Summary

Substrate is a semiconductor moonshot that founder James Proud has described as having 1% odds, while its potential payoff has become large enough to attract financing. The company claims particle-accelerator-powered X-ray lithography could halve manufacturing costs, eliminate some multi-patterning, and produce high-resolution layers comparable to those of leading foundries. Ben Thompson remains explicitly uncertain—“It’d be nice to see more evidence”—but highlights Proud’s posture of operating with “100% conviction” despite the long odds.
Trying to disrupt ASML and TSMC simultaneously sounds absurd, yet the industry’s integration makes that strategy internally coherent. Thompson argues that lithography is the central step around which the entire fabrication process is optimized, so a radically different tool probably cannot simply be dropped into an incumbent fab. Substrate can buy conventional etching, coating, and doping equipment, but probably must rebuild how those pieces work together “from first principles.”
ASML and TSMC’s co-evolution created formidable incumbent lock-in. Lithography now represents perhaps 30%-35% of fabrication cost; Thompson roughly estimates current machines near $250 million and the next generation closer to $500 million, while stressing that he is “pulling these numbers out of my rear end.” A roughly $30 billion fab that is not filled and running leaves its owner in serious trouble, making TSMC “almost the brake on the AI bubble” because it bears enormous upfront equipment risk.
The strongest defense of today’s AI bubble is not durable GPUs but coordinated invention across many improbable projects. Big Tech spending flows through NVIDIA to TSMC and ASML, while simultaneously making outsiders such as Substrate financeable: “A lot of people working on a lot of hard problems all at the same time.” Thompson expects excess and failures, yet argues that the future will depend on what “emerges from the rubble” and could not otherwise have been attempted.
A thermodynamic chip promising theoretically 10,000-times-more-efficient image generation is the episode’s best specimen of bubble-enabled experimentation. Its proposed mechanism excites a chip according to a probability distribution, then uses the system’s physical decay to compute probabilistic outputs. It sounds “totally insane,” but Thompson sees a conceptual fit between probabilistic AI and probabilistic hardware that could better align computation with probabilistic outputs than today’s binary, deterministic chips.
Semiconductor controls, ideological ambition, and AI mania are all forcing functions against industrial stasis. Thompson says a new US foundry would otherwise go bankrupt “100 times over” while descending the learning curve; chip controls instead motivate China to absorb uneconomic development costs, while without them TSMC could “rule the roost forever.” Breaking that equilibrium may require “religious fervor”: Substrate is ideological about the US winning, while AI companies act as if they are “inventing God.”

Deep dive

1. Substrate’s X-ray wager is plausible enough to examine, not proven enough to endorse

Andrew Sharp frames Substrate’s claim precisely: channel light from a particle accelerator through a car-sized tool, print layers comparable to those of leading foundries, and cut semiconductor manufacturing costs in half. Thompson’s answer to whether it has legs is deliberately blunt: “I don’t know.”
X-ray lithography itself is not new. EUV reduced the width of light to 13 nanometers, but still requires extraordinary mirrors and sometimes multi-patterning; every additional pass adds complexity and cost. X-rays are theoretically thinner, potentially removing that repetition and lowering costs.
The unresolved questions are load-bearing: whether Substrate has achieved a consistent X-ray source, whether radiation disrupts other fabrication steps such as doping, and whether problems that confronted the approach in the 1980s and 1990s have genuinely been overcome. An old idea becoming viable 40 years later is “totally plausible,” not established.
Proud himself reportedly told investors there was a 1% chance of success but said he would operate with 100% conviction; he suggested the percentage had risen somewhat as breakthroughs emerged. Sharp found him more credible for frankly acknowledging the odds.
Thompson acknowledges criticism that his Proud interview lacked technical pressure, but says the relevant details were never going to be disclosed: “To the extent there’s something real, it is obviously a massively valuable secret.”

2. ASML won by optimizing the system, not merely selling a better machine

Thompson traces ASML’s ascent against Canon and Nikon to its partnership with TSMC, contrasting that relationship with Nikon’s alignment with Intel. “The rise of TSMC is the rise of ASML, and vice versa.”
The shift from 200-millimeter to 300-millimeter wafers captures the difference. Intel could tolerate less attention to throughput and yields because its monopoly margins covered inefficiency; ASML and TSMC redesigned the process together so larger wafers moved fast enough to deliver their intended volume gains.
Lithography now accounts for roughly 30%-35% of the cost, according to Thompson, while foundries are in the $20 billion range and the next generation may cost $30 billion-$50 billion. He roughly put current machines near $250 million and the next generation near $500 million, while warning that he was “pulling these numbers out of my rear end.” A $30 billion fab that is not filled and running leaves its owner in serious trouble, which is why he called TSMC “almost the brake on the AI bubble.”
A Taiwan earthquake—2001, Thompson believes—became another inflection point: ASML helped restore TSMC’s machines within weeks, while Japanese equipment took months. ASML later did the difficult commercialization work on EUV and became the only supplier capable of making those machines.

3. Replacing lithography may require rebuilding the foundry around it

Sharp’s pushback is the obvious one: Proud is proposing to disrupt both ASML and TSMC, despite two decades of failed challenges to either company. “That seems very ambitious.”
Thompson’s rebuttal is architectural. Lithography is the most expensive, complex, precision-dependent fabrication step, and every surrounding process has been tuned to it—much like the Windows-Intel partnership. Substrate need not reinvent etching or doping equipment, but radically changing the center means reintegrating everything around it.
Incumbent strength may create the opening. TSMC and Intel possess enormous expertise in the existing method, yet are “pot committed” to it; Thompson sees little chance that TSMC could seriously pursue a wholesale alternative while meeting current demand. Dominance makes a completely new approach more viable because “there’s not gonna be a response.”

4. Bubble economics coordinates inventions that normal markets would reject

Thompson used to associate bubbles primarily with waste and reputational danger. Something like Pets.com exists today, but it took 20 years, a recession, and a “cold frost ice age in tech” for the underlying model to emerge.
The conventional productive-bubble payoff is stranded infrastructure: bankrupt railroad builders still leave tracks, failed telecom companies leave dark fiber, and power investment can leave plants operating for 50 or 100 years. GPUs lack that longevity, creating a harder question about what today’s durable residue will be.
His newer framing, drawn from Byrne Hobart’s Boom, is “coordinated creation, coordinated innovation, coordinated invention.” Tech companies’ tens or hundreds of billions flow into NVIDIA, then TSMC and ASML, while the scale of the opportunity supplies capital for many interdependent experiments whose successes might justify many failures.
The thermodynamic-chip example carries the argument: excite a chip according to a chosen probability distribution, let it decay toward neutral energy, and use that physical process for computation. The claimed theoretical result—image generation 10,000 times more efficiently—might fail, but “in a normal period, no one” would fund the attempt.

5. A probabilistic technology cycle demands probabilistic analysis

Thompson expects the mania to break: “At some point we’re gonna say, ‘Wow, we got a little crazy then.’” He nevertheless thinks it is “almost certainly” true that valuable technologies will emerge from the wreckage only because the bubble funded them.
That changes his own analytical posture. Stratechery’s first decade offered relatively deterministic chains—Aggregation Theory, stable Big Tech lanes, A-to-B-to-C reasoning—but today he is attempting “deterministic analysis on top of probabilistic outcomes.” Whether Proud succeeds is unknowable; whether it is important and valuable that someone is attempting it feels much clearer.
The geopolitical version of the same mechanism is chip controls, which motivate China to absorb development costs that ordinary economics would reject. Substrate’s declaration that it is “an ideological company” and thinks the US should win invites demands for technical proof, but Thompson argues that difficult breakouts can require the same “religious fervor” driving AI companies to spend as though they are “inventing God.”

Character.AI’s Teen Chatbot Crackdown + Elon Musk Groks Wikipedia + 48 Hours Without A.I.Robinhood: Mobile First, Margins Later - [Business Breakdowns, EP.233]