The Next Big Thing: Quantum Computing — September 2026

Prepared 2026-09-14 for the NBT page on gendler.net. Portfolio weights are as of the 2026-09-11 close. Every figure traces to a primary source: a regulatory filing, a company release, or a peer-reviewed paper. Where something could not be confirmed, this page says so instead of filling the gap.

Revised 2026-09-25. The insider-selling section is rebuilt from Form 4 filings, the Infleqtion lock-up is settled from primary documents, Xanadu's lock-up date is confirmed from its CEO's filing, and the holdings disclosure is corrected.

This document is not investment advice. It is the reasoning behind positions already held, written down so it can be checked against reality later. The author holds the eight securities in the quantum book. He may also hold other securities named here; any such holding is incidental to the thesis.

Public disclosure: this page publishes portfolio weights only, as percentages of the quantum book. No share counts, no cost basis, no dollar totals. Company-level public figures (revenue, cash, market cap, guidance) may appear.

---

0. The thirty-second version

For the reader who runs a business and has a meeting in five minutes.

• The physics has stopped arguing. Three independent, peer-reviewed experiments have shown quantum error rates falling as machines grow. That was the make-or-break question, and it is answered.
• The engineering has not. Nobody has a machine that runs a commercially useful job better than a conventional computer. The gap is measured in years and in hundreds of thousands of qubits.
• So the sector is priced on the 2030s. Seven of eight holdings trade between 60 and 560 times trailing sales. That is not a multiple of a business; it is the premium on an option.
• This book is eight names, one factor. Trapped ion, superconducting, neutral atom, photonic, annealing, and a software layer, with IBM as ballast. Measured over the summer, the pure-plays moved together with a correlation of 0.6 to 0.9 and a beta to the S&P 500 near 4.
• The nearest risks are scheduled, not scientific. Lock-up expiries, registered government stakes, equity facilities, and a Federal Reserve meeting on September 16 with a rate hike priced at better than even odds. Long-duration assets do not enjoy that.
• What would change the mind behind this page is written down in §11, in advance, so it cannot be rewritten afterwards.

---

1. What a quantum computer actually is

Ordinary computers store bits that are 0 or 1. A quantum computer stores qubits. Before you look, a qubit can be 0, 1, or a weighted blend of both, which is superposition. Qubits can also be correlated so that measuring one constrains another, which is entanglement. Neither is magic. They are the raw materials of a different kind of arithmetic.

Useful quantum algorithms work by interference. The machine arranges the many possible answers so that the wrong ones cancel and the right one is more likely when you finally measure. That is the whole idea in one breath. It is not "trying every answer at once," and it is not parallel universes. Relative phase is what makes cancellation possible; without it, superposition is just a coin flip.

Scott Aaronson's plain-language account of why this is hard to explain without phase and interference is the best next stop; IBM's "What is quantum computing" pages, NIST's explainer, and the Qiskit Learning modules follow. The textbook definitions live in Nielsen and Chuang.

Hardware teams build qubits several ways: trapped ions, superconducting circuits, arrays of neutral atoms, photons, and a still-contested topological design. The labels matter for engineering trade-offs (speed against coherence, cryogenics against room temperature, fab-friendly against exotic) but not for the core idea. A scoreboard of which approaches this book owns appears in §6.

Why it is hard

Qubits are fragile. Heat, vibration, and stray electromagnetic fields scramble the state; physicists call it decoherence. Every useful machine built so far is noisy. Counting physical qubits in a press release therefore tells you almost nothing about whether a computation will finish correctly. A bigger chip that is still too noisy is not progress. It is a louder error generator.

The only known path to a machine that stays correct while it computes is quantum error correction: many imperfect physical qubits encode one more-reliable logical qubit. Think gross versus net. Below a critical physical error rate, the threshold, adding physical qubits reduces the logical error. Above it, adding qubits makes things worse. Crossing that threshold, and then keeping logical error falling as systems grow, is the engineering problem the entire sector is now organized around.

That crossing has been shown three times in peer-reviewed work. Google's Willow chip ran a distance-7 error-correcting code below threshold in December 2024 (Nature, doi:10.1038/s41586-024-08449-y (opens in a new tab)), with logical error falling as the code grew. USTC matched the result in December 2025 at a weaker suppression factor (Physical Review Letters, doi:10.1103/rqkg-dw31 (opens in a new tab)). A Harvard / MIT / QuEra architecture published in November 2025 (Nature, doi:10.1038/s41586-025-09848-5 (opens in a new tab)) put below-threshold memory, logical entanglement, logical magic states, and entropy removal in one system, the most complete demonstration to date. These are milestones on the error-correction road, not finished utility machines. The overhead is still brutal: optimistic estimates talk in hundreds of thousands of physical qubits for hard commercial workloads, while today's largest error-corrected demonstrations use on the order of a hundred.

What "advantage" means

A headline that a quantum chip beat a supercomputer has a poor survival rate. Classical algorithms often catch up, sometimes within months. Prefer "narrow benchmark under dispute" until a claim has held up under independent scrutiny. The sector has therefore moved its own goalposts toward verifiable advantage and toward fault tolerance: errors found and fixed faster than they appear, for a long computation.

As of this writing, nobody has shown a logical qubit with error low enough for hard commercial work (about one error in a million operations or better), a full universal gate set at that quality feeding a real algorithm, or any error-corrected computation that clearly beats the best classical method on a problem anyone cares about outside a narrow benchmark. Until that arrives, valuations in this sector are a claim on the 2030s, not a multiple of 2026 sales.

What you can buy time on today are noisy intermediate-scale machines, useful for research, algorithm development, and proving out control stacks, not yet for most commercial jobs. John Preskill's 2018 framing of that era (Quantum 2, 79, doi:10.22331/q-2018-08-06-79 (opens in a new tab)) still fits. Everything below assumes these ideas and asks what they imply for a portfolio already concentrated in the companies building the hardware, the fabrication capacity, and the error-correction lead.

---

2. The thesis in one paragraph

Quantum computing has crossed from "does the physics work" to "can the engineering scale". As the last section set out, the precondition for error correction has been met; the engineering has not. Three independent groups have now shown error rates falling as systems grow, which is the precondition for everything else. The sector's revenue does not remotely justify its valuations today, and the honest case for owning it is not current earnings but position: the companies that hold the hardware, the fabrication capacity and the error-correction lead when fault tolerance arrives are unlikely to be assembled cheaply afterwards. That is a bet on a decade, priced daily by a market that reprices it violently on news.

---

3. Why these holdings

Anyone who has bought entitled land before the road was built will recognize the shape of this. You are not paying for cash flow. You are paying for a position that will be expensive to assemble once the road is visible, and you accept that the carry is ugly in the meantime. Five claims, then the weights.

Modality first, ticker second. The book spans trapped ion (IONQ, QNT), superconducting (IBM, RGTI, and QBTS's gate-model path), neutral atom (INFQ), annealing (QBTS), photonics (XNDU), and a hardware-agnostic software layer (HQ). Topological is the serious modality not held, and it is the one under formal challenge in Nature (see §9, Microsoft). This is diversification across an unsettled technical question, not a claim that every approach wins.

The 2026 listing wave is the opportunity. Four of eight names could not have been bought in January: INFQ (2026-02-17), HQ (2026-03-20), XNDU (2026-03-27), QNT (2026-06-04). The book is a bet that public-market access arrived before the engineering was finished.

IBM is ballast, not a quantum earnings story. About a quarter of the book. Profitable, diversified, dividend-paying, and, measured over the summer, essentially uncorrelated with the rest (§8). IBM still discloses no standalone quantum revenue. It dilutes a sector de-rating; it does not hedge one. Ballast does not steer.

IONQ is the commercial concentration. About a quarter of the book. Trailing revenue leader among pure-plays; the FY26 guide is SkyWater-combined (see §4). The thesis accepts acquired growth and a large non-cash warrant mark as the price of owning the revenue outlier.

HQ is a small software satellite (about 3%). If the winner-take-all story is that compilers outlive any one chip, this is the expression of it: pre-revenue by design, long lock-ups, thin float, and a stock that moves like a coiled spring in both directions.

Shape of the book (2026-09-11 close): pure-play quantum is about 74% of the book; IONQ plus IBM is about 52%.

Weights move with price. Anything published from this table carries its as-of date.

---

4. The holdings

Financials are Q2 2026 (quarter ended 2026-06-30) unless noted.

IBM — the anchor

The only profitable, diversified position: about $69.1B trailing revenue, $8.1B cash, roughly 3.3× sales, a 2.8% dividend. IBM discloses no quantum revenue at all, so this is not a quantum earnings story. It is a $10B five-year commitment (June 2026), 90-plus deployed systems, the Qiskit software standard, and a purpose-built quantum wafer foundry in Albany with a proposed $1B CHIPS award. Starling, the 200-logical-qubit fault-tolerant target, remains set for 2029.

Dated deployment (2026-09-10 IR): a Swiss quantum innovation hub with Lockheed Martin at ETH Zurich, and the first IBM Quantum System Two at CSCS Lugano carrying Nighthawk r2, targeted to be operational by end-2026 (IBM Quantum blog (opens in a new tab)). Nighthawk r2 is already on the IBM Quantum Platform. Kookaburra, the first qLDPC memory-and-logic module, remains a 2026 roadmap item and is unconfirmed; IBM's November Quantum Developer Conference is the likely checkpoint.

Ballast has its own weather. IBM pre-announced a Q2 mainframe shortfall on 2026-07-14 and the stock fell 25% in a session, its worst day in the two-year window, then cut full-year guidance to 4–5% constant-currency growth. Its July advantage claims drew a classical rebuttal within two weeks. The position is a quarter of the book because it is uncorrelated, not because it is calm.

QNT (Quantinuum) — the technical leader

Trapped ion, and the error-correction leader by published results. $8.0M quarterly revenue against $2.1B cash and FY26 guidance of $28–32M: the widest gap between technical standing and revenue in the portfolio, and the reason its roughly $12.8B market value is a statement about 2030, not 2026. Honeywell and Cambridge Quantum retain the great majority of the equity through a non-economic Class B structure.

Government money arrived wearing a term sheet. A CHIPS R&D award of up to $100M was finalized 2026-09-08. Per the 8-K, Commerce received Class A shares priced at the lower of the IPO price less 20% or the award-date close less 15%, with a resale shelf due within three months and the right to demand underwritten offerings. Cash comes in three tranches ($56M up front, then $32M and $12M on milestones) with a clawback up to the full disbursed amount. Not a free grant.

Roadmap: Sol 2027, Apollo (fault-tolerant) 2029, neither slipped so far. Logical-qubit counts cited on this page are from March and September 2026 preprints and await peer review. A Russell 1000 / 3000 addition effective 2026-09-21 is reported from FTSE Russell's preliminary list but not yet confirmed on the index provider's own site.

IONQ — the revenue leader

The commercial outlier: $246.5M trailing revenue. FY26 guidance was raised twice along a clear IR chain:

• 2026-08-05 (Q2): FY26 $280–290M, explicitly ex-SkyWater.
• 2026-09-08: FY26 $450–460M, SkyWater from the 2026-07-31 close through year-end, minus estimated intercompany.

Much of the year-over-year growth is therefore acquired foundry rather than organic quantum-platform revenue; treat the combined guide as a sum of parts, not a pure quantum raise. The $1.87B GAAP quarterly loss is about $1.65B of non-cash warrant marks, not operations. Roughly $2.0B cash pro forma. Superion 256 launched 2026-09-08 with customer deliveries in 2027. The same Investor Day produced a compiled "blueprint" for breaking 256-bit elliptic-curve signatures, which is a resource estimate on paper, not a demonstration, and a quantum-safe networking agreement with Congruity360 (value reported by secondary coverage at about $8M; not stated in the release headline).

Two cautions. 256-qubit systems at 99.99% were promised for 2026 and are now "first half of 2027." Wolfpack Research published a short report on 2026-02-04 alleging cancelled Pentagon earmarks had supplied most of 2022–2024 revenue; IonQ called it false and misleading, and the shares fell 8% that day and 14% the next. IonQ was absent from the 2026-09-08 CHIPS public-book equity tranche (it bought a foundry instead).

INFQ (Infleqtion) — neutral atoms with real revenue

The only holding with meaningful revenue outside computing: quantum sensing, meaning atomic clocks, RF sensors, and a NASA gravity-gradiometer program. $13.5M quarterly revenue, FY26 guidance about $45.1M, roughly $581.8M liquid, no debt, and the lowest sales multiple of the pure-plays at about 70×. Neutral atoms have the strongest peer-reviewed error-correction record of any modality, though Infleqtion's own technical claims are company statements rather than fault-tolerance demonstrations.

Its Q2 was restated on 2026-08-17 and the 10-Q filed a day late. The company calls the errors immaterial, and cash was unaffected, but it is the kind of thing that matters if it repeats. Infleqtion's May 2026 CHIPS item was a letter of intent only, not a finalized award; do not conflate it with the September 8 finals. On the tape it is the healthiest chart in the pure-play sleeve (§8), a chart whose largest sellers in late May were a director's funds and its own CEO (§10.6), which is worth exactly as much as any three-month chart is worth.

QBTS (D-Wave) — the contested one

Revenue is flat at about $3.1M a quarter; FY2025's $24.6M was inflated by a single system sale. Bookings tell a better story ($35.5M in H1 2026, $40.7M backlog). Its Science advantage claim has been contested continuously for 18 months: a narrow benchmark under dispute, not a settled result. Earlier thinking on this page, in January 2026, said avoid; the counter-argument is that the Quantum Circuits acquisition (January 2026) bought a gate-model path, so this is no longer a pure annealing bet.

A CHIPS R&D award of up to $100M was finalized 2026-09-08, again as minority Commerce equity at about a 15% discount with resale rights. The CFO retired effective 2026-09-02 (8-K filed 2026-08-19; the company states no disagreement over financial reporting). He had sold $16.8M of stock between 05-22 and 06-15, none of it under a plan, and held about 1.1 million shares at his last filing. The stock's 9% drop came a week later, on 2026-08-26, a day the whole sector fell, so the press attribution to D-Wave alone is an inference.

RGTI (Rigetti) — superconducting, behind the leaders

$5.1M quarterly revenue on $541.3M cash, no debt. Cepheus-1-108Q is generally available at about 99.1% median two-qubit fidelity, below the 99.5% originally promised, and delivered in April 2026 rather than end-2025. No logical-qubit or below-threshold result. The roadmap language softened in August 2026 from a dated late-2027 target to "~1,000 qubits over roughly a three-year time horizon". Rigetti was also not among the eleven teams DARPA advanced to Stage B of its benchmarking program in November 2025, while IonQ, Quantinuum and Xanadu were.

A CHIPS R&D award of up to $100M was finalized 2026-09-08 (Commerce equity, same structure as QBTS and QNT). A September paper in Physical Review Applied on qubit-efficient optimization is real science and does not change the roadmap picture. Of everything in this portfolio, this is the name whose story has moved backwards.

XNDU (Xanadu) — photonics, earliest stage

Photonic, room-temperature optics, and the first pure-play photonic company to list. $1.5M quarterly revenue, $312.8M cash, a new roadmap (2026-08-31) to 1,000-plus logical qubits by 2031, and a lithography collaboration with ASML announced 2026-09-09. Photonics has manufacturability results but no logical-qubit demonstration; it is the furthest from the error-correction frontier of anything held.

The near-term issue is supply, not operations: a large resale shelf (the stock fell 61% in a day when it was digested on 2026-05-04), a $300M equity facility with $232.8M of capacity left, and an insider standoff that ended on 2026-09-22, the day the CEO made his first open-market sales in sixty days (§17). Its loss also doubled quarter on quarter, which makes it the shortest runway in the book.

HQ (Horizon Quantum) — the software bet

Not a hardware company: a compiler and IDE stack whose pitch is automatically accelerating conventional code on quantum machines. Zero revenue last quarter, $113.3M cash, and a $115.2M loss that is almost entirely a non-cash warrant remeasurement. Primary lock-up language for shares is 24 months, to about 2028-03-19; near-term supply is therefore minimal, but the tradeable float is thin and moves hard: realized volatility near 160% annualized, and a 20% intraday drop on 2026-09-14 with no release or filing to explain it. Its FY2025 annual report carried going-concern language tied to the SPAC's liquidation deadline, resolved by the merger closing. If the winner-take-all framing is right that software matters as much as hardware, this is the portfolio's expression of it, at the earliest possible stage.

---

5. A short glossary

---

6. Where the technology stands

The axis that matters is error correction. Everything else, from qubit counts to "quantum advantage" headlines to roadmap art, is downstream of whether logical error rates fall as machines grow.

The precondition has been met, three times (Google 2024, USTC 2025, Harvard/MIT/QuEra 2025, cited in §1).

Quantinuum has the most logical qubits doing actual computation: 48 error-corrected and 94 error-detected (March 2026 preprint), and a logical two-qubit error of 2.8 × 10⁻⁴ without post-selection on Helios (September 2026 preprint). Both await peer review.

What nobody has shown, as of today: a logical qubit below about 10⁻⁶ error, a full universal gate set at scale feeding real algorithms, or any error-corrected computation that beats a classical computer on a problem that matters outside a narrow benchmark.

Modality scoreboard

The portfolio therefore spans every serious modality except topological, which is the one whose central claim is under formal challenge in Nature. That is either deliberate diversification across an unsettled technical question, or an admission that nobody yet knows which approach wins. Both readings are fair.

---

7. Advantage claims, honestly kept

"Quantum advantage" claims have a poor survival rate, and a thesis that ignores this is decoration.

Amit Hagar's July 2026 preprint argues that with one exception, every flagship claim since 2019 was classically matched within 18 months. Scott Aaronson concedes the catch-ups but not the conclusion. Both are right about the record, which is why the sector's own attention has moved to verifiable advantage and fault tolerance.

---

8. The tape: what the market is saying

A thesis about a decade still has to live through each quarter, and the quarter just ended was unkind. This section measures rather than opines. Closes are Yahoo Finance data through 2026-09-11. Volatility is the standard deviation of the last 60 daily returns, annualized; betas and correlations are computed over the 68 sessions from 2026-06-05, the first day all eight names traded together. Live prices on this page come from the market-data feed, not from these tables.

Trend

The four 2026 listings do not yet have 200 sessions of history.

Three things a chartist would say out loud. First, every pure-play is below its 50-day average except INFQ, and the 2021 cohort (IONQ, RGTI, QBTS) is 17–22% below its 200-day: a sector in a downtrend, not a pullback. Second, the sector's closing peaks cluster in early June 2026, the week Quantinuum listed. The largest IPO in the sector's history marked the top, which is the oldest pattern in the book. Third, the S&P 500 is 2% from its high while these names are 40–80% from theirs. That divergence is not about qubits. It is about the discount rate, and the Federal Reserve meets on September 16 with a hike priced at better than even odds.

One irony worth a sentence: the ETF called quantum (QTUM) is up 34% this year, because it mostly holds large-cap semiconductors and software. The companies called quantum are down 18–36% year to date. The index and the industry are not the same trade.

Volatility and correlation, 2026-06-05 to 2026-09-11

The window is the 68 sessions in which all eight names have traded.

What the numbers say:

• Seven tickers, closer to one trade. IONQ, RGTI and QBTS move with correlations of 0.88 to 0.94; the average pairwise correlation across all seven pure-plays is 0.61. The 2026 listings (QNT, XNDU, HQ) are less coupled, partly because lock-ups and thin floats give them their own weather. Do not read that as diversification; read it as noise.
• IBM is the only thing in the book that does not move with the book. Correlation with every pure-play is within a rounding error of zero over the window. That is why it is a quarter of the portfolio. Its own 70% realized volatility is almost entirely one session (the 25% drop of 2026-07-14); on a 20-day basis it runs about 32%.
• At the 09-11 weights, the whole book carried about 62% annualized volatility and a beta to the S&P 500 of 2.75. The S&P itself ran 14% over the same window. A 1% down day in the index has, on average, been a 2.75% down day here.
• The book fell about 31% over the window (weights held constant, a simplification). Every holding was down except HQ (+28%); IONQ was the worst at −44%.

Dated sessions, for the record

None of this is a signal to act. It is the weather the thesis has to survive, measured so the next revision of this page can check whether it did.

---

9. Names in the crosshairs

A watchlist, not a shopping list. Why each is not in the quantum book.

Alphabet (GOOGL). Google Quantum AI published the first below-threshold surface-code result (Nature, 2024-12-09) and the longest-surviving advantage-style claim in the scorecard above (OTOC / Quantum Echoes, Nature, 2025-10-22). The best superconducting error-correction paper is therefore not in the book. GOOGL is a diluted mega-cap: quantum is not the equity story. Watch the lab; do not pretend a 26% IONQ weight is a Willow substitute.

Microsoft (MSFT). Azure Quantum and the Atom Computing partnership are real. The headline modality, topological / Majorana, is under formal challenge: Henry Legg, Nature Matters Arising (2026-06-24, doi:10.1038/s41586-026-10567-8 (opens in a new tab)), with a Microsoft reply in the same issue (doi:10.1038/s41586-026-10568-7 (opens in a new tab)). That is why topological is the modality not held. MSFT is also a diluted mega-cap.

NVIDIA (NVDA). CUDA-Q and hybrid classical–quantum workflows matter if fault tolerance arrives. NVDA is not a qubit company. The 2025 write-up treated it as the AI giant adjacent to quantum; the 2026 book chose the machines instead. Keep it as pick-and-shovel context.

ASML and TSM. EUV lithography and leading-edge foundry are the classical chip stack. Useful if quantum control electronics go to CMOS at volume (IonQ's SkyWater thesis is the book's own version of this). Adjacent IR: the Xanadu–ASML lithography collaboration for photonic hardware (2026-09-09). They are not a quantum-computing bet.

FormFactor (FORM). The 2025 mid-cap pick-and-shovel: cryogenic probe and test. Still a clean watch if superconducting and ion-trap volumes rise. Not a substitute for a modality holding.

Honeywell (HON). Honeywell (with Cambridge Quantum's legacy) still holds the great majority of Quantinuum through non-economic Class B. The book already owns QNT Class A. Holding HON for quantum is a second, diluted claim on the same residual. Watch as the parent, not as a second Quantinuum.

Quantum Computing Inc. (QUBT). On the October 2025 ticker list. Weaker peer-reviewed record and a smaller commercial footprint than anything in the current eight. Off the page unless a primary filing changes that.

Pasqal (PSQL). Neutral-atom peer to INFQ. Business combination closed 2026-08-27; Nasdaq listing 2026-08-28. Aramco already runs a Pasqal machine in Dhahran with a commercial QCaaS layer (Aramco IR, 2026-05). Two weeks in, the stock is about 70% below its post-listing high. First-class crosshair: same modality family as INFQ, newer listing, lock-up and dilution not yet sized in the research pack. Too new to size; no weight.

IQM (IQMX). First European quantum company on a major US exchange; ADSs from 2026-07-02. Superconducting full-stack; geographic and liquidity profile differs from the US core book. Named as a 2026 listing outside the book; no more than that.

Still private, still material.

• PsiQuantum: Omega chipset in Nature (2025); DARPA US2QC final validation with Microsoft (2025-02); NIST CHIPS final award up to $100M on 2026-09-08, the private fourth award alongside RGTI, QBTS, and QNT. Watch an IPO; no ticker.
• QuEra / Harvard / MIT: the strongest peer-reviewed neutral-atom fault-tolerance paper (2025-11-10) is not Infleqtion's. INFQ is the listed sensing-plus-atoms expression; QuEra is the science lead.
• Atom Computing: Microsoft partnership; Magne (Denmark / QuNorth) targeted early 2027; May 2026 CHIPS LOI cohort.

---

10. Risk register

10.1 Valuation — a snapshot, not a fact

At the 2026-09-11 close, one vendor's numbers (stockanalysis.com). They move daily; the live page should read the feed. This table exists to make the shape of the problem visible.

A business executive reads that column the way a landlord reads a cap rate of 0.2%: the price is not about this year's rent.

10.2 The supply calendar

The most under-appreciated risk in a portfolio of recent listings: the sellers are scheduled.

A new class of seller appeared this month. In the first eight days of September 2026 the Commerce Department finalized CHIPS R&D awards of up to $100M each for Rigetti, D-Wave, Quantinuum, and PsiQuantum (private). For the three public names in this book, Commerce took minority equity stakes struck at roughly 15–20% discounts, with resale registration rights and milestone-gated disbursement. Government money arrived as registered stock, not as free grants. IonQ is absent from this tranche. Infleqtion's May item remains an LOI only.

10.3 The next ten days

Two of the three market events land in the same week and point in opposite directions for two different names. That is the sector in miniature.

10.4 Cash runway

Quarters of runway at 2026-06-30, on an operating-cash basis and on a quarterly-loss basis. Where the two diverge sharply, the company is collecting customer or government cash ahead of expense.

Nobody here is close to running out. That is the single most reassuring number in this document, and it is a direct consequence of the 2026 listing wave: these companies raised money while the window was open.

10.5 Concentration — what seven pure-plays actually buy

Seven pure-plays are about 74% of the portfolio, and they buy less diversification than seven names suggests, because they trade on the same drivers. §8 measured it: the 2021 cohort is correlated at 0.9, the whole sleeve at 0.6, and the book's beta to the index is 2.75. The clearest single-day evidence is 2026-08-26, when QBTS fell 9% on its CFO news and RGTI, IONQ and HQ fell 5–9% with no news of their own.

Twelve-month peak-to-trough ranges of roughly 75–85% for RGTI, QBTS, XNDU and HQ, and about 70% for IONQ, tell you what a de-rating looks like. Because the names are correlated, those drawdowns coincide rather than offset. The lock-up expiries in §10.2 also cluster within a few months of each other.

IBM is the only ballast, at 25.85% of the book, and it does not hedge quantum-specific disappointment; it dilutes it. Four holdings (RGTI, QBTS, XNDU, HQ) publish no revenue guidance at all, so there is no periodic external checkpoint to re-anchor against. Roadmaps and government awards fill that role, and those are correlated across names too: all three CHIPS finals landed in the same five days.

10.6 What has actually gone wrong, last twelve months

Not hypothetical risks. Observed events.

Insider selling, from Form 4 filings: Insiders of the eight holdings reported about $350M of open-market sales in the twelve months to 2026-09-22, and 96% of it was sold between 2026-05-21 and 06-15. About $7.8M, 2%, was sold under Rule 10b5-1 plans; none of the sales named below was. Shares withheld for tax on vesting ($26.4M) are not sales and are excluded. Xanadu files no Form 4s (§17).

Infleqtion is most of the total. Funds managed by Maverick Capital, whose managing partner David Singer sits on Infleqtion's board, sold about 16.8 million shares for $262.5M in four sessions from 05-21 to 05-29, roughly 7.6% of the shares then outstanding, and distributed several million more to their limited partners. In the same fortnight the CEO sold $13.1M, most of it shares from options exercised at $0.90 that day; his filing states he sold under 9% of his holding and keeps about 3.4% of the company. The chief revenue officer sold $1.6M, a director's venture fund $0.9M, and on 06-04 the CTO $2.1M. None of it waited for the 180-day lock-up bound of 2026-08-12, because that bound was never reached: the company's bylaws release the lock-up early once the volume-weighted price holds at or above $12.00 for fifteen trading days in any 180, and that condition was met in April (§17). Insiders were free from mid-April, the first-quarter results landed 05-14, and the selling began 05-21.

D-Wave's CEO sold $18.0M on 06-08, all of it 687,627 shares exercised that day at under a dollar; his previously held shares were untouched. Its CFO sold $16.8M between 05-22 and 06-15, $9.1M as a same-day exercise-and-sell and $7.7M from shares already held; his last sale was 06-15, the retirement 8-K followed on 08-19, and he still held about 1.1 million shares. Rigetti's then-CTO, now COO, sold $13.3M, $12.7M of it on 05-29, when a fifth of the shares came from options exercised at $0.27 and the rest cut his holding by more than half; Rigetti's CEO sold $4.5M, and one director sold $2.1M of warrants while another sold $4.3M under a plan adopted in 2025. IBM's software chief sold $5.8M on 08-26. IonQ shows no officer sales. Quantinuum shows none at all: eleven directors and officers bought $24.7M at the $60 IPO price on 06-05, $15.0M of it by one director, and its lock-up runs to about 12-01. No revenue guidance has been missed by any holding that gives guidance, and no going-concern language appears in any current filing.

---

11. What would confirm it, and what would break it

Dated milestones to watch:

IBM's Kookaburra has not been confirmed delivered with under four months left in the year. Silence at the November conference would be informative.

Exit signals, stated in advance (this is the part a thesis written after the fact always omits):

• A logical-qubit roadmap slipping twice in a row from the same company. Rigetti is already one slip in.
• Revenue guidance missed at IONQ, INFQ or QNT, the three names where guidance exists to miss.
• Going-concern language, or an equity facility restarted at depressed prices.
• A credible classical rebuttal of the last surviving advantage claim.
• Insider selling concentrated immediately after a lock-up expiry, rather than under pre-arranged plans. Fired, at Infleqtion, 2026-05-21. Its bylaws released the lock-up early in April on a price trigger (§17). The first-quarter results cleared on 05-14, and in the ten sessions from 05-21 the CEO, CTO, chief revenue officer, a director's fund and a director's firm sold 99% of the year's insider volume, none of it under a Rule 10b5-1 plan. The selling did not follow the release immediately, but it did follow the first open trading window after it, which is the earliest these holders could practically have sold. The test was written to catch insiders selling as soon as they are able, without the pre-commitment of a plan, and that is what the filings show. The Xanadu window opens this month and cannot be watched through Form 4s (§17).
• A second restatement anywhere in the book.

What would confirm it is quieter: a peer-reviewed logical error rate an order of magnitude below today's best; a fault-tolerant module shipped on the date it was promised; a guide raised on organic, not acquired, revenue. Any one of those would be worth more than a year of headlines.

---

12. The bear case

Stated at full strength.

1. Classical algorithms keep catching up. The record through September 2026 supports the pattern; only Google's OTOC result has held for a year.
2. The overhead is brutal. Optimistic March 2026 estimates still need hundreds of thousands of physical qubits, or 10,000 atoms using codes whose logical gates have never been demonstrated. Today's largest error-corrected computations use about 100 physical qubits.
3. Roadmaps slip. IonQ 2026 → 2027. Rigetti twice. Microsoft moved to 2029.
4. Revenue is not tracking valuation. Quantinuum: $8M a quarter at roughly $12.8B. D-Wave: $5.9M for the half year. The 2021 SPAC cohort (IonQ, Rigetti, D-Wave) traded below listing prices for years, and 2026 produced a larger cohort by the same mechanism. Two of the four 2026 listings already trade below their reference price.
5. A headline modality may be built on sand. Microsoft's topological claim is under formal challenge in Nature. Not in the book, but it is evidence about how far a well-funded quantum claim can travel before scrutiny catches it.
6. The tape agrees with the bears, for now. Every pure-play in a downtrend, the sector top coinciding with its biggest IPO, and a central bank leaning toward higher rates. Long-duration assets have no friends in that room.

The physics objection is losing to data. The engineering objection is intact and is the one the optimists themselves cite. The commercial objection is fully supported by the record. The market objection is loud and, like all market objections, temporary in one direction or the other.

---

13. The nearer-term driver: post-quantum cryptography

PQC is the one quantum-adjacent market with hard deadlines and no dependency on any quantum computer working. NIST's standards are final (FIPS 203/204/205), and Executive Order 14412 of 2026-06-22 sets federal deadlines of 2030 for key establishment and 2031 for signatures, with contractor obligations attached. NSA's CNSA 2.0 requires compliant national-security acquisitions from 2027-01-01. FIPS 140-2 modules stop qualifying for new federal procurement on 2026-09-21.

The resource estimates keep shrinking, which is the part that moves regulators: Google's March 2026 preprint put 256-bit elliptic-curve breaking under 500,000 physical qubits, and IonQ's September 2026 blueprint is the same genre. Both are paper, not hardware. The mandates are law either way.

None of the eight holdings is primarily a PQC vendor, so this is context rather than a position: it funds the sector's credibility and keeps government money flowing regardless of when fault tolerance arrives.

---

14. The named challenger: robotics

Quantum is the NBT thesis. Robotics is the named rival for "next platform" mindshare and capital, not a technical substitute for qubits.

Tesla's Optimus production path is IR: the Q1 2026 update describes Gen 3 as the first mass-production design and Fremont S/X lines converting for first-generation builds (ir.tesla.com). September multi-thousand-unit supplier-order headlines are not Tesla IR; treat them as rumor until the company says so. Figure AI remains private; Figure 02/03 at BMW Spartanburg is partner-backed (BMW Group press, 2026-06-25).

The point for this page: a decade-long quantum book can be right on physics and still lose the "next big thing" argument if humanoid deployment outruns fault tolerance. That is competing attention and capital, not a qubit replacement.

---

15. Investor resources

Every link below was checked against the right registrant. The regulatory index matters more than the IR page: XNDU and HQ are foreign private issuers filing 20-F and 6-K rather than 10-K and 10-Q, so they report less often than the rest.

Sector resources: the National Quantum Initiative (opens in a new tab), DARPA's Quantum Benchmarking Initiative (opens in a new tab), the Quantum Economic Development Consortium (opens in a new tab), McKinsey's Quantum Technology Monitor 2026, Quantum Computing Report (opens in a new tab), and NIST's Post-Quantum Cryptography project (opens in a new tab). No stock-promotion sites and no newsletters selling picks.

Implementation note: six of the eight IR sites — Rigetti, Xanadu, Horizon, Quantinuum, IonQ and D-Wave — answer HTTP 403 or a bot interstitial to anything that is not a person at a browser, including a real headless Chrome. Only IBM and Infleqtion answer cleanly. SEC EDGAR serves fine but requires a User-Agent carrying contact details, per its own access policy. An automated link checker will report six healthy links as broken unless it can tell a bot-block from a 404.

Key primary links — technology

• Google Quantum AI, Nature 638, 920–926 (2025); online 2024-12-09: doi:10.1038/s41586-024-08449-y (opens in a new tab)
• Google Quantum AI, Quantum Echoes / OTOC, Nature 646, 825–830 (2025-10-22): nature.com/articles/s41586-025-09526-6 (opens in a new tab)
• Bluvstein et al., Nature 649, 39–46 (2025-11-10): doi:10.1038/s41586-025-09848-5 (opens in a new tab)
• He et al., Phys. Rev. Lett. 135, 260601 (2025-12-22): doi:10.1103/rqkg-dw31 (opens in a new tab)
• Legg, Nature Matters Arising (2026-06-24): doi:10.1038/s41586-026-10567-8 (opens in a new tab)
• Microsoft reply, same issue: doi:10.1038/s41586-026-10568-7 (opens in a new tab)
• Preskill, Quantum 2, 79 (2018): doi:10.22331/q-2018-08-06-79 (opens in a new tab)
• IBM Quantum roadmap: ibm.com/roadmaps/quantum (opens in a new tab)
• IBM Swiss System Two / Nighthawk (2026-09-10): ibm.com/quantum/blog/swiss-innovation-hub (opens in a new tab)

Key primary links — holdings IR and CHIPS

• IonQ FY26 outlook (2026-09-08, SkyWater-combined): ionq.com news (opens in a new tab)
• IonQ Q2 2026 (ex-SkyWater guide): investors.ionq.com (opens in a new tab)
• NIST / Commerce CHIPS finals (2026-09-08): Quantinuum, Rigetti, D-Wave, PsiQuantum; see nist.gov news for each awardee
• Quantinuum 8-K on the CHIPS award and share issuance (2026-09-08): SEC (opens in a new tab)
• Quantinuum 424B4: SEC (opens in a new tab)

Key primary links — policy

• NIST Post-Quantum Cryptography: csrc.nist.gov/projects/post-quantum-cryptography (opens in a new tab)

---

16. Earlier thinking

This page is the third pass at the same question, and the answer has moved. In October 2025 the thinking was built around large-caps (Nvidia, Microsoft, Alphabet, ASML, TSMC), one mid-cap test-equipment name, and four pure-plays: quantum as a satellite to an AI book. By January 2026 it had become nine names across three time horizons, with an explicit instruction to avoid D-Wave on the grounds that annealing is not the winner-take-all prize.

The book held today is different: eight names, seven of them pure-play quantum, with IBM the only diversified, profitable position. D-Wave is held despite January's "avoid", because its Quantum Circuits acquisition bought a gate-model path, and it is sized accordingly, at about 5%. The large-cap hedges the earlier thinking leaned on (Alphabet, Microsoft, Nvidia, Honeywell) are not in the book; they sit in the crosshairs instead. Four of the eight listed in 2026, so the portfolio is a bet on the listing wave, made when the sector opened itself to public money.

Recording the reversals is the point. A thesis that remembers only what it got right is a press release.

---

17. How this was researched, and what is not verified

The facts on this page were assembled in two passes, on 2026-09-11 and 2026-09-14, from SEC filings (10-K, 10-Q, 8-K, 20-F, 6-K, prospectuses), company investor-relations releases, peer-reviewed journals, arXiv preprints labelled as such, government program pages (NIST, DARPA, the White House, the Federal Register), and one vendor's market-data snapshot. Financial press was used only to date events. Where two sources disagreed, the primary filing won and the disagreement is noted below.

What a reader should treat as uncertain:

• The XNDU lock-up date is confirmed by a filing, though never announced. It was derived as ~2026-09-22 from filed "up to 180 days" language after the 2026-03-26 close. The CEO's Schedule 13D amendment, filed 2026-09-24, reports open-market sales dated 2026-09-22 and no other transactions in the prior sixty days, so his first sales came on the derived date. Neither that filing nor the company states the lock-up itself, so the date rests on the sales rather than on an announcement.
• The QNT lock-up is 180 days after the 424B4 prospectus date, so ~2026-12-01/02, not ~November 30. Confirm whether the clock starts 2026-06-03, 06-04, or 06-05 before treating either day as exact.
• HQ: primary shares lock-up language is 24 months, to ~2028-03-19. Do not publish a 2027-03-19 sponsor date until a distinct 12-month sponsor exhibit is quoted.
• INFQ's lock-up: settled, and it did not run to the 180-day bound. Two instruments govern it, and conflating them is what left this open. The registration rights agreement of 2026-02-13 locks holders for a flat 180 days, to 2026-08-12, with no price trigger and release only by board waiver. The bylaws carry the trigger: section 7.5 defines a "Triggering Event" as the volume-weighted average price holding at or above $12.00 for any fifteen trading days within any 180, and exempts the lock-up "from and after" it. The Maverick funds, the CEO's holding entities and the CTO all signed as Insiders, so they were bound until that event. Computed from daily prices since the 2026-02-17 first trade, the fifteenth qualifying day falls on 2026-04-17 measuring by closing price, 04-20 by the daily typical price, and no later than 04-27 counting only days whose entire intraday range cleared $12.00. The release therefore came in April on any measure, four months before the bound, which confirms the April date this page previously declined to publish on secondary sourcing. The trigger is self-executing and the company filed no 8-K announcing it, which is why it only ever appeared in secondary sources. Residual uncertainty: true VWAP needs intraday data, so the exact day inside April is approximate; the month is not.
• The Quantinuum Commerce share issuance is confirmed from the 8-K body (Items 1.01 and 3.02); the press-release exhibit alone does not mention it. Share counts are deliberately kept off this page.
• The QNT Russell 1000 / 3000 addition (2026-09-21) is secondary only as of this writing; the index provider's standing PDF is still the June list. Confirm on lseg.com or drop the line.
• Insider-sale figures in §10.6 are computed from Form 4 XML on EDGAR for 2025-09-22 to 2026-09-22 (read 2026-09-22), and whether a sale was under a Rule 10b5-1 plan is taken from each filing's own checkbox, not inferred. Where a fund and one of its partners each filed the same trades, the trades are counted once. Xanadu files as a foreign private issuer and has no Form 4s on EDGAR. Its CEO's sales surface in Schedule 13D amendments because he holds more than 5%; other insiders' activity is unverified here, and its Canadian insider filings have not been read.
• Several 2026 technical results are arXiv preprints that have not cleared peer review, including Quantinuum's logical-qubit counts and IonQ's elliptic-curve blueprint.
• Whether IBM delivers Kookaburra in 2026 is unknown; November QDC is the checkpoint.
• Market caps and multiples are one vendor's snapshot at one close; the tape tables use a second vendor's closes, which differ by small amounts (and by about $4 for IBM). Neither is authoritative.
• The HQ 20% intraday drop on 2026-09-14 is an observation after the snapshot date with no identified cause; the session had not closed when this was written.
• IONQ's combined FY26 guide bridge (organic vs SkyWater vs intercompany) should be confirmed from Investor Day slides or an 8-K exhibit before treating any specific split as precise.
• All §8 statistics are computed, not quoted, and depend on the window chosen (2026-06-05 to 09-11). A different window gives different numbers; the direction of every conclusion is robust to that, the decimals are not.

---

18. Disclaimer

This is a personal investment thesis published for informational purposes. It is not investment advice, and its author is not a licensed financial adviser. The author holds positions in the eight quantum-book securities, and may hold other securities named here, incidentally to the thesis. Quantum computing is a speculative sector in which a single position can lose more than half its value in one session, as one of these holdings did in May 2026, and in which the profitable anchor of this book lost a quarter of its value in one day. Past performance does not indicate future results.