Case Study: Technical Due Diligence on a Quantum Startup

Executive Summary

An investment committee is considering a $40M Series B in a quantum hardware startup. The deck is impressive: a novel qubit modality, a published Nature paper, a Fortune-500 pilot customer, and a roadmap to 10,000 qubits by 2031.

This case study performs the technical diligence. It works through six questions that separate a real technical position from a well-told story, and reaches a recommendation. The framework applies to evaluating research groups, vendor claims, and internal projects as much as to investments.

Skills applied

• Assessing hardware claims against physical constraints (§31.9).
• Evaluating benchmark results and their baselines.
• Distinguishing pilot revenue from product revenue.
• Identifying the binding scaling constraint of an architecture.

The company

QubitWorks — 45 employees, $28M raised, a modality they describe as "hybrid spin-photonic." Claims: 64 qubits, 99.4% two-qubit fidelity, 10,000 qubits by 2031, three paying customers.

Question 1: Is the fidelity claim measured or projected?

Asked: how was 99.4% measured, on how many qubit pairs, and what was the median versus worst?

Answer: 99.4% is the best pair, measured by interleaved randomized benchmarking. Median across the device is 97.1%; the worst usable pair is 94.2%.

Assessment. Not dishonest — best-pair reporting is common — but the operative number for circuits is the median, and 97.1% is roughly 5× worse than the leading platforms. At 97.1%, a 100-gate circuit has fidelity $0.971^{100} \approx 0.05$.

The general rule: always ask for median and worst, not best. A circuit is limited by the qubits it uses, not the best one on the chip.

Question 2: What is the scaling constraint, and is it acknowledged?

Asked: what physically prevents this architecture from reaching 10,000 qubits, and what is the plan?

Answer: a clear one. Their modality requires optical addressing, and beam-steering optics scale poorly beyond a few hundred sites. Their plan is integrated photonic delivery, which is in development with a partner foundry, prototypes expected in 18 months.

Assessment: strong. A team that can articulate its own binding constraint precisely, and has a specific plan with a timeline, is a materially better sign than one claiming no fundamental obstacles. Teams that cannot name their scaling constraint either do not understand it or are hiding it.

Question 3: What does the Nature paper actually show?

Asked: what was demonstrated, and what did the reviewers push back on?

The paper reports a 12-qubit entangled state with fidelity 0.71, verified by parity oscillation (a technique from Chapter 5's GHZ case study — and the correct one, not a computational-basis histogram).

Assessment: solid but modest. 12 qubits at 0.71 fidelity is real, credible work, comparable to what several groups achieved 3–5 years ago on other platforms. Verified properly. It does not demonstrate a lead; it demonstrates competence.

Question 4: What are the customers actually paying for?

Asked: what do the three customers receive, what do they pay, and would they renew?

Assessment: this is not product revenue. All three are exploratory or grant-funded. No customer is buying computation because it solves a problem faster or cheaper. That is the norm across the entire industry — but it must be counted honestly, because it means revenue reflects strategic curiosity and government funding, both of which can evaporate quickly.

The diagnostic question for any such contract: would the customer renew if the vendor's hardware did not improve? For exploration contracts, usually not.

Question 5: Does the roadmap arithmetic work?

Claim: 10,000 qubits by 2031, five years out.

Sustained ~3× per year. For comparison, superconducting platforms have grown roughly 2× per year over the past decade, with far more capital.

But the harder question is fidelity. 10,000 qubits at 97% median fidelity is not useful for anything — no algorithm and no error-correcting code works at that error rate. The roadmap gives qubit counts and is silent on fidelity targets.

Asked: what is the fidelity roadmap? Answer: 99.9% by 2029.

That is the number that matters, and it is not on the slide. Ask for it every time.

Question 6: Is the team's composition right?

45 employees: 18 physicists, 9 electrical/control engineers, 6 software, 4 optics, 3 cryogenics, 5 business.

Assessment: physics-heavy. Chapter 29's cost analysis showed that the quantum processor is ~7% of a system's capital, with control electronics, software, and integration dominating. A 40% physicist ratio suits a research group; a company shipping systems typically needs more engineers than physicists. This is a common pattern in university spin-outs and a predictable source of execution risk.

The recommendation

Technical position: credible but not leading.

Recommendation: pass at this valuation, with a specific re-engagement trigger — revisit if they demonstrate median two-qubit fidelity above 99.5% across 100+ qubits, or if the integrated photonic delivery prototype works on schedule. Both are checkable in 18 months.

Note what did not drive the decision: the modality being novel, the Nature paper existing, or the Fortune-500 logo. Those are the elements of the pitch that carry least information.

Discussion Questions

1. Best-pair fidelity was 2.3 points above median. Why is the median the operative number, and why is best-pair reporting so common?
2. Knowing your own scaling constraint was scored as a strong positive. Why is that more informative than the constraint itself?
3. Exploratory revenue is the industry norm. How would you distinguish a company that will convert it from one that will not?
4. The roadmap gave qubits but not fidelity. Construct the fidelity roadmap the company would need for their qubit roadmap to matter.

Your Turn: Extensions

• Take a real quantum company's public materials and answer all six questions from available information.
• Compute the maximum useful circuit depth at 97.1% versus 99.5% two-qubit fidelity.
• Compare the team-composition ratios of a hardware startup against an established vendor.
• Draft the five questions you would put to a vendor before a purchase decision.

Key Takeaways

• Ask for median and worst gate fidelity, never best — circuits are limited by the qubits they use.
• A team that can precisely name its own scaling constraint and has a dated plan is a stronger signal than one claiming no obstacles.
• Exploratory and grant-funded contracts are not product revenue; ask whether the customer would renew absent hardware improvement.
• Qubit-count roadmaps without matching fidelity targets are not roadmaps — 10,000 qubits at 97% fidelity is useless.
• Team composition should reflect that the processor is a small fraction of the system; physics-heavy companies carry execution risk.