A household freezer does not earn its keep through a triumph of thermodynamics every time it preserves dinner. It earns it through maintenance: a compressor, a thermostat, a door seal and an electricity bill. Quantum computing has arrived at the same industrial stage. The compelling laboratory result is no longer the difficult object; the difficult object is a machine that stays cold, stays calibrated and stays coherent long enough to finish a job no classical server can finish at any price.
Two developments define that threshold. Google's Willow processor demonstrated surface-code error correction below threshold in research published in Nature in December 2024, while NIST finalized its first three post-quantum cryptographic standards in August of that year, converting a research forecast into a procurement obligation.
The Scarcity Has Moved Down the Stack
Industry coverage counts qubits the way early aviation counted horsepower. In a superconducting system, the constrained input is not the transmon; it is the millikelvin budget and the coaxial count. A dilution refrigerator delivers on the order of hundreds of microwatts at its mixing chamber, and every added qubit draws heat down a wiring harness that must also fit inside a vacuum can.
The consequence for the category is that capital is pricing the wrong asset. Headline qubit count is a derived metric. The scarce assets are refrigeration capacity, low-noise amplification, RF routing and the calibration labor that holds a device inside its error budget. Value will consolidate around the plumbing — dilution systems, cryo-CMOS control, packaging and real-time decoding — while bare processor ventures compete on specifications their suppliers effectively set.
Eight Years to Make an Ocean Pay
The Atlantic telegraph cable of 1858 operated for roughly three weeks before the line died. The idea had been proved; the insulation, the conductor and the receiver had not been industrialized. A durable transatlantic service waited until 1866, and William Thomson's sensitive mirror galvanometer mattered as much as the cable itself.
The parallel is uncomfortable for this valuation cycle. A logical qubit that outlasts its physical components is the 1858 moment, not the 1866 moment. The intervening years were not wasted physics; they were materials, splicing, testing and maintenance. Buyers who treat a threshold-crossing result as a delivery date repeat a nineteenth-century error. Investors who wrote off the entire category after the first failure missed the years when the economics actually arrived.
What a Twofold Error Reduction Actually Buys
Primary-source finding: In Google's Nature surface-code experiment, each increment in code distance suppressed the logical error rate by roughly a factor of two — the point at which redundancy becomes a remedy rather than a liability.
Read that as an accounting rule, not a speed claim. The unit a buyer should procure is logical volume: encoded qubits multiplied by the circuit depth they sustain at a stated fidelity. A machine with more physical qubits and a worse suppression factor can be the less capable machine. Trapped-ion results reported by Oxford's physics group in 2025 extended the fault-tolerance claim from storing an encoded qubit toward executing a protected algorithm, which is precisely why platform comparisons are escaping the qubit-count frame.
The Counter-Case to the Refrigerator Thesis
The hardware-first reading has a genuine weakness. Resource estimates are functions of algorithms as well as devices: better arithmetic circuits, magic-state preparation and distillation have repeatedly reduced the projected cost of the same factoring or simulation task without a single new fabrication run. Classical progress cuts the other way, with tensor-network and Schrödinger-method simulators repeatedly raising the size at which a quantum device must establish advantage. A plan assuming hardware scaling alone will produce a commercial workload is as fragile as one assuming the workload is already proven.
The Invoice Arrives Before the Machine
The second market is already billing customers, and it does not require a qubit. NIST's FIPS 203, 204 and 205 standardized ML-KEM, ML-DSA and SLH-DSA; federal planning under National Security Memorandum 10 is organized around 2030 and 2035 migration milestones.
Primary-source specification: FIPS 203 defines a 1,184-byte encapsulation key and a 1,088-byte ciphertext for ML-KEM-768. FIPS 204's ML-DSA-87 produces a 4,627-byte signature, against roughly 64 bytes for a P-256 ECDSA signature.
That is not a configuration flag. Hybrid TLS handshakes grow, path MTU behavior shifts on constrained and satellite links, hardware security modules encounter key-slot and RAM limits, and code-signing pipelines must absorb signatures an order of magnitude larger. Certificate-transparency logs and embedded devices with fixed flash budgets are the quiet casualties. This is a hardware and protocol exercise wearing a cryptography label.
Panic Is Not a Migration Plan
The contrary case deserves equal weight. Vendors profit from describing every session token as a quantum casualty, yet most firms do not hold data whose useful life outlasts a cryptanalytically relevant machine. Exposure concentrates in health records, geophysical surveys, signals intelligence and infrastructure secrets — information captured now and decrypted later.
Primary-source estimate: Gidney and Ekerå's 2019 calculation factored a 2,048-bit RSA integer using an estimated 20 million noisy qubits over eight hours — a conditional result, not a shipment date.
Hybrid constructions also retain the classical security margin if the lattice component disappoints. Inventory first; forklift upgrades second.
What to Buy, and What to Defer
- Inventory by data lifetime, not system criticality. Thirty-year infrastructure secrets justify urgency; a rotating session token does not.
- Test the hybrid handshake on your worst link. Measure fragmentation, handshake latency and HSM behavior before standardizing a suite.
- Buy algorithmic agility. Pair a lattice key-establishment path with a hash-based signature fallback for long-lived artifacts, accepting SLH-DSA's slower signing and larger payloads.
- Demand a logical-volume curve from quantum suppliers, tied to a named workload and a classical baseline. Reject bare qubit counts.
- For individuals, prefer services already publishing hybrid post-quantum key exchange; treat "quantum-proof" as marketing, not certification.
Six Months Out: A Market for Reliable Operations
By the spring of 2027, credible procurement documents will specify logical qubits, circuit depth and decoded throughput rather than raw device size. Real-time decoding and cryogenic control will be the constrained line items. Enterprise budgets will tilt from quantum pilots toward cryptographic migration, because one deadline is contractual and the other is a roadmap.
Dispersion will widen. A handful of full-stack programs will begin resembling utilities, selling a guaranteed error budget; most point solutions will resemble components, priced against incumbent instrumentation. The category will not have been disproved. It will have acquired an invoice, a maintenance schedule and a procurement department — which is how infrastructure becomes real.