Quantum-computing SaaS on a single VPS — market research

Date: 18 September 2026
Question: Is any quantum-computing SaaS viable on this Ubuntu VPS (94 GB RAM, no QPU, nginx 80/443 only), for US customers, legally clean, and profitable in 12 months?
Verdict: KILL. Do not ship a quantum product on this VPS. The commercial layer that makes money requires hardware, hyperscaler contracts, or PhD-depth domain embedding we do not have. Everything that can run here is already free, better, and owned by IBM, Google, AWS, Xanadu, or qBraid.

This is not a “maybe education” hedge. Education is also a kill.

1. Market size and commercial reality in September 2026

Separate three numbers that get mixed in press releases.

Use vendor revenue. The trillion-dollar figure is a consulting TAM, not a bill anyone pays.

TQI segment split (Q3 2026, base case)

McKinsey’s QC company revenue path is similar in order of magnitude: $1.1–1.4B (2025) → $1.5–2.0B (2026) → $3.2–4.4B (2028). TQI’s 2030 bear/bull is $3.2B / $9.5B. The spread is one variable: when error correction makes fault-tolerant machines commercially useful.

Ignore the $3.5B–$74B “QCaaS market” PDFs from MarketIntelo / SNS Insider / ResearchIntelo. They mix hardware leases, PQC, consulting, and vapor. TQI’s $0.25B QCaaS in 2025 is the honest number. Cloud access is the fastest-growing slice, not a $10B 2026 market.

Geography of vendor revenue (TQI, 2030): Americas 59%, EMEA 32%, APAC 9%. McKinsey’s adopter mix is inverted: Europe 43% of analyzed customers, US 29%, Asia 22%. US buyers exist; they buy from IBM, IonQ, AWS, Azure, Quantinuum — not from a Contabo/IONOS VPS in Germany.

What “commercial tipping point” actually means

McKinsey’s 2026 Monitor title is “A commercial tipping point.” The facts underneath are more modest:

• >300 large companies collaborating with quantum vendors (Airbus, JPMorgan, Boehringer Ingelheim cited).
• 33% of analyzed firms spend >$10M/year on quantum; 7% spend >$50M; one budget hits $200M.
• 72% of activity is at privately owned companies (shift off public labs).
• $12.6B invested in quantum-tech startups in 2025 (6.3× 2024); >90% into computing; ~60% of 2025 volume in the ten largest deals.
• Capital is concentrating, not democratizing. This is a market for IBM, IonQ, Quantinuum, Google, AWS, PsiQuantum, and a handful of $100M+ software houses.

IBM’s public line (Sep 2026 hardware page): “IBM and partners delivered advantage as promised in 2026.” Read the fine print. IBM’s Quantum Advantage Tracker (Algorithmiq, Flatiron, BlueQubit) is hunting first verified examples of advantage on synthetic / doped-Clifford / sampling workloads. Independent write-ups of Nighthawk dual tests (June 2026) and the UChicago doped-Clifford result (Aug 2026) are explicit: no commercial application, no customer economics moved. Quantum Litmus’s rolling commercial-readiness window as of Sep 2026 is still 2030–2033.

IBM Quantum Starling — 200 logical qubits, 100 million gates, Poughkeepsie data center — remains 2029. Two cryogenic modules were linked in Aug 2026. That is engineering, not a product we can sit in front of.

Commercial reality, one sentence: in September 2026, paying customers buy access to someone else’s fridge, national-lab / Fortune-100 PoCs, and professional services. They do not buy a 32-qubit statevector notebook from an unknown SaaS.

2. What actually makes money today vs 2029

Money today (2025–2026)

Hardware and system sales. TQI tracks 204 disclosed deals (2013–early 2026): 144 system sales, 18 leases, 9 cloud-access contracts, 2 software sales. IBM is first on tracked value (~$732M, ~$66M average). IQM leads on count (25). IonQ sold a Superion 256-qubit system to Cambridge in Q1 2026. D-Wave closed a $10M / 2-year Fortune-100 QCaaS contract (Jan 2026) and a $100M US Commerce/CHIPS R&D agreement (Sep 2026). This is capital equipment and sovereign procurement. We cannot do it.

Cloud QPU access (QCaaS). IBM Quantum Platform, AWS Braket, Azure Quantum, IonQ Cloud, D-Wave Leap, Quantinuum via Azure. IBM list prices verified Sep 2026:

That is the product enterprises pay for: minutes on a real QPU. IBM’s free Open Plan already includes unlimited simulators and 10 minutes on 100+ qubit hardware. A VPS cannot undercut “free real hardware.”

IonQ guided $280–290M for 2026 before folding in SkyWater (foundry). After the July 31 close, FY2026 guidance is $450–460M — most of the raise is not quantum computing. Motley Fool (8 Sep 2026) is blunt about this. Public “quantum” revenue is being padded with fabs, networking, and cybersecurity roll-ups.

Professional services. TQI: services were $0.61B in 2025 and barely grow to $1.03B in 2030 (11% CAGR). Integration work does not scale. It also requires domain PhDs sitting with the customer. Not a VPS subscription.

Software that is actually attached to a QPU. Classiq (~$200M raised, ~103 employees, “tens of millions” revenue rumored, AWS Marketplace, BMW/Citi/Rolls-Royce logos). Q-CTRL (error suppression / Fire Opal, defence navigation). Riverlane (error-correction stack). Quantum Machines (control electronics). These are compiler / control / QEC companies with hardware partnerships. Novo Holdings (Sep 2026) warns that compilers, runtimes, and general developer tools are likely to be absorbed as features by hardware vendors and clouds, not durable independent businesses — unless one becomes the Arm/CUDA of quantum, which Classiq is trying to be with $200M.

Software that is not attached to a QPU dies. Zapata Computing (the cautionary tale): SPAC 2024, operations ceased Oct 2024, entire staff terminated, rebranded Zapata Quantum, $0 revenue in 2025 and H1 2026, $15M rescue round Apr 2026, OTC, going-concern language, accumulated deficit ~$118M. They were the “hardware-agnostic quantum software / application layer” story. The market did not pay.

Money in 2029–2035 (if roadmaps hold)

• IBM Starling (2029): 200 logical qubits / 100M gates. First plausible window for narrow chemistry / materials kernels inside classical workflows.
• Novo Holdings (11 Sep 2026): ~70% of $13.9B private QC capital 2014–2025 went to hardware. Application layer is underfunded relative to future value, not relative to today’s willingness to pay. Their investable company is one that (a) picks a molecular bottleneck, (b) formulates it, (c) reduces resource requirements, (d) validates against experiment, (e) embeds in a pharma workflow before FTQC arrives. That is a 5–10 year, PhD-staffed, customer-by-customer grind. Early revenue looks like custom scientific work, not a $29/mo SaaS.
• TQI: QCaaS becomes 33% of vendor revenue by 2030 because buyers rent QPUs. The rent still goes to IBM/IonQ/AWS.
• Fault-tolerant commercial chemistry/finance: late 2020s / early 2030s at the optimistic end; Quantum Litmus still says 2030–2033. Betting a 12-month P&L on 2029 hardware is how Zapata burned $118M.

What does not make money

• Circuit classrooms, Bloch-sphere toys, “learn Grover’s algorithm” playgrounds.
• Generic quantum notebooks (IBM sunset its own Quantum Lab in Nov 2024 and pointed learners at qBraid).
• CPU statevector simulators past ~30 qubits (see §5).
• “Quantum advantage as a service” marketing. Advantage in 2026 is a benchmark, not a SKU.
• Thin resale of Braket/IBM minutes. AWS and IBM already sell those. Reselling model access as a prepaid wallet also violates the brief.

3. Verdict: KILL

KILL the entire quantum-computing SaaS direction for this VPS.

Not “kill QCaaS but keep a classroom.” Kill all of it as a CryptoBook pivot. It fails the brief on demand, wedge, 60-second value, 12-month profitability, and honesty. Building it would be security-theater-adjacent: a black/pink site that looks like quantum because the word is hot, while the actual compute is numpy on 94 GB of RAM.

Instant-kill checklist vs BRIEF.md

The brief already flagged this: “A quantum SaaS on this VPS is likely a kill unless it is purely educational/software with no quantum hardware claim.” The research confirms the unless-clause fails too.

4. Product shapes — all dead, including the “education” escape hatch

Evaluated as if we were trying to survive. None survive.

A. QCaaS / “run on a quantum computer” — KILL

We have no QPU. Renting IBM/Braket/IonQ and wrapping a UI is a reseller. IBM PayGo is $96/min; Open Plan is free. AWS Braket and Azure Quantum already aggregate IonQ, Rigetti, QuEra, IQM, Quantinuum. qBraid already does notebooks + multi-QPU with a $0 / $20 / $100 seat model and is the IBM-recommended lab host. We would be a worse qBraid with one server in Germany.

B. Hosted quantum simulator — KILL

Physics of this box:

Statevector RAM ≈ 16 bytes × 2^n (complex128).

So the honest ceiling is ~31 qubits if we leave room for OS, nginx, Postgres, and concurrent users. One 32-qubit job monopolizes the machine. Tensor-network / MPS tricks can go higher on structured circuits and collapse on generic ones.

This is what a laptop already does. Qiskit Aer, Cirq, PennyLane, CUDA-Q, QuTiP are free and local. IBM Open Plan simulators are unlimited. Classiq ran a 36-qubit simulation on Oracle Cloud (May 2026). NVIDIA sells GPU-accelerated simulation. A CPU VPS in 2026 is not a product; it is a homework server.

C. Algorithm notebooks / “Colab for quantum” — KILL

IBM Quantum Lab was sunset November 2024. Replacement: qBraid Lab. qBraid Free = 4 GB / 2 vCPU / 25 CPU-hours; Standard $20/mo; Pro $100/mo; GPU SKUs at $3–5/hr. Google Colab + pip install qiskit is free. We would be cloning a company IBM already anointed, at lower RAM than their Pro tier, with no QPU backends.

D. Quantum circuit classroom / Bloch sphere / “60-second Grover” — KILL

This is the brief’s candidate D, and it is the one people reach for when they want a soft landing.

• IBM Qiskit Global Summer School 2026 (13–24 July) is free, two weeks, Discord, badges, a beginners-only track. Registration filled in days.
• IBM Learning courses and the Qiskit textbook are free. IBM Composer is a visual circuit playground, free, on real or simulated backends.
• Coursera / edX / MIT / Q-CTRL Black Opal / SpinQ classroom hardware (~$5k Gemini Mini, 200+ institutions) already occupy education.
• The paying education buyers are universities buying kits or site licenses, not consumers swiping a card after 60 seconds. That is a sales-cycle business.
• Willingness-to-pay for “another Bloch sphere” is approximately zero. The job-market signal is “know Qiskit,” which IBM teaches for free because they want developers on their hardware.

A pretty Virtus-themed circuit toy can be built in a weekend. It will not be profitable in 12 months. It will not have high demand. It is a portfolio piece, not a SaaS.

E. Application-layer algorithms (chemistry, portfolio, logistics) — KILL

Novo Holdings (Sep 2026) is the best argument for this layer, and even they say:

• Value is in identifying a bottleneck, formulating it, reducing resources, validating, embedding.
• Early FTQC will not model a drug end-to-end; it will do a small electronic-structure kernel.
• Early revenue is custom scientific work, reused into a platform over years.
• Classical DFT/CCSD and AI already occupy the budget; quantum must beat them on a specific calculation.
• Life sciences is attractive if you are Novo-scale and already sit inside pharma workflows.

We have no computational chemists, no experimental validation loop, no QPU, no 5-year runway. Phasecraft (Novo co-led $34M Series B, Sep 2025) and Algorithmiq exist to do exactly this. Zapata tried the generic version and went to zero revenue.

F. Compiler / “describe the algorithm, we synthesize the circuit” — KILL

That is Classiq 1.0 (Feb 2026), $200M+ in, AI agents that write programs (Apr 2026), AWS Marketplace. Horizon Quantum is building languages (Helium/Hydrogen/Beryllium). A two-person VPS rewrite is not a wedge.

G. Post-quantum cryptography tooling — not quantum computing; still a weak idea

PQC is classical cryptography (NIST ML-KEM / ML-DSA / SLH-DSA) that defends against future quantum computers. It is the only “quantum” word US companies are actually being forced to budget for:

• EO 14412 (22 Jun 2026): federal HVAs to PQC key-establishment by 31 Dec 2030, signatures by 31 Dec 2031.
• Federal contractors will see procurement clauses. SMB urgency in 2026 is still low.
• Vendors: PQShield, SandboxAQ, IBM, Cloudflare, AWS, Google, Microsoft, OpenSSL/liboqs. Crowded.
• Brief forbids fake compliance certificates and “we make you AI Act compliant.” “We make you PQC-compliant” is the same landmine.
• Crypto discovery / CBOM products (Qtonic and others) are security consultancies, not a 60-second playground.

If someone later wants a classical crypto product, research it as cryptography, not as quantum computing. Do not smuggle it in under this verdict.

5. Demand, competition, regulation, cost — honest assessment

Demand

Real demand (paying): Fortune 100 / national labs / defence for QPU minutes, on-prem systems, and services. Deal sizes: $10M QCaaS contracts, $28.8k–$250k IBM annual floors, $66M average IBM system deals. Buyer is a quantum working group with a 2029 roadmap, not a developer who found us on Product Hunt.

Fake demand (press): “300 companies adopting,” “$2.7T by 2035,” “quantum advantage delivered 2026.” Adoption means PoCs and cloud accounts. Advantage means a sampling benchmark. The $2.7T is value-at-stake if hardware works.

Consumer / indie-dev demand for paid simulators or classrooms: near zero. IBM, Google, Xanadu, and qBraid saturated it with free or $20/mo goods. Qiskit is the resume keyword; we cannot become the resume keyword.

US customers specifically: TQI books 59% of 2030 vendor revenue in the Americas, but that revenue is captured by US-headquartered hardware and cloud vendors. A German VPS with a .space domain is not in the procurement set.

Competition (anyone we would actually meet)

A Langfuse clone is “not genius” per the brief. A Qiskit clone is worse: Qiskit is MIT-licensed and IBM wants you to use it.

Regulation (export controls, ITAR, cloud loopholes)

This is not theoretical.

EAR / BIS (Sep 2024 IFR, still in force 2026):

• ECCN 4A906 controls quantum computers at qubit/error thresholds (from 34 fully-controlled connected working physical qubits with C-NOT error ≤ 10⁻⁴, stepping up). Related assemblies, control electronics, software (4D906) and technology (4E906) for development/production of those systems.
• Also 3A901 / 3D901 (cryo CMOS, related software), dilution refrigerators, wafer probers.
• License exceptions for countries with equivalent controls (UK, NL, JP, AU, NO, etc.). Worldwide NS/RS licensing on the high-end items.
• Academic NISQ toys often sit below the 4A906 thresholds, but “check the current parameters, do not assume.”

ITAR: defence-funded quantum sensing / comms / some algorithms can be USML. A public SaaS that “helps design quantum algorithms for crypto-breaking” is how you get a conversation with DDTC you do not want. Shor’s algorithm notebooks as a product aimed at “break RSA” would be insane. Even as a classroom, marketing that way is reckless.

Remote Access Security Act (RASA): House passed H.R. 2683 on 12 Jan 2026, 369–22. Senate companion S. 3519 pending as of mid-2026. 2024 draft text explicitly included remote access to quantum computers. Even if the enacted version focuses on AI chips, BIS is being handed the authority to treat cloud access as an export. A non-US VPS offering “quantum computing” to the world, including Country Group D, is the fact pattern they are closing.

EO 14413 (22 Jun 2026), “Ushering in the Next Frontier of Quantum Innovation”: national QC-ADDS effort, supply-chain, harmonized export controls and outbound-investment restrictions with allies, expanded FBI QIST counterintelligence team. Combined with the Jan 2025 outbound-investment rule already barring US-person investment in Chinese quantum computing.

Practical meaning for this VPS:

• A purely classical open-source-simulator website serving US persons is probably EAR99 / not 4A906 — we are not exporting a QPU. That is the best case.
• The moment we proxy IBM/Braket/IonQ, we inherit their AUPs, KYC, and export screening. IBM and AWS already block embargoed destinations. We would have to too, as a reseller, with no legal team.
• The moment we market “cryptographically relevant” anything, we are in the national-security conversation.
• Hosting in Germany, selling to US customers, advertising quantum: extra KYC, geo-blocking, and ToS work for a product that does not make money.

Regulation is not the primary kill (demand is). It is a secondary kill: even the toy has a compliance tail we should not grow.

Cost (ours and the user’s)

Our cost to pretend:

• VPS is already paid. Node/Python/Postgres exist. A Qiskit-Aer playground is a few days of work.
• The real cost is opportunity: 12 months of CryptoBook pivot spent on a market where Zapata did $0 after $118M, while agentic-AI SaaS is where Bain’s $4–6B captured / $100B opportunity actually sits in 2026.
• GPU for >30-qubit or noisy simulation: not on this machine; renting A100s destroys the “single VPS, high margin” constraint.
• US legal review for export/ITAR/ToS if we say the word “quantum” loudly: thousands of dollars, recurring.

User cost vs alternatives:

• IBM Open Plan: $0, real 156-qubit Heron, unlimited simulators.
• pip install qiskit on a laptop: $0, 28–30 qubits.
• qBraid Standard: $20/mo, notebooks + QPU proxy.
• IBM PayGo if they actually need a QPU: $96/min, no middleman.

There is no price we can charge where the user is better off and we have margin. That violates brief item 6 twice.

12-month P&L sketch (why “profitable in 12 months” is fantasy)

Assume a miracle: 2,000 signups, 2% convert to $12/mo classroom Pro, 40 paying seats = $480 MRR. Churn vs free IBM is high. Stripe + support + any GPU burst and we are at best coffee money. Enterprise “quantum training for your team” is a sales-call product (violates item 5) and collides with IBM’s free Summer School and paid IBM Quantum Network.

No honest model reaches a meaningful SaaS margin in 12 months without lying about hardware or reselling QPU minutes.

Bottom line

KILL.

Quantum computing in September 2026 is a hardware-and-cloud market of ~$1.5–2B vendor revenue, concentrating into IBM, IonQ, Google, AWS, Azure, Quantinuum, and a few $100M software specialists. Fault-tolerant machines that could support a real application layer are IBM Starling 2029, not this year. Novo Holdings is right that applications are underfunded — and wrong as a cue for us: that layer needs domain scientists, experimental validation, and QPU access, and Zapata already died trying the software-only version.

On this VPS, without a QPU, the only honest product is a ≤32-qubit classical simulator or a circuit classroom. Both are free from IBM/Google/Xanadu/qBraid, including free time on 100+ qubit hardware. Education does not save it. PQC is a different product and a different research memo.

Do not build quantum. Spend the pivot on an agentic-AI wedge that can actually collect a subscription on one box.

Sources (accessed 18 Sep 2026)

1. McKinsey, Quantum Technology Monitor 2026: A commercial tipping point (28 Apr 2026), https://www.mckinsey.com/capabilities/mckinsey-technology/our-insights/mckinsey-quantum-technology-monitor-2026-a-commercial-tipping-point — also PDF excerpts and DE press release: >$1B 2025 QC revenue, $1.5–2.0B 2026, $3.2–4.4B 2028; $12.6B 2025 investment (6.3×); $1.3–2.7T value at stake 2035; 300+ adopting companies.
2. The Quantum Insider, How big is the quantum computing market? (15 Sep 2026, Q3 2026), https://thequantuminsider.com/how-big-is-the-quantum-computing-market/ — $1.79B 2025 vendor revenue; hardware/cloud/services split; $5.41B 2030 base.
3. The Quantum Insider, Novo Holdings Report: Quantum Investment Must Move Beyond Hardware to Build Applications (11 Sep 2026), https://thequantuminsider.com/2026/09/11/novo-holdings-report-quantum-investment-must-move-beyond-hardware-to-build-applications/ — 70% of $13.9B 2014–2025 private capital into hardware; application-layer control-point thesis; life-sciences caveat.
4. IBM Quantum hardware & roadmap, https://www.ibm.com/quantum/hardware — Starling 2029, 200 logical qubits / 100M gates; “advantage as promised in 2026.”
5. IBM Quantum Nighthawk r2 blog (31 Aug 2026); IBM Quantum pricing https://ibm.com/quantum/pricing — Open / $96/min / Flex / Premium.
6. IonQ FY2026 guidance $450–460M incl. SkyWater (8 Sep 2026); Motley Fool analysis that most of the raise is not QC.
7. Zapata Quantum SEC / 10-Q 2026: $0 revenue 2025 and H1 2026; ~$118M accumulated deficit; operations ceased Oct 2024.
8. BIS ECCN 4A906 / 4D906 / 4E906 (6 Sep 2024 IFR); IISS Securing the quantum stack (Jul 2026); CASRAI export-control guide (updated 14 Sep 2026).
9. Remote Access Security Act, H.R. 2683 passed House 12 Jan 2026 (369–22); S. 3519; Freshfields / Latham analyses. 2024 draft included quantum computers.
10. EO 14412 & EO 14413 (22 Jun 2026) — PQC federal deadlines 2030/2031; QC-ADDS; export-control harmonization.
11. qBraid pricing (docs.qbraid.com / qbraid.com/pricing) — Free / $20 / $100; IBM-recommended notebook host after Quantum Lab sunset Nov 2024.
12. Classiq: $200M+ raised; Classiq 1.0 Feb 2026; AWS Marketplace; 36-qubit OCI sim May 2026.
13. Quantum Litmus archive (Sep 2026) — commercial readiness window still 2030–2033.
14. IBM Qiskit Global Summer School 2026 — free, 13–24 Jul 2026.
15. TQI How many quantum computers have been sold? (15 Sep 2026) — 204 deals, 2 software sales.