Source Foundry raised $400M from a distressed hedge fund and Sequoia. The target: ASML’s near-total monopoly on advanced chip lithography. The odds: near zero. The reward: a paradigm shift in AI hardware.
Here is the data. ASML’s EUV tools cost $150M+ each, require 50+ engineers per shift, and consume enough power to light a small town. They are the only path to 3nm and below. Every AI chip—NVIDIA H100, AMD MI300, Google TPU—runs through a Dutch fab built on a single optical architecture developed over 30 years. That is a single point of failure. And that is exactly what Leopold Aschenbrenner’s fund and Sequoia are betting $400M to break.
Context: Why Now? The semiconductor industry is in a structural bottleneck. AI training compute demand doubles every 3-4 months, but lithography improvements have slowed. ASML’s High-NA EUV is delayed, costs spiraling toward $400M per tool. The global supply of advanced nodes is constrained by the number of EUV machines ASML can ship—about 50 per year. Each tool is a 10-ton behemoth with 100,000+ parts. The physics itself is hitting limits: 13.5nm wavelength, 0.55 NA optics, extreme vacuum. The next step—Hyper-NA EUV—is not expected until 2028 and may never be economically viable.
Enter Source Foundry. Founded in 2025 by a Stanford materials scientist, hidden in stealth, and now backed by Aschenbrenner’s personal fund ($400M) and Sequoia ($100M). The company’s name is a tell: “Source” likely refers to the light source—the core of any lithography tool. The claim: a “simpler, cheaper, faster” lithography process. No details. No patents yet. No customers. Just a lab demo that convinced two of the most sophisticated tech investors in the world to take a high-conviction bet.
Core: The Technical Case From my own experience reverse-engineering DeFi yield contracts, I know that when a materials scientist leads a hardware startup, the breakthrough is almost never in optics. Optical lithography is a 50-year-old field with diminishing returns. The real innovation in lithography—if it comes—will be in the materials: resists, masks, or the interaction between light and photoresist. Source Foundry’s founder is a materials scientist, not an optical engineer. That is a signal.
What are the plausible alternates? Three paths: (1) Nanoimprint lithography (NIL)—mechanically stamping patterns—used by Canon for niche applications, but throughput is 1/10th of EUV. (2) Multi-beam electron beam direct write—ASML’s eBeam, but also slow. (3) Directed self-assembly (DSA)—block copolymers that self-organize into patterns—promising but still lab-only. The most likely candidate: a hybrid approach using a novel resist that can be exposed with a simpler, cheaper light source, perhaps a high-harmonic generation (HHG) source or a free-electron laser miniaturized to fit in a fab. HHG can produce coherent EUV light from a tabletop setup, but currently at low power (~1W vs. ASML’s 250W). If Source Foundry has cracked power scaling, it changes everything.
Quantitative narrative: ASML’s R&D budget is $4.5B/year. Source Foundry’s total funding is $500M—less than one quarter of ASML’s annual R&D. To close that gap, the new architecture must be fundamentally simpler. A 10x reduction in parts count, a 5x reduction in vacuum requirements, a 2x reduction in energy consumption. That is what “simpler, cheaper, faster” implies. But no one has ever built a high-volume lithography tool with fewer than 50,000 parts. The complexity is inherent to the physics.
Contrarian: The Unreported Angle The mainstream narrative is: “ASML is a moat, Source Foundry is a long-shot.” True, but incomplete. The real blind spot is the institutional lock-in. ASML is not just a tool supplier; it is a co-developer with TSMC, Samsung, and Intel. They share R&D, IP, and process recipes. A new lithography tool would require rewriting the entire process flow—thin films, etch, metrology, yield management. The risk of a fab line being down for weeks to retool is measured in billions of dollars. No foundry will risk that for a startup with no track record.
But here is the contrarian twist: The bottleneck is so acute that the largest AI chip buyers—Google, Microsoft, OpenAI—are desperate. They are investing in their own chip design (TPU, Maia, etc.) and now in manufacturing innovation. Aschenbrenner’s fund is explicitly a “compute bottleneck” thesis. The real prize is not a better EUV tool; it is a sovereign lithography capability for the US. ASML is Dutch, and the US government has already used export controls to block ASML exports to China. If the US wants to build its own advanced chip fabs without relying on a foreign monopoly, it needs a domestic alternative. Source Foundry, if successful, becomes a strategic asset—like how the US funded DARPA projects to create GPS and the internet.
Infrastructure-first critical lens: The supply chain for EUV is a spiderweb of single-source suppliers. Zeiss for optics, Cymer for light source, VDL for wafer stages. Source Foundry cannot replicate that. But if their tool uses a different light source and simpler optics, they can bypass Zeiss. That is exactly what a materials scientist would do: design a resist that works with a simpler, broad-spectrum source, eliminating the need for 13.5nm mirrors. The mirrors are the most expensive, most complex part of EUV. If Source Foundry can make a resist that works at 13.5nm with a 10x lower power requirement, and use a cheap, off-the-shelf laser, they collapse the cost structure.
Takeaway: What to Watch The next 18 months are critical. Source Foundry needs to produce a prototype that can image a 5nm half-pitch pattern at >10 wafers per hour. Anything less is a lab curiosity. Watch for patent filings in the US and Europe—if they start filing broadly on novel resist chemistries, that is a signal. Also watch for hiring of process integration engineers from TSMC and Intel. No one has ever scaled a new lithography technology from lab to fab without a deep partnership with a foundry. The most likely outcome: a proof-of-concept, a partnership with a US-based company like GlobalFoundries or Intel, and a $2B+ Series C. But the risk of complete failure is 90%. The question is not whether Source Foundry will succeed, but whether the AI chip industry can afford to wait another decade for an alternative to ASML.
s congestion is the real bottleneck. And it is not just about chips. It is about the physics of light, the inertia of supply chains, and the strategic interests of nations. This $400M bet is a signal that the market senses something breaking. The cheetah runs fast, but the hunted must be in sight.
