No website that matters. No published papers. No disclosed patents. No whitepaper. No customers. No revenue. No product.
That is the complete public dossier on Source Foundry, a semiconductor lithography startup that has raised $500 million from Sequoia Capital and technology-strategy investor Leopold Aschenbrenner. Founded in 2025 and led by Stanford materials scientist Abdulmalik Obaid, the startup's stated ambition is to build lithography tools that are "simpler, cheaper, faster" than the extreme ultraviolet machines ASML spent three decades perfecting.
I have audited 2017 ICO whitepapers with more verifiable substance than this filing cabinet of air. Those were scams. This is physics โ and physics is less forgiving than any fraud.
The public sees a funding round. I see a probability distribution with a long left tail. The ledger doesn't lie โ but in this case, the ledger is empty. What remains analyzable is the structure of the bet: the investors, the timing, the founder's background, and the historical weight of the technology under siege. Structure dictates fate. That structure is where the truth lives.
Context
ASML holds 100% of the EUV lithography market. That monopoly began in the 1990s, when the company bet its survival on extreme ultraviolet technology that most of the industry considered a scientific dead end. It took roughly two decades and billions in failed prototypes โ early sources were too weak, optics absorbed too much light, the vacuum environment destroyed everything โ until the first high-volume production system reached TSMC in 2018.
The moat is not purely technological. ASML's supply chain is a fortress built from Zeiss's atomically precise mirrors and Cymer's laser-driven tin plasma sources. Its relevant customer base is exactly three companies โ TSMC, Samsung, Intel โ all of whom hold equity in ASML, join roadmap collaborations, and prepay billions for tool reservations. The installed EUV fleet has processed tens of millions of wafers. High-NA EUV, the next-generation leap for sub-2nm nodes, is already entering early production.
The economics are extreme. Each EUV tool sells for $150โ200 million. ASML ships roughly 50โ60 per year, generating about $10 billion in annual EUV revenue at gross margins near 50%. The company spends over โฌ4 billion annually on R&D โ more than three-quarters of Source Foundry's entire war chest.
The demand side is unprecedented. AI accelerators are the largest single source of incremental demand for leading-edge nodes; every major hyperscaler is locked into TSMC's 3nm for the next three years. CoWoS advanced packaging, not lithography, has been the flashpoint bottleneck for AI GPU supply โ but packaging is a downstream constraint. The upstream constraint, the one that caps how many advanced-node wafers exist at all, is EUV tool output. That is the pressure valve Source Foundry claims to target. A different, cheaper lithography path would expand the entire industry's capacity envelope, not merely substitute for it.
The challenger: a Bay Area stealth company, founded in 2025. Aschenbrenner, author of the influential "Situational Awareness" essay on AI and US-China competition, seeded it with $100 million, then followed with $400 million more even as his fund reportedly faced liquidity pressures. Sequoia Capital joined at significant scale. Combined capital: $500 million.
The motivation is transparent. AI compute demand is colliding with advanced-node capacity. NVIDIA, AMD, Google โ every significant AI player depends on TSMC's 3nm and 2nm nodes, which are capped by EUV availability. The chip shortage of the AI era is a lithography shortage. Challenging ASML is not technology hubris. It is strategic desperation with a balance sheet attached.
Core: The Teardown
The physics gap. Quantify it. ASML: twenty years from research inception to the first production EUV tool. Source Foundry: founded in 2025, sitting in concept-to-early-prototype stage. Under the most generous scenario โ validated prototype in three years, pilot customers in five, volume ramp in eight โ this company is one to two complete technology lifecycles behind the incumbent.
The validation gap is more damning. ASML's current NXE:3800E systems have demonstrated yield and uptime across millions of processed wafers. Source Foundry has published nothing: no resolution data, no throughput numbers, no resist sensitivity curves. In the lithography industry, silence is the loudest confession. The interval between laboratory demonstration and fab-class qualification is where 90% of advanced lithography concepts die. ASML nearly perished in that interval multiple times between 2000 and 2015, surviving on decades of DUV cash flow. Source Foundry has no such annuity. The money stops. The clock does not.
The capital equation. $500 million sounds like a war chest. In ASML terms, it is six weeks of the incumbent's annual R&D burn โ and ASML's R&D is incremental optimization of a working system. Source Foundry is funding an invention from zero. A conventional independent lithography effort requires $20โ50 billion across a decade. Source Foundry has raised 1โ2.5% of that. There is exactly one coherent explanation: the company believes its approach eliminates the complexity that drives ASML's cost base. A materials breakthrough in photoresist, mask technology, or light-source physics could collapse the requirement for the entire precision-optics stack. That is the only arithmetic that makes $500 million rational. Either that, or the check-writers are irrational. Both options remain on the table.
The supply chain trap. A modern lithography tool is a systems-integration marvel: laser-driven plasma sources, multilayer mirrors, vacuum stages positioned within a fraction of an atom, particle control beyond hospital standards, and orchestration software binding it together. Every subsystem is custom. Every supplier is locked into ASML's orbit. A new entrant cannot recreate that ecosystem in five years, or ten. The only viable strategy is circumvention: a machine that deliberately avoids the hardest components. This is why the materials-science angle matters. If Source Foundry's edge is a resist that works with cheaper, lower-numerical-aperture optics โ or a compact high-harmonic-generation source that skips the laser hall โ the ASML fortress becomes irrelevant. If they are building another optical projection tool, they are already irrelevant.
The name whispers the thesis. "Source Foundry" โ a foundry for a new kind of light. High-harmonic generation, firing femtosecond lasers into gas cells, has produced coherent EUV radiation in university laboratories for years. The hard problem is scaling brightness, repetition rate, and uptime to fab requirements. That problem has never been solved. It may be unsolvable. But it is the right battlefield for a materials scientist with $500 million and no respect for orthodoxy.
The alternatives, eviscerated. History is hostile to lithography revolution. Nanoimprint lithography, pursued seriously by Canon, stamps patterns mechanically and has never beaten defect economics at scale. Multi-beam e-beam direct write has the resolution but writes at glacial speeds. Directed self-assembly of block copolymers has the chemistry but lacks placement accuracy. Computational lithography optimizes the current regime; it does not escape it. None killed ASML. All had competent engineers, patient capital, and decades of grace. Source Foundry's differentiation can only reside at the unspoken intersection of materials and source physics. That is the best place to attack. It is also, statistically, the most likely place to die.
The customer concentration trap. The addressable buyer universe is three companies. TSMC, Samsung, and Intel are deeply entwined with ASML: equity stakes, joint development, prepayments, decades of process qualification. Their operations teams have zero incentive to pilot an untested machine that could erase billions in throughput while the industry's most reliable tool waits in the next bay. Physics can be conquered; corporate risk committees are entropic. Every fraction of yield risk maps to analyst guidance, stock price, and executive bonus. No operations executive loses their career by declining a discount on a novel lithography tool. They lose it by approving a pilot that fails. The incentive asymmetry is total. The only plausible early adopter is a greenfield fab โ likely a CHIPS Act-funded US facility whose mandate is supply security, not next-quarter margin.
The patent wall. ASML's intellectual property is not a wall; it is a mountain range. Tens of thousands of patents cover EUV source geometry, collector optics, pellicles, vacuum transport, alignment, metrology, process control. Any conventional leading-edge path means litigation the startup cannot fund. The only escape is a physically distinct route. I have seen this dynamic in DeFi, where protocols claimed composability while violating the invariants the market depended upon. The lesson transfers: when the incumbent owns the interface, innovation must happen outside the interface. Source Foundry's decision to lead with a materials scientist rather than an optical physicist suggests they understand this. Whether their understanding survives contact with the actual physics is unresolved.
The balance sheet autopsy. A $500 million raise against zero revenue and zero product implies a pre-money valuation in the $1.5โ2.5 billion range under conventional dilution assumptions. That valuation is not real. It is an option price on a tail outcome. A serious lithography development program burns $100โ200 million per year. Runway: three to five years. Without a flagship milestone in that window โ a working prototype, external validation, a pilot agreement โ the company raises again at depressed terms or dies. This is the anatomy of an all-or-nothing venture gamble. I have seen the same shape in crypto: leveraged into a single narrative, no covenants, no salvage value. Sometimes the leverage pays. More often, the ledger records the loss.
The geopolitical layer. Aschenbrenner's published thesis describes US AI dominance as conditional on breaking the Dutch single point of failure. ASML resides in the Netherlands. Dutch political dynamics govern every export-control decision on advanced lithography. For Washington, that is unacceptable strategic dependency. If Source Foundry succeeds, it creates a US-controlled alternative. Export controls become a US decision, not a Dutch negotiation. There is also the China dimension. Each side of the semiconductor decoupling now treats lithography as the decisive choke point. Washington restricts ASML exports indirectly through Dutch policy; Beijing spends billions attempting its own EUV program with no credible breakthrough. A US-controlled alternative would give Washington an export-control lever that is entirely its own โ no negotiation with The Hague required. That is a strategic asset priced far above any financial model. If Source Foundry fails, the loss is a rounding error in the context of the AI race. This is why Aschenbrenner, amid his own fund's liquidity stress, added $400 million. This was never a pure financial investment. It is insurance. Insurance is priced to absorb total loss.
Contrarian: What the Bulls Get Right
The skeptics' confidence masks a real asymmetry. ASML's moat is integrated complexity, not fundamental physics. The moment a simpler physical path exists, the moat evaporates. A materials-science thesis could obviate the entire Zeiss optical stack. You do not defeat the king by storming his castle; you raise a castle of different material.
Demand is real. The AI compute curve is exponential, advanced-node capacity is the binding constraint, and every EUV tool is sold out years in advance. A cheaper, faster, simpler machine โ even at modestly lower performance โ would find buyers among insurgent fabs and subsidized US facilities. The geopolitical tailwind is beyond dispute: Washington would subsidize domestic lithography for a decade if it meant ending dependence on Dutch export controls.
And the Sequoia data point cannot be rationalized away. Sequoia does not write nine-figure hardware checks on PowerPoint. Their diligence is the most disciplined in the industry. The inference is inescapable: they saw something physical โ lab data, a prototype, a proprietary material โ that is not in the public domain. That information asymmetry should make every confident skeptic hesitate.
Takeaway
The public sees the spark; I track the fuel lines. The fuel lines are coherent: a materials-science founder, a source-focused name, unprecedented conviction from smart capital, and a geopolitical imperative that reshapes the payout. This is the structure of a genuine paradigm attempt.
But the ledger doesn't lie. It records: no product, no data, no customers, and a capital base equal to six weeks of the incumbent's R&D burn. Based on my audit experience โ ICOs, DeFi collapses, stablecoin autopsies โ the modal outcome here is failure, with success probability under fifteen percent. Yet that tail, if it lands, rewrites the semiconductor order for fifty years.
Watch three milestones: published patent filings, lab data entering the public domain, a first pilot customer. If none appear by 2028, the structure will have made the decision. Structure usually does.