How to Tell a Quantum Chokepoint From an Inconvenience

Helium-3 turns up on nearly every quantum sovereignty briefing as a chokepoint, and the reasoning looks sound. It is produced as a decay product of tritium, the supply is tied to a small number of national programs, and nothing substitutes for it inside a dilution refrigerator, the machine that holds superconducting and spin qubits below 10 mK. All of that is accurate. It is also, for most quantum programs, the wrong item to put at the top of a risk register. In a modern dry dilution refrigerator the helium-3 circulates in a sealed loop, gets pumped back into storage at the end of a cooldown, and is used again on the next one. You buy a charge once, with the machine. Losses come from leaks and from mistakes during warm-up, not from running experiments.

That reclassification changes what you should do about it. A consumable needs a stockpile, an annual budget line, and a supplier relationship someone renews. A one-time capital charge needs a leak-check procedure, a recovery protocol, trained hands during every warm-up, and one spare bottle in a locked cabinet. Same underlying scarcity, two entirely different responses, and only one of them is worth what it costs.

Most dependency reviews we are asked to comment on don’t make that distinction. They rate items by how rare the item is and how many suppliers exist, which produces a page of red, and a page of red prioritizes nothing. What follows is the set of five questions we teach for separating a genuine chokepoint from an expensive nuisance, and what each answer changes about the work that comes next.

The all-red review

Rarity is the wrong primary axis, and it fails in both directions.

It over-flags. Enriched isotopes, exotic crystals, and single-vendor instruments all light up red, the review circulates, and the mitigation budget gets spread thin across a dozen items rather than concentrated on the two that would actually stop delivery. Everything gets a watch item, an owner, and a quarterly slide. Nothing gets fixed.

It also under-flags. The dependency that halts a program is often unremarkable on a scarcity axis: a wire-bonding step that only one contract shop performs to the tolerance you need, a calibration service written into the acceptance criteria, a firmware signing key held by a vendor, a specific low-outgassing epoxy. A part can be cheap, catalog-listed, and available from four distributors, and still stop you cold because the only engineer who can install it correctly works for the company that sold it.

The discriminating question is what happens to the mission if the item is withheld tomorrow, and how long until you are running again.

Two definitions, and the space between them

A chokepoint is a dependency whose loss stops delivery of the mission, with no substitution path that completes inside your decision horizon. Both conditions are required. Loss without a stop is an annoyance. A stop with a nine-month workaround, on a program with three years of runway, is a schedule problem with a price attached.

An inconvenience is everything else that hurts. It costs money, it slips dates, it forces a redesign, it makes a program officer’s month unpleasant. What makes it an inconvenience is that a path exists and someone can put a number on it.

Expense is not a third category. A component that costs four times what it should is a budget conversation, and budget conversations have known participants and known outcomes. Treating price as a sovereignty risk is how reviews end up recommending domestic production of things that are merely overpriced.

The genuinely hard cases sit at the boundary and move across it. A dependency that is an inconvenience at three machines becomes a chokepoint at thirty, because supplier capacity rather than supplier existence turns into the binding constraint. The classification is valid for a scale, a mission, and a date, and it needs redoing when any of the three changes.

Five tests

Substitution

Ask at three levels, in order, and stop at the first yes.

Same function, different vendor: can another supplier’s part drop in, and what does acceptance cost? Different part, same function: is there another way to do this job with hardware you can obtain? Different architecture: can the system be built so the requirement disappears?

Cryogenics answers no, no, and yes. At the millikelvin stage there is no functional substitute for dilution refrigeration. Adiabatic demagnetization refrigerators reach comparable temperatures in bursts rather than continuously, and pulse-tube coolers alone stop several orders of magnitude short. But a trapped-ion or neutral-atom machine never needs those temperatures at all, so the requirement vanishes at the architecture level.

That third answer counts only if you can still act on it. Architecture substitution is a five-year decision made once, near the start. For a program already committed to superconducting qubits, the honest answer to the substitution test is no.

Independence

Count suppliers, then count the layer underneath them. Two vendors buying the same subsystem from one upstream source are one supplier wearing two logos.

Dilution refrigerators are built by Bluefors, Oxford Instruments, Leiden Cryogenics, ICEoxford, Maybell Quantum, and Zero Point Cryogenics, among others. That reads as a competitive market. Every dry system among them needs a pulse-tube cryocooler for its 4 K stage, and that market has far fewer entrants. Bluefors, the largest of the fridge vendors, announced its acquisition of Cryomech, one of the main cryocooler makers, in December 2022 and completed it in May 2023. Count at the wrong layer and a concentrated dependency looks diversified.

Independence also has a legal dimension and a knowledge dimension. Two vendors inside the same export-control jurisdiction fail together the day the rule changes. Two vendors whose lead engineers trained in the same three laboratories will hit the same technical wall at the same time. Genuine independence means separate upstreams, separate jurisdictions, and separate engineering lineages.

Recovery time against your horizon

This test is a comparison, never an absolute, and it is the one most often skipped.

Recovery time means time to working, not time to delivered. For a cryogenic system, that includes crating and shipping, site preparation, wiring, the first cooldown, thermometry calibration, and acceptance testing. For a second-source component it includes requalification, the engineering work needed to prove the substitute performs inside spec in your system rather than the vendor’s. Requalification is frequently longer than procurement and is almost never on the schedule.

Then set that against the decision it feeds. A fourteen-month recovery on a machine needed for a demonstration in nine months is a stop. The same fourteen months against a five-year build-out is a purchase order placed early. So the identical item earns different ratings in two programs, and both ratings are correct.

Rather than quoting a market-wide lead time, ask your vendor for the current figure in writing, and record it next to the date of the milestone that depends on it. The gap between those two numbers is the rating.

Jurisdiction, or who can say no

The question is not where a thing is made, it’s who has the authority to stop it moving.

Licensing authority follows design origin, controlled content, and tooling, not the factory postcode. A subsystem assembled domestically under a foreign license, using foreign-origin controlled parts, remains a foreign dependency, and the review that marks it green because the building is local has recorded the wrong fact.

Controls in this area attach to specifications rather than to product categories, which produces an effect procurement teams find counterintuitive: a supplier can be open to you at one performance level and closed at the next, so a capability that was freely importable during prototyping becomes licensable at production spec. The US Bureau of Industry and Security published an interim final rule covering quantum computers and associated equipment in September 2024 (89 FR 72926), with several partner governments moving on similar lines. Anyone rating a cryogenic or control-electronics dependency should verify the current classification for the exact configuration on the purchase order, not the family it belongs to.

Cloud access raises a jurisdiction question before it raises a technology one. Germany’s IBM Quantum System One at Ehningen, operated in partnership with the Fraunhofer-Gesellschaft, sits on German soil and is still built and serviced by IBM. Which of those two facts governs depends on the worry: physical access and data residency point one way, long-term maintenance and spare parts point the other.

Regeneration

If you lost the capability tomorrow, could money rebuild it, and on what timescale?

Some capabilities are purchasable. Inventory, tooling, test equipment, and service contracts all regenerate at the speed of a procurement cycle. Some do not. A team that has taken the same class of machine through dozens of cooldowns has knowledge that exists nowhere in writing, and no amount of budget compresses the years it took to acquire. Hiring accelerates; experience doesn’t.

Apply the test to documentation as well as people. If your operating procedures are a vendor’s manual plus four people’s memories, the regeneration answer is worse than the org chart suggests.

Reading the five answers together

Substitution and recovery time decide the classification. Fail both and you have a chokepoint. Fail one and you have a constraint that inventory, contract terms, or schedule float can absorb.

The other three change how you respond rather than whether it’s a chokepoint. Independence and jurisdiction set the probability you ever get tested. Regeneration sets what your fallback costs and how long it stays a problem after the immediate crisis passes.

Test	Reads as chokepoint	Reads as inconvenience
Substitution	No functional alternative at any layer you can still reach	Alternative exists at a performance or cost penalty you can quantify
Independence	Several vendors, one upstream, one jurisdiction	Genuinely separate supply paths and engineering lineages
Recovery vs horizon	Time to working exceeds the decision that depends on it	Absorbed by schedule float or existing inventory
Jurisdiction	One authority can halt it, and your specification sits near a control threshold	Uncontrolled, or controlled by an authority with predictable licensing
Regeneration	Capability lives in people and takes years to rebuild	Capability is documented, tooled, and purchasable

The dilution refrigerator, run through all five

Substitution. None at the millikelvin stage for a superconducting or spin-qubit program. Architecture-level substitution exists and is unavailable to a program already committed. Fail.

Independence. Six or more vendors at the top layer, materially fewer at the cryocooler layer beneath them, and most of the vendors clustered in a small number of allied jurisdictions. Weak, with the weakness one level down from where reviews usually look.

Recovery time. Long relative to research schedules, and longer than the delivery quote suggests once installation, wiring, and first cooldown are counted. Get the number from the vendor and set it against the milestone.

Jurisdiction. Cryogenic systems appear in dual-use control lists in multiple jurisdictions, and classification depends on configuration. Verify per purchase order.

Regeneration. This is where most reviews stop too early, because acquisition risk feels like the whole risk. Owning the machine is not the same as being able to keep it running. Who replaces a failed heater? Who finds and repairs a leak in a mixing chamber line? Who recalibrates thermometry after a component swap? If every answer is a vendor engineer on a plane, then the operating dependency is a separate line item from the purchase, and it stays open long after the acquisition risk closes.

The verdict is two chokepoints, not one, and they take different mitigations. Acquisition is a chokepoint for any program that must field hardware inside two or three years, and it responds to early ordering, a second qualified supplier, and a spare unit. Maintenance is a chokepoint on a longer clock that no stockpile touches, and it responds only to cryogenic engineers you employ and keep. A program that hedges the first and ignores the second has bought a machine it can’t service.

Talent fails a different test

Talent is the dependency you can’t hold in inventory, and it behaves oddly under the five tests.

Substitution at the person level is called hiring, and it works for roles where the skill is documented and transferable. It fails for roles where the skill is one specific machine’s behavior on a bad day. Independence applies literally: if your three cryogenic engineers all came from the same vendor, and that vendor can hire them back, you have one supplier. Jurisdiction applies too, through clearance eligibility, visa status, and deemed-export rules, which treat disclosing controlled technology to a foreign national inside your own country as an export requiring a license.

Regeneration is where talent scores worst. A doctoral pipeline runs on a timescale of years and moves nothing inside a program’s planning horizon. Conversion is the only lever that acts fast enough: taking RF engineers, cryogenic technicians, control-systems engineers, and semiconductor process engineers who already have the adjacent skills, and training them into quantum hardware roles over months rather than years. That pathway is also how most working quantum engineers arrived, which is worth knowing when a program is staring at a hiring plan it can’t fill. QuantumCareers.com maps the routes in more detail.

So talent shows up as a chokepoint in nearly every honest review, and it’s the one item a procurement office cannot close with a purchase order. It closes through training and retention, which is a different budget and a different owner, and that reassignment should happen in the same meeting where the rating is agreed.

Seven ways reviews get the classification wrong

Price mistaken for scarcity. A part that costs four times what it should is a budget item. Domestic production is an expensive answer to a cheap problem.

Lead time mistaken for unavailability. Long lead times are only a chokepoint when they exceed the horizon of the decision they feed. Otherwise they are a reason to order sooner.

“The only vendor we’ve qualified” mistaken for “the only vendor.” These are different sentences. The real dependency is the requalification cost for the alternative, and that cost is knowable before you need it.

Vendors counted at the wrong layer. Six suppliers of a system that all buy the same subsystem from two sources is a two-supplier dependency. Count until you reach a layer with genuine plurality.

Domestic assembly mistaken for domestic control. Where the license, the design, and the controlled content originate governs who can halt the shipment.

Consumables and capital charges scored identically. The helium-3 in a sealed dilution refrigerator circuit is a fixed charge bought with the machine, and rating it as an annual consumable produces a stockpiling program nobody needed.

Acquisition risk closed while operating risk stays open. A signed contract closes half the exposure on any complex instrument. The maintenance half outlives it and rarely appears on the same page.

Your chokepoints are not the field’s

Rate against a mission and a modality, meaning the qubit technology the program has actually committed to: superconducting circuits, trapped ions, neutral atoms, photonics, spins in silicon.

For a government running a portfolio across three of those, an embargo on dilution refrigerators is severe and partial. Parts of the portfolio keep operating, and the national capability degrades rather than stops. For a company that builds only superconducting machines, the same embargo is total. Reverse the case and it holds: a narrow-linewidth laser at 729 nm is irrelevant to a superconducting program and existential to a calcium ion program, and isotopically enriched silicon-28 matters enormously to spin-qubit work and not at all to anyone else.

The same component, three programs, three ratings, none of them wrong. What is wrong is a review that rates components in the abstract, because the resulting document tells no particular reader what to do.

From rating to action

Chokepoint mitigations, roughly in order of cost:

1. Design it out. Available only early, and only at the architecture or modality level. The cheapest mitigation and the one with the shortest window.
2. Qualify a second source before you need it. Pay the requalification cost during calm conditions, when the engineering time exists and nobody is improvising.
3. Hold inventory or a spare unit. Works for parts, not for services or people, and only for as long as the stock lasts.
4. Build a minimum domestic or allied capability. Enough to maintain, verify, and if necessary fabricate at low volume. Not enough to compete on price, and not intended to.

Inconvenience mitigations are contract terms, schedule float, and a named owner who watches the item and escalates when the rating changes.

One capability sits underneath all of these and gets funded last: the ability to test and characterize what you import. A program with its own metrology and acceptance testing can accept components from a wider set of suppliers, because it is not extending trust it can’t check. That is usually the cheapest sovereignty investment available, and it improves your position against every chokepoint on the list at once.

The reason to separate chokepoints from inconveniences at all is that mitigation budgets are finite. Every dollar hedging an annoyance is a dollar not spent on the dependency that would stop the program, and an all-red heatmap guarantees that trade gets made badly.

Building the judgment in-house

The five tests are easy to state and hard to apply. Applying them takes someone who knows what a pulse tube does, what requalification involves, why a control threshold attaches to a specification rather than a product line, and what a cryogenic engineer’s job actually consists of on a Tuesday. That is a hardware and standards literacy problem before it is a procurement problem, and it is why so many dependency reviews get outsourced to a consultancy and then sit unused: the organization can’t argue with the ratings it paid for.

We build our programs for exactly that gap, so that the procurement leads, program managers, and policy analysts responsible for these decisions can run the assessment themselves and defend the results. For the deeper technical background on the quantum industrial base, PostQuantum.com covers the full stack layer by layer. For structured training, our current catalog is at quantumacademy.com/.