Keller and Zeloof's Indie Fab Bet, a Bottom-Up Challenge to Semiconductor Concentration

The semiconductor industry has two kinds of legends. There are institutional heroes — engineers whose names are buried in corporate org charts but whose work shapes every device we touch. And then there is Jim Keller, who somehow moved between AMD, Apple, Tesla, and Intel while leaving a defining architectural fingerprint at each stop. Zen, the A4, the FSD silicon — the list reads like a one-person portfolio of modern computing history. When someone with that resume decides to enter the fabrication business alongside Sam Zeloof, the engineer who famously built chips in his garage using refurbished 1970s-era equipment, the industry pays attention.

What makes this pairing unusual is not just the credentials. It is the direction. In 2026, the dominant conversation around semiconductor sovereignty runs through government budgets, trade policy, and billion-dollar fab construction deals. The European Chips Act has mobilized national treasuries. TSMC's Dresden facility is under construction. Intel is navigating its European roadmap with subsidy support. All of it is top-down: states deciding that chip manufacturing is too strategic to leave entirely to market forces, then writing enormous checks to attract or build the necessary capacity. Keller and Zeloof are pointing in the opposite direction entirely.

The Garage Logic Applied to a Larger Stage

Sam Zeloof's original garage project attracted attention precisely because it made the impossible seem merely difficult. Modern semiconductor manufacturing requires cleanrooms, extreme ultraviolet lithography systems costing upward of $200 million per unit, and process control at sub-nanometer tolerances. Zeloof's answer was to work with what he could actually access — older photolithography techniques, repurposed equipment, and an enormous amount of patience — and produce something that worked, even at feature sizes that Intel had left behind decades earlier.

The philosophical implication of that work was not lost on the engineering community. If the barrier to chip manufacturing is primarily capital and regulatory access rather than fundamental physics, then the concentration of advanced fabrication in three companies — TSMC, Samsung, and Intel Foundry — is a contingent historical fact, not an inevitable structural law. Indie fab is, in its most ambitious reading, an attempt to test that hypothesis at a credible scale.

Where this logic runs into friction is at the boundary between the philosophically possible and the economically viable. Advanced node fabrication below 7nm requires EUV lithography, and EUV requires ASML. There is no workaround, no substitute, no garage version of a system that weighs 180 tons and contains 100,000 individual components. Export controls backed by US and Dutch governments mean that EUV access is itself a geopolitical instrument, not simply a market transaction. An indie fab that opts out of EUV is, by definition, opting out of the front lines of logic scaling.

Where the Indie Thesis Actually Holds

The honest version of the indie fab argument is not that it will challenge TSMC at 2nm. It is that the semiconductor manufacturing stack is far larger and more diverse than advanced logic alone. Mature nodes — 28nm and above — still account for the majority of chips shipped globally by unit volume. Specialty processes for MEMS devices, photonics, power semiconductors, and biosensor chips are poorly served by fabs optimized for high-volume consumer silicon. Emerging material platforms — gallium nitride, silicon carbide, indium phosphide — represent genuine white space where a credentialed, nimble manufacturer could carve out defensible territory.

This is also where the Keller-Zeloof pairing makes most sense analytically. Keller's career has been defined not by incremental node shrinks but by architectural decisions that extract more value from a given process generation. If the indie fab thesis is "we design the process alongside the architecture" — optimizing co-design in ways that large fabs serving dozens of customers structurally cannot — that is a real differentiation argument. The opportunity is not to out-scale TSMC but to do something TSMC's customer-agnostic model makes impossible.

The capital structure, however, remains the central constraint. Even modest cleanroom facilities capable of processing wafers at legacy nodes require hundreds of millions of dollars in equipment, real estate, and operational overhead. Venture capital can absorb some of that, particularly with Keller's name on the pitch deck, but the timelines for fab-level returns are measured in decades, not the four-to-seven-year windows that most funds underwrite. The financial model has to be solved before the physics model matters, and solving it without government subsidy narrows the viable product space considerably.

2026 and the Semiconductor Sovereignty Paradox

The simultaneous emergence of top-down and bottom-up approaches to semiconductor sovereignty in the same calendar year illuminates a genuine paradox at the heart of the industry. Both movements are responding to the same underlying anxiety: the dangerous concentration of advanced manufacturing in Taiwan, the supply chain fragility exposed by the pandemic-era chip shortage, and the growing use of semiconductor access as geopolitical leverage.

But their implied theories of change are almost incompatible. The EU approach accepts the basic logic of capital concentration and tries to redistribute where that capital is deployed geographically. The indie fab approach questions whether concentration is structurally necessary at all, at least for significant portions of the manufacturing stack. One is playing the existing game on new terrain; the other is trying to change what the game is.

Neither approach is likely to fully succeed on its own terms within any near-term horizon. The EU's billions will not produce TSMC-equivalent process capability within a politically relevant timeframe, and the indie fab will not displace merchant fabs in any volume market. But if both survive long enough to generate institutional knowledge, trained engineers, and proof-of-concept data, the semiconductor ecosystem of 2035 may look meaningfully different from the one that existed before this wave of experiments began. That is the real wager Keller and Zeloof are making — not that they will win the fab race, but that running it changes what winning looks like.