Quantum Hardware Roadmaps
A quantum hardware roadmap is a builder’s published multi-year plan for the machines it intends to ship, and the leading ones, from IBM, Google, IonQ, and Quantinuum, now converge on a shared destination: a large, fault-tolerant quantum computer built from error-corrected logical qubits around the end of this decade. Read carefully, these roadmaps are useful evidence, because they show measurable progress on the axis that actually gates a cryptographic attack, which is error correction rather than raw qubit count. Read carelessly, they become a countdown, which is the mistake to avoid, because a roadmap is a plan under a marketing deadline, and quantum dates have slipped before. The honest way to use them is the hype filter: weigh what a vendor has demonstrated far more heavily than the year printed next to a future milestone.
Source: IBM, “IBM lays out clear path to fault-tolerant quantum computing,” June 10, 2025, ibm.com.
The short version:
- The leading roadmaps agree on the goal, a fault-tolerant machine with hundreds of logical qubits, and most publish target years around 2029 for a first large-scale system.
[OPERATOR VERIFY]every forward-looking date; roadmaps slip. - IBM targets Starling in 2029 (200 logical qubits, 100 million gates), building through Loon (2025), Kookaburra (2026), and Cockatoo (2027).
[OPERATOR VERIFY] - Google reached a below-threshold logical qubit with Willow in 2024, the milestone that matters, and frames a useful error-corrected machine around 2029, scaling toward roughly one million physical qubits.
[OPERATOR VERIFY] - Trapped-ion builders, IonQ and Quantinuum, lead on gate fidelity and publish their own late-decade fault-tolerance targets on different physics.
[OPERATOR VERIFY] - None of this is a CRQC. Breaking RSA-2048 needs thousands of logical qubits and billions of error-free operations, well beyond every roadmap’s first fault-tolerant machine.
- The date that governs your risk is your data’s shelf-life under harvest-now-decrypt-later, not a vendor’s milestone year.
Why read a roadmap through the hype filter?
A roadmap deserves the hype filter because it mixes two very different things in one graphic: milestones a company has already delivered, which are evidence, and milestones it intends to deliver, which are targets. The delivered ones tell you where the hardware actually is. The future ones tell you where a company hopes to be under commercial and competitive pressure, and that pressure is exactly what makes a printed date optimistic. Forward-looking quantum timelines have a documented history of slipping, so the correct weight on a future milestone year is real but heavily discounted.
The filter also keeps you reading the right unit. A roadmap headline that promises a large qubit count in a given year is a claim about width, and width is the axis that misleads. What decides whether a machine can break cryptography is depth run under error correction, which shows up in a roadmap as logical-qubit counts, gate fidelities, and below-threshold demonstrations. When a roadmap leads with physical qubit totals and stays quiet about logical qubits and error rates, that emphasis is itself worth noting. When it commits to a specific logical-qubit count with a demonstrated error-correction result behind it, that is the substantive part.
What is IBM’s roadmap?
IBM publishes the most detailed public roadmap, and in June 2025 it laid out a staged path to a fault-tolerant machine it calls Starling, targeted for 2029. The intermediate systems each test one hard piece of the fault-tolerance problem before the pieces are assembled, and IBM ties the plan to the qLDPC error-correcting code, which aims to cut the physical-to-logical overhead well below the surface code’s.
| System | Target | Role |
|---|---|---|
| Loon | 2025 | Test qLDPC architecture components, including long-range “C-couplers” |
| Kookaburra | 2026 | First modular processor combining quantum memory with logic |
| Cockatoo | 2027 | Entangle two Kookaburra modules via “L-couplers” |
| Starling | 2029 | 200 logical qubits running 100 million quantum gates |
| Blue Jay | future | 2,000 logical qubits running 1 billion operations |
Every year and figure in that table past 2025 is a target, so treat each as [OPERATOR VERIFY] and re-check it against IBM’s current roadmap rather than this snapshot. The substantive read is the structure: IBM is sequencing toward error-corrected logical qubits explicitly, which is the right axis, and Starling’s 200 logical qubits, if delivered on time, would still be short of the thousands a CRQC needs.
Source: IBM, “IBM lays out clear path to fault-tolerant quantum computing,” June 10, 2025, ibm.com; The Quantum Insider, “IBM Offers Roadmap Toward Large-Scale, Fault-Tolerant Quantum Computer,” June 10, 2025, thequantuminsider.com.
What is Google’s roadmap, and why is Willow the milestone that matters?
Google organizes its work as a six-milestone roadmap toward a large error-corrected machine, and the milestone worth weighting is one it has already cleared rather than one it has promised. In December 2024, Google’s Willow chip, 105 superconducting qubits, demonstrated quantum error correction below the surface-code threshold, meaning that as the team made the error-correcting code larger, the logical error rate went down rather than up. That is the property a scalable fault-tolerant machine ultimately needs, and it is the kind of result the hype filter says to weight heavily, because it is a demonstrated, peer-reviewed error-correction result, not a projection.
Source: Rajeev Acharya et al. (Google Quantum AI), “Quantum error correction below the surface code threshold,” Nature 638, 920-926, 2025, arXiv:2408.13687; Google, “Meet Willow, our state-of-the-art quantum chip,” December 9, 2024, blog.google.
The forward-looking half of Google’s roadmap is the part to hold loosely. Its final milestone is a large-scale error-corrected computer connecting on the order of one million physical qubits, folded into a much smaller number of logical qubits, and Google leadership has framed a useful error-corrected machine around 2029. Both the million-qubit figure and the 2029 framing are targets, so treat them as [OPERATOR VERIFY] and re-confirm against Google’s current roadmap. The steady read is that Willow proved the error-correction approach can scale the right way, while the machine that runs a real attack is still milestones and years out.
Source: Google Quantum AI, “Our quantum computing roadmap,” quantumai.google/roadmap.
What are the trapped-ion roadmaps, IonQ and Quantinuum?
IonQ and Quantinuum build on trapped ions rather than superconducting circuits, which changes the roadmap shape, because ion qubits lead the field on gate fidelity and coherence while scaling to large numbers is the harder problem. Both publish late-decade fault-tolerance targets, and both should be read with the same discount on future dates.
Quantinuum released a five-year roadmap toward a fault-tolerant machine it calls Apollo, targeted around 2029, moving through Helios in 2025 and Sol in 2027. It reported that its Helios system carries 98 physical qubits, 48 logical qubits, and a two-qubit gate fidelity of 99.921%, which is among the highest fidelities any modality has demonstrated. IonQ delivered its 2025 milestone of AQ 64 (an application-oriented benchmark) on its Tempo system, and publishes a stepwise plan through roughly 10,000-qubit chips in 2027 and multi-chip modules with on the order of 1,600 logical qubits in 2028, toward a target of millions of physical qubits and tens of thousands of logical qubits by 2030.
Every future figure here is a vendor target under competitive pressure, so treat the 2027, 2028, and 2030 milestones as [OPERATOR VERIFY] and prefer the demonstrated numbers (Helios fidelity, IonQ’s AQ 64) over the projected ones. The genuine point is that trapped ions pursue the same fault-tolerant destination from a fidelity-first direction, so no single modality’s roadmap is the only path.
Sources: HPCwire, “Quantinuum Introduces Helios Quantum System as Roadmap Advances Toward Apollo,” November 5, 2025, hpcwire.com; IonQ, “IonQ Hits AQ 64 Milestone Ahead of Schedule,” 2025, ionq.com.
How do the roadmaps compare?
Laid side by side, the leading roadmaps converge on a fault-tolerant machine around the end of the decade while differing on the physics and the path. The table reflects each builder’s published plan as of 2026, and every future entry is a target rather than a delivered result.
| Builder | Modality | Demonstrated milestone | First fault-tolerant target | Notable future figure |
|---|---|---|---|---|
| IBM | Superconducting | 1,121-qubit Condor (2023) | Starling, 2029 [OPERATOR VERIFY] | 200 logical qubits [OPERATOR VERIFY] |
| Superconducting | Below-threshold logical qubit, Willow (2024) | Useful machine ~2029 [OPERATOR VERIFY] | ~1 million physical qubits [OPERATOR VERIFY] | |
| Quantinuum | Trapped ion | Helios, 48 logical qubits, 99.921% fidelity (2025) | Apollo, ~2029 [OPERATOR VERIFY] | Hundreds of logical qubits [OPERATOR VERIFY] |
| IonQ | Trapped ion | AQ 64 on Tempo (2025) | Late-decade [OPERATOR VERIFY] | Millions of physical qubits by 2030 [OPERATOR VERIFY] |
Source: IBM, June 10, 2025, ibm.com; Google, quantumai.google/roadmap; Quantinuum, November 5, 2025, hpcwire.com; IonQ, 2025, ionq.com.
The convergence around 2029 is striking and should be read with care, because a shared target year across competitors reflects a shared ambition as much as a shared certainty. What the table shows more reliably is direction: every serious builder is now sequencing toward logical qubits and error correction, which is the axis that matters, and the demonstrated-milestone column is where the real evidence lives.
Do these roadmaps mean a code-breaker is close?
No, and the gap between a first fault-tolerant machine and a code-breaker is the whole reason. The most-cited peer-reviewed estimate puts breaking RSA-2048 with Shor’s algorithm at roughly 6,100 logical qubits, realized as about 20 million noisy physical qubits in a 2021 construction and driven under 1 million in a 2025 optimization, running a computation billions of operations deep without an uncorrected error. Set that against the roadmaps: a first fault-tolerant system with 200 logical qubits, if delivered on schedule, is more than an order of magnitude short in logical qubits and further still in the sustained error-free depth an attack needs.
Sources: Craig Gidney and Martin Ekerå, “How to factor 2048 bit RSA integers in 8 hours using 20 million noisy qubits,” Quantum 5, 433, 2021, arXiv:1905.09749; Craig Gidney, “How to factor 2048 bit RSA integers with less than a million noisy qubits,” 2025, arXiv:2505.15917.
So a roadmap milestone landing on time would be real progress toward a CRQC without being one. The resource estimates keep dropping as the engineering improves, which means the threat timeline is a moving band rather than a fixed wall, and credible expert and government assessments still span roughly 2030 to 2040 and beyond. The roadmaps sharpen that band a little; they do not collapse it into a date.
How should a security team actually use a roadmap?
A security team should use a roadmap as one input into a Mosca’s-theorem risk calculation, never as the deadline itself. The deadline that governs your migration is set by your own data: how long it must stay confidential, plus how long your migration takes, weighed against when a capable machine plausibly arrives. A vendor roadmap informs only that last term, and it informs it as a band, so the sound posture is to plan around lead time rather than bet on a printed year.
The practical discipline is three-part. Track the demonstrated milestones, because a below-threshold logical qubit or a new fidelity record is real evidence that shortens the plausible band. Discount the future dates, because they slip and re-checking them is the point of the [OPERATOR VERIFY] flags above. And keep migrating regardless, because harvested data is exposed the day a CRQC turns on, whenever that is, and a large-estate migration takes years. A roadmap that accelerates should tighten your schedule; a roadmap that slips should not relax it, because your harvesting exposure is live today.
Common misconceptions
- “A vendor roadmap is a countdown to breaking RSA.” It is a plan for a fault-tolerant machine, not a code-breaker, and its future dates are targets under commercial pressure. Breaking RSA-2048 needs far more logical qubits and depth than any first fault-tolerant system on these roadmaps.
- “The company with the most physical qubits is winning the race.” Physical qubit count is width, and it does not compare across modalities. The axis that decides capability is logical qubits below the error-correction threshold, which is why Google’s Willow result matters more than a raw count.
- “They all say 2029, so 2029 is when quantum breaks encryption.” A shared target year reflects shared ambition, and even a delivered 2029 fault-tolerant machine is not a CRQC. Independent expert timelines still span 2030 to 2040 and beyond.
- “If the roadmaps slip, I can relax my migration.” No. Your exposure is set by harvesting and your own migration time, not by the vendor’s schedule. Data recorded today is at risk whenever a machine eventually arrives.
- “Trapped-ion builders are behind because they have fewer qubits.” Trapped ions lead on gate fidelity and coherence, which are the axes that build logical qubits. Their smaller physical counts reflect a harder scaling problem on different physics, not lower quality.
Questions people ask
When do the roadmaps say a fault-tolerant quantum computer arrives? Most leading builders publish targets around 2029 for a first large-scale fault-tolerant machine, IBM’s Starling and Quantinuum’s Apollo among them. Treat every such date as a target that has slipped before, and re-verify it against the current roadmap rather than a snapshot.
Does a 2029 machine break RSA? No. A first fault-tolerant system on these roadmaps is targeted at a few hundred logical qubits, while breaking RSA-2048 needs thousands of logical qubits and billions of error-free operations. It would be real progress toward a CRQC while still falling well short of one.
Which company is closest to breaking encryption? None is close, and the question is better asked in logical qubits than physical ones or brand names. The builder that first fields thousands of clean logical qubits below the error-correction threshold gets there, and as of 2026 all are far from that, with Google having demonstrated a single below-threshold logical qubit.
Why do the roadmaps keep the physical-qubit numbers so large? Because building one reliable logical qubit takes many physical ones under error correction, on the order of hundreds to thousands each depending on hardware quality. A million-physical-qubit target folds down to a much smaller number of logical qubits.
Should a vendor roadmap change my migration plan? Only at the margin. A roadmap accelerating or a new error-correction milestone landing should tighten your Mosca band, but the migration itself is driven by your data’s confidentiality lifetime and your own timeline, so harvesting keeps the work urgent no matter what a roadmap says.
How reliable are these dates? The demonstrated milestones are reliable because they happened and were peer-reviewed or independently reported. The future dates are vendor targets under competitive pressure and have a track record of slipping, which is why each carries an [OPERATOR VERIFY] flag here and belongs in a plan only as a band, not a fixed year.
Go deeper
The hype filter these roadmaps are read through: How to Tell Real Quantum Progress From Hype is the board-grade checklist for a quantum headline, physical versus logical qubits, fidelity over count, demo versus fault-tolerant machine.
The hardware underneath: Quantum Hardware Modalities (superconducting versus trapped-ion versus photonic versus neutral-atom) · Logical vs Physical Qubits · Quantum Error Correction · Quantum Fidelity · The Threshold Theorem · NISQ (Noisy Intermediate-Scale Quantum)
The threshold and the timeline: Cryptographically Relevant Quantum Computer (CRQC) · Quantum Threat Timeline · Quantum Resource Estimation · Mosca’s Theorem
Everything here is the map, given freely. When your team needs the roadmap dates translated into a risk band and a dated migration plan for your own systems, that’s what an alignment briefing is for.
Last verified 2026-07-12 · Maintained by Addie LaMarr, LaMarr Labs.