Qubit fabrication & foundry

Superconducting and photonic qubits don't come out of standard fabs. They need angstrom-level Josephson-junction deposition, loss-obsessed surface engineering, ultra-low-loss waveguides, integrated single-photon detectors, and exotic electro-optic films — barium titanate, thin-film lithium niobate — bonded onto 300mm CMOS. Production-grade versions of those process lines exist at roughly two US merchant foundries, and both just got marked to market.

Who owns the choke. GlobalFoundries manufactures PsiQuantum's million-qubit-targeted Omega photonic chipset on the Fab 8 300mm line in Malta, NY, runs Diraq's cryo-CMOS and spin-qubit work on 22FDX, and in May 2026 formalized the franchise as Quantum Technology Solutions — a dedicated quantum unit launched with a signed Letter of Intent for $375M of proposed CHIPS Act funding (under which Commerce will take ~1% equity once finalized) and a partner list that reads like the industry: Google Quantum AI, Microsoft Quantum, NVIDIA, PsiQuantum, Quantinuum, Quantum Motion. One nuance on the exotic films: PsiQuantum manufactures its 300mm barium-titanate wafers at its own California facilities, then integrates them with GF CMOS — part of the capability is captive to the customer, not the foundry. The other line that matters is SkyWater's DoD-Trusted Minnesota fab, which has fabricated D-Wave's qubits for over a decade — including the chips behind D-Wave's March 2025 quantum-supremacy claim — has run 200mm development and photonics work for PsiQuantum since 2023 (the production-scale Omega ramp sits at GF Fab 8), and is co-developing QuamCore's superconducting SFQ controllers. Rigetti's Fab-1 is the only other merchant-accessible dedicated US qubit fab — Google's Santa Barbara line and AWS's Ocelot facility are captive — and it mostly serves Rigetti.

Why the moat holds. Qubit lines are qualified in coherence times and photon loss, not yield curves, and that data is tacit — it lives inside decade-long co-development loops between foundry process engineers and qubit designers. Layer on trusted-foundry accreditation for government work and the economics of running specialty modules at sub-scale volumes, and a challenger needs years of money-losing co-development plus an anchor tenant that doesn't exist: every credible anchor (PsiQuantum, Diraq, Quantum Motion, Google) is already attached to GF or SkyWater. imec runs comparable flows but is neither American nor production-scale; IBM and Intel keep internal lines that serve nobody else.

Be honest about the size. Industry-wide quantum foundry revenue is still measured in tens of millions of development dollars, not wafer ramps; neither GF nor SkyWater discloses a quantum line item. The chokepoint is real while the market is small — which is exactly why scarce accredited capacity gets bought rather than competed with. IonQ proved it in January 2026, agreeing to acquire SkyWater for $35.00 a share, roughly $1.8B in cash and stock; shareholders approved on May 8 and close is expected by Q3 2026. The only listed near-pure quantum foundry is being taken off the board at a strategic premium.

What forces the next repricing. Three things stack in H2 2026. GF's first post-launch earnings prints (Q2, early August) start putting bookings and capacity commitments against a quantum unit that is currently a press release backed by federal equity. DARPA's Quantum Benchmarking Initiative moves its 11 Stage B teams toward Stage C decisions by late 2026 — funded fault-tolerant prototype builds for the winners, with the silicon-spin and photonic teams (Diraq, Quantum Motion, Xanadu, Photonic) all dependent on advanced-CMOS foundry partners. And the National Quantum Initiative Reauthorization is through committee in both chambers, queuing five more years of federal demand. Own the scarcity, not the revenue: GFS is the foundry with the government signing onto its cap table, SKYT is deal paper converting into IONQ, and AMAT's quantum toolchain collaborations are free optionality, not earnings.