempt matters regardless.