Fab2
A “fab that mass-produces fabs” — Jim Keller and Sam Zeloof’s decentralized bet against Big Foundry orthodoxy
Fab2 (formerly Atomic Semi, Inc.) is a US semiconductor manufacturing-equipment startup founded in 2022, underwriting an unconventional thesis: mass-producing compact, modular fabrication plants using tools it builds itself. The company rebranded from Atomic Semi to Fab2 in July 2026 and relocated headquarters from California to Austin, Texas. From an underwriting standpoint, this is a textbook founder-led deep-tech situation in which personal reputational capital, rather than disclosed commercial traction, is doing the heavy lifting on early-stage pricing.
First came to industry attention for building lithography equipment in his family’s garage as a teenager. In 2018, at roughly 17, he produced the “Zeloof Z1” — widely regarded as the first homebrew lithographically fabricated microchip — and followed it in 2021 with the “Z2,” which reached roughly 100 transistors at approximately 300nm features. He co-founded Atomic Semi with Jim Keller in 2022 while studying electrical engineering at Carnegie Mellon University (2018–2022). Self-taught, hands-on fabrication expertise is the company’s core technical asset — but a CEO with a thin conventional operating track record is a variable worth watching as the organization scales.
One of the industry’s most credentialed chip architects, with stints spanning DEC’s Alpha processors, AMD’s K7, K8, and Zen architectures, Apple’s A4 and A5 chips (via the P.A. Semi acquisition), Tesla’s Autopilot silicon, and a term as Intel senior vice president. He currently also serves as CEO of AI chip company Tenstorrent Inc. — a dual-mandate that is a material governance overhang we flag explicitly in the risk section below.
Fab2’s self-defined category — “fab fab,” a fab that manufactures fabs — is a deliberate inversion of the batch-processing, 300mm-wafer model that defines the leading-edge foundry business. The company’s technical core is electron beam (EB) lithography, which patterns directly without a mask and targets prototype chips completed within hours. The trade-off is unavoidable: EB writing is materially slower than EUV exposure of full 300mm wafers, and is structurally unsuited to the hundreds-of-millions-of-units annual cadence run by TSMC or Samsung. We read the addressable market as effectively capped at prototyping, R&D, and low-volume specialty runs — not a displacement threat to leading-edge foundry capacity.
Builds every fab-constituent component in-house — pumps, valves, sensors, gas piping, nanometer-precision actuators, control boards — and mass-produces the assembled compact fabs themselves. EB-lithography-based, software-defined fabs pattern chips smaller than a standard wafer.
Rebranded from “Atomic Studio,” a browser-based collaborative electronic design automation (EDA) tool covering layout, schematic capture, and simulation — positioned to compress the design cycle in tandem with the hardware side.
Three-Site Footprint, Texas-Weighted: ① Austin — new R&D and production headquarters, roughly 120,000 sq ft (~11,000 m²). ② Lockhart — the core “fab fab” manufacturing site, roughly 30,000 sq ft (~2,800 m²). ③ San Francisco — the original “garage fab,” roughly 25,000 sq ft (~2,300 m²), retained from the company’s founding. Fab2 is investing roughly $3.9M converting an Austin warehouse into office and R&D space, with phase-two construction targeted for completion in February 2027. Four years operating out of California followed by a hiring-focus shift to Texas is an unambiguous relocation signal.
Competitive Landscape — Contrast With Terafab: Fab2’s Texas build-out runs in parallel with Tesla and SpaceX’s “Terafab,” a single mega-campus in the same Austin region targeting one terawatt of annual compute capacity, with total construction costs estimated up to $119 billion. Tom’s Hardware frames the two as “contrasting answers to the same question” of how the US should expand chipmaking capacity — concentrated mega-campus investment versus decentralized, replicable small fabs. The two models are not direct competitors on volume, but they represent competing capital-allocation bets on the same underlying policy question, in the same geography.
Fab2 has re-rated roughly 37x in the roughly three years between its 2023 seed and its September 2026 Series A close. With no disclosed revenue or ARR in public sources, we read this re-rating as driven predominantly by founder-brand premium — particularly Jim Keller’s standing — layered on top of macro capital flows chasing the “US semiconductor reshoring” theme, rather than by underwritten commercial metrics.
First institutional round, led by the OpenAI Startup Fund, raising approximately $15M at an approximate $100M valuation. Angel participation from Naval Ravikant, Nat Friedman (former GitHub CEO), and Fred Ehrsam (Coinbase co-founder) signals early-stage capital drawn primarily from founder-adjacent, high-signal networks rather than traditional deep-tech specialists.
A $500M Series A led by Fundomo values the company at $3.7B — roughly a 37x step-up from the seed. The syndicate is unusually diverse for a deep-tech round: Corner Capital, Paradigm, Duquesne Family Office, Maverick Silicon, StepStone Group, Protagonist, and UDC Ventures joined institutional-side, with seed angel Naval Ravikant participating again. The blend of crypto/macro-adjacent capital (Paradigm) alongside traditional PE and family-office money (Duquesne, StepStone) is notable — it reads as thematic capital rotation into “American manufacturing reshoring” narratives as much as a conviction underwrite of Fab2’s specific unit economics.
Funding figures in this section are sourced from press coverage of a private company (WOWTALE, Tom’s Hardware, BigGo Finance, among others) and are not confirmed against SEC filings or audited financial statements. Reported Series A investor rosters diverge across sources — PitchBook’s company profile separately references Beyond Earth Ventures, Corner Capital Management, Eclipse Capital, Permutation, and Bogaroo as 5 of 13 total investors, a list that does not fully reconcile with the WOWTALE-sourced syndicate cited above. The precise lead-investor attribution, final cap table composition, and valuation methodology remain unconfirmed by the company; we present the divergent sources side by side rather than silently reconciling them.
Fab2’s positioning rests less on demonstrated technical differentiation than on founder credibility and a deliberate carve-out into a niche that leading-edge foundries structurally cannot serve. This is a classic early deep-tech risk/return profile: a proven founding team underwriting an unproven business model.
Jim Keller’s tenure across DEC, AMD, Apple, Tesla, and Intel, combined with Sam Zeloof’s proven homebrew-lithography track record, is the dominant factor explaining a 37x re-rating absent disclosed revenue. This is founder-premium pricing in its purest deep-tech form.
In-house design and manufacture of pumps, valves, lithography tools, and vacuum systems could, in theory, deliver a cost and lead-time edge over the traditional equipment supply chain (ASML, Applied Materials). The flip side is meaningful execution risk from having to run multiple complex subsystems in parallel with no outsourcing safety valve.
TSMC and Samsung are optimized for hundreds-of-millions-of-units annual production and structurally cannot serve rapid, hours-scale prototyping demand. Fab2 targets that white space directly, sidestepping head-to-head competition with mega-campus projects like Terafab.
Multi-strategy institutional participation from Paradigm, StepStone Group, and Duquesne Family Office in the $500M Series A suggests reasonably strong access to follow-on capital for what will be a capital-intensive build-out phase.
We view Fab2 as a high-risk, high-return deep-tech position defined by the combination of a credentialed founding team, an unproven business model, and near-total financial opacity at this stage. The following risk factors are included as standard analytical practice, whether or not each was explicitly flagged in source material.
EB lithography’s mask-free, direct-write process is materially slower than EUV exposure, which structurally excludes Fab2 from TSMC/Samsung-scale volume production. This effectively caps the addressable market at prototyping and R&D niches — a real constraint on long-term TAM, not merely an early-stage limitation.
“Mass-producing replicable fabs” has no industry precedent, and public sources show no disclosed revenue, named customers, or production contracts. The current valuation prices in customer acceptance of an unprecedented business model before that acceptance has been demonstrated.
Co-founder Jim Keller simultaneously serves as CEO of a separate AI chip company, Tenstorrent Inc. With both companies in capital-intensive hardware scale-up phases, a potential conflict in management bandwidth and decision-making focus cannot be ruled out.
A well-funded cohort of rivals is chasing the ASML/TSMC incumbency from adjacent angles: Substrate (roughly $100M seed led by Founders Fund and others, $1B+ valuation, particle-accelerator X-ray lithography), xLight (free-electron-laser EUV source, $40M Series B plus a CHIPS Act loan letter of intent), and Source Foundry ($500M from Situational Awareness at a $5B valuation). Fab2’s technical differentiation may prove less durable than the current mark implies given this competitive density.
Running parallel in-house development across pumps, valves, lithography, and vacuum systems creates cascading schedule risk from any single subsystem delay. Fixed-cost exposure across three simultaneous sites (Austin, Lockhart, San Francisco) compounds the burn-rate profile ahead of any confirmed revenue.
As noted in Section 03, Series A investor rosters diverge across sources, and no audited financials or SEC filings exist to independently verify the valuation. There is no disclosed operating metric that supports the 37x step-up from seed to Series A.