Control electronics & cryo-CMOS

Every funded quantum roadmap, regardless of qubit modality, terminates in the same place: racks of arbitrary waveform generators, readout digitizers, and FPGA feedback engines, typically several RF/baseband channels per qubit at four-to-five figures per channel. The control stack routinely costs more than the quantum chip it drives, and unlike the chip it scales linearly — the superconducting I/O bottleneck literature is blunt that cabling, heat load, and the footprint, cost, and calibration burden of rack-scale instrumentation become prohibitive past a few thousand channels.

Who owns it. A five-vendor merchant oligopoly, so settled that when NVIDIA launched NVQLink in October 2025 it named exactly five control partners: Keysight, Quantum Machines, Qblox, QubiC, and Zurich Instruments (NVIDIA newsroom). Quantum Machines says its systems run at more than half of all companies building quantum computers. Keysight's QCS sits inside Fujitsu/RIKEN's 256-qubit machine and anchors the world's largest commercial control install (1,000+ qubit support) at Japan's AIST G-QuAT center. Zurich Instruments (Rohde & Schwarz) answered in March 2026 with ZQCS, promising 1,000+ channels per rack. The moat is real: sub-nanosecond synchronization across thousands of channels, microsecond-latency error-correction feedback, and years of co-calibration against a customer's specific qubits. Swapping control vendors means re-bringing-up the entire machine — switching costs compound with qubit count.

What breaks it. Physics. You cannot run a million coax lines into a dilution refrigerator, so control must eventually move inside the cryostat. Intel owns the most credible cryo-CMOS silicon: Horse Ridge II drives up to 16 spin qubits from 4 K on a 22nm FinFET logic process and was deployed with the 12-qubit Tunnel Falls system at Argonne in January 2026. If qubit makers integrate control on-die, the merchant rack market gets disintermediated exactly when it should inflect. The counterweight is brutal power math — millikelvin stages tolerate only picowatt-scale dissipation per channel (order-of-magnitude tens of picowatts in the multiplexer literature; demonstrated mK devices report ~200 pW static), orders of magnitude below today's cryo-CMOS — so room-temperature racks plus multiplexing keep winning for at least the next five years, and the incumbents are co-opting the threat (all five joined NVQLink; Zurich Instruments is already running GPU-coupled QEC demos with IQM).

Be honest about size. The merchant control market is plausibly low hundreds of millions of dollars today. Quantum Machines has raised $280M lifetime; Qblox's Series A was $26M; Keysight doesn't break quantum out of ~$5.4B of revenue, which tells you it's a sliver. The chokepoint is real while the market is small — every roadmap routes through it, but listed-equity exposure is diluted (KEYS, FORM, NVDA) or private.

Repricing catalyst. DARPA moved 11 companies into QBI Stage B in November 2025; the yearlong stage requires each finalist to detail a credible, risk-mitigated R&D plan toward utility scale by 2033 — which should surface whose control electronics each architecture leans on (our inference; DARPA mandates the plan, not vendor disclosure). Layer on the NQI Reauthorization advancing through Congress with an explicit quantum supply-chain-resilience mandate, and H2'26 is when Washington starts mapping this choke by name. The scarcity of investable pure-plays is itself the trade: the cleanest exit for this oligopoly is strategic M&A by a KEYS-class acquirer — likely the first moment public investors get direct exposure.