Quantum Hardware: State of the Art

This is the page on this site most likely to age. The previous quantum computing subpages cover physics, algorithms, error correction principles, and architecture fundamentals — all of which have shelf lives measured in years or decades. This page is a snapshot of where actual quantum hardware sits in mid-2026, and the snapshot will be wrong in detail within a year. The intention is to refresh it periodically; the framing throughout is “here is what is true today, not what will be true tomorrow.”

The audience for this page is a senior engineer or technical decision-maker who wants the concrete picture: which vendors operate which systems, what those systems can actually do, where the cloud platforms provide access, and how the capability frontier compares to the application thresholds that matter (cryptographic relevance, chemistry relevance, the ongoing post-quantum migration timeline). It is the practical complement to the Quantum Computing umbrella and the deep dives in Qubit Architectures, Quantum Algorithms, Quantum Mechanics Fundamentals, and Quantum Error Correction.

The metrics that actually matter

Raw qubit count is the most-cited quantum computing metric and one of the least informative. A 1000-qubit system with 95% two-qubit gate fidelity is less capable for most algorithms than a 50-qubit system with 99.9% fidelity, because the larger system cannot execute deep enough circuits to do anything useful. The metrics that actually track capability are several:

Two-qubit gate fidelity is the dominant determinant of useful algorithm depth. Below 99% fidelity, the system is above the surface code threshold and cannot benefit from error correction. Above 99%, error correction starts to work; above 99.9%, the per-qubit overhead for error correction becomes manageable.

Single-qubit gate fidelity matters but is usually high enough not to be the bottleneck — 99.9%+ is typical in any production system.

Coherence times (T1, T2) determine how long quantum states survive between operations. For superconducting systems, T1 ≈ 100-500 μs is typical and T2 is somewhat shorter. For trapped-ion and neutral-atom systems, coherence times are seconds to minutes. The relevant operational ratio is coherence time to gate time, which gives the rough number of operations that can run before decoherence dominates.

Connectivity matters because algorithms that need to entangle distant qubits pay a cost in SWAP operations on nearest-neighbor architectures, accumulating gate errors. All-to-all connectivity (trapped ions) or reconfigurable connectivity (neutral atoms) is structurally advantageous over fixed nearest-neighbor connectivity (superconducting).

Measurement fidelity determines how accurately the final classical readout reflects the underlying quantum state. Typical numbers are 95-99% for individual qubit measurements in production systems.

Quantum volume (IBM’s composite metric) attempts to combine qubit count, gate fidelity, and connectivity into a single number that scales with algorithm capability. Quantum volume of 2^n means the system can reliably execute an n-qubit-by-n-depth random circuit. Modern IBM systems achieve quantum volumes in the range of 2^9 to 2^11; Quantinuum systems exceed 2^15.

Algorithmic qubits (AQ) is IonQ’s metric, measuring the number of qubits that can be used together effectively in a benchmark suite of algorithms. AQ is more conservative than raw qubit count and tracks practical capability better. Current production AQ counts in IonQ systems are in the 30-40 range.

CLOPS (Circuit Layer Operations Per Second) is IBM’s throughput metric, measuring how many circuit layers can be executed per second. Relevant for variational algorithms and other workloads with many circuit executions.

Logical qubit count is the metric that will eventually replace physical qubit count as error correction matures. As of mid-2026, no commercial system has been advertised with logical qubit specifications in the standard way physical qubit counts are advertised, but the metric will appear within the next few years as fault-tolerant systems come online.

The right framing: physical qubit count is one number among many, and a system marketing only its qubit count is signaling that the more meaningful numbers are not flattering. Modern vendor disclosures include the broader set of metrics, with the field gradually converging on common reporting standards.

IBM

IBM operates the largest quantum hardware program by total accessible qubit count and has the most aggressive publicly-stated roadmap.

Current production systems (mid-2026):

• Heron r1 (133 qubits) — first deployed late 2023, focused on improved fidelity over predecessors.
• Heron r2 (156 qubits) — deployed in 2024, with two-qubit gate fidelities pushed below 0.5% error per gate on the best edges. The basis for current IBM Quantum production access.
• Quantum System Two — modular architecture connecting multiple Heron processors with cryogenic couplers. First deployed in late 2023 at IBM Yorktown Heights; additional installations have followed at partner facilities.

Notable systems on the trajectory:

• Eagle (127 qubits, 2021) — first IBM system above 100 qubits, still operating for legacy applications.
• Osprey (433 qubits, 2022) — large but with mixed quality; effectively retired from active production.
• Condor (1,121 qubits, 2023) — the largest single-chip superconducting processor at the time of announcement. Engineering testbed rather than production system.

Announced roadmap:

• Flamingo (~1,386 qubits across multiple chips, 2025-2026) — modular scaling using Quantum System Two architecture.
• Kookaburra (~4,158 qubits across three Flamingo processors) — next generation modular system.
• Blue Jay (~10,000 logical qubits, targeted 2033) — the long-term roadmap target, explicitly transitioning from physical to logical qubit counts.

Two-qubit gate error rates on best Heron r2 edges: approximately 0.3-0.5% per gate. The best edges of the best Heron systems are now solidly below the surface code threshold, with the implication that IBM can begin meaningful error correction work on these systems.

Access: IBM Quantum cloud platform with a free open-access tier (limited monthly compute) and paid Premium / Enterprise tiers for higher throughput and access to the latest hardware. Programming primarily through Qiskit, the open-source quantum software framework.

Google Quantum AI

Google’s program has prioritized error correction demonstrations over raw qubit count scaling, and the December 2024 Willow result is currently the most significant single demonstration in the field.

Current production system:

• Willow (105 qubits, December 2024) — the system that demonstrated below-threshold quantum error correction. Two-qubit gate fidelities around 99.5%, surface code experiments at distances 3, 5, and 7.

Notable predecessor:

• Sycamore (53 qubits, 2019) — the system used for the original “quantum supremacy” demonstration. Subsequent classical algorithm improvements eroded the claimed gap but the demonstration remains historically significant.

Roadmap: Google’s public roadmap is less detailed than IBM’s in terms of specific qubit counts and dates. The strategic direction is explicitly toward fault-tolerant computing, with intermediate milestones organized around logical qubit demonstrations rather than physical qubit milestones. The published targets include “long-lived logical qubit” (achieved with Willow at distance 7) and “first useful fault-tolerant algorithm” without committed dates.

Access: Google’s quantum hardware has historically been accessed primarily through research partnerships and Google internal use. Limited external access through Google Quantum AI partnerships. Programming through Cirq, Google’s open-source quantum framework. The contrast with IBM’s open cloud access is deliberate — Google’s program is research-oriented rather than commercial-platform-oriented.

Quantinuum

Quantinuum (the result of the merger between Honeywell Quantum Solutions and Cambridge Quantum in 2021) operates the trapped-ion systems with the highest published per-qubit fidelities in any production quantum computing platform.

Current production systems:

• H1 (20 qubits) — first generation system, still operating for some workloads.
• H2 (56 qubits) — current flagship system. Two-qubit gate fidelities above 99.9% on the best operations. Has demonstrated fault-tolerant logical qubit operations with logical fidelity exceeding the underlying physical fidelity, a critical milestone on the path to fault tolerance.

Notable announcements:

• Helios — next-generation system in development, with public roadmap targeting larger qubit counts and continued fidelity improvements.

Two-qubit gate fidelities: 99.9%+ in published demonstrations, with the best operations approaching 99.95%. This is the highest published two-qubit fidelity in production quantum hardware.

Quantum volume: Quantinuum has consistently led the industry in published quantum volume numbers, exceeding 2^15 (32,768) in 2023 and continuing to advance.

Architecture: Linear ion trap with QCCD (Quantum Charge-Coupled Device) architecture — ions are physically shuttled between specialized zones for gates, storage, and measurement. The QCCD approach provides a clean scaling path through inter-zone movement rather than through larger single traps.

Access: Available through Microsoft Azure Quantum and through Quantinuum’s direct cloud platform. Programming through TKET (Quantinuum’s compiler) and Qiskit / Cirq via cross-framework adapters.

IonQ

IonQ is the other major trapped-ion company, with a different architectural emphasis from Quantinuum (Penning-style traps initially, with newer systems using related but distinct trap designs).

Current production systems:

• Forte (32 qubits) — production system through 2023-2024 with Algorithmic Qubits (AQ) of 35. AQ is IonQ’s metric measuring usable algorithmic capacity, more conservative than raw qubit count.
• Forte Enterprise — productized variant of Forte for commercial customers.

Announced systems:

• Tempo — next-generation system in development, targeting 64+ algorithmic qubits.

Algorithmic Qubits trajectory: AQ counts have grown from 11 (2020) through 22 (2022) to 35 (2024). The metric attempts to capture practical capability across a benchmark suite of algorithms.

Two-qubit gate fidelities: approximately 99.5-99.8% on production systems, somewhat below Quantinuum’s published numbers but competitive with the broader field.

Access: Available through AWS Braket, Microsoft Azure Quantum, and Google Cloud Marketplace. Programming through IonQ’s native SDK plus integration with Qiskit, Cirq, and PennyLane.

Atom Computing

Atom Computing operates neutral atom systems and was the first to break the 1000-qubit barrier in any quantum computing architecture.

Notable systems:

• 1,180-atom system (announced October 2023) — first quantum computer of any architecture to exceed 1,000 qubits. 2D array of neutral atoms with demonstrated coherence across the full array.
• Newer systems (2024-2025) with expanded arrays and improved fidelities.

Strategic positioning: Atom Computing has partnered with Microsoft on Azure Quantum integration and is one of the platforms targeted for Microsoft’s fault-tolerant quantum computing roadmap. The collaboration combines Atom Computing’s neutral atom hardware with Microsoft’s qubit-virtualization software stack.

Two-qubit gate fidelities: approximately 99.5%, with improvements ongoing through 2024-2025.

Access: Through Microsoft Azure Quantum.

QuEra Computing

QuEra operates neutral atom systems with a different architectural emphasis than Atom Computing — historically more focused on analog mode operation (continuous Hamiltonian evolution) and recently extending to gate-based digital operation.

Current production systems:

• Aquila (256 atoms) — production analog-mode system, available through AWS Braket. Used primarily for combinatorial optimization problems mappable to neutral atom Ising models.
• Newer systems demonstrating logical qubit operations and extended gate-based capabilities.

Notable results: QuEra has published demonstrations of logical qubit operations on neutral atoms in 2023-2024, including high-fidelity logical operations using reconfigurable atom arrays.

Access: Through AWS Braket primarily.

Pasqal

Pasqal operates neutral atom systems with strong integration into European supercomputing infrastructure.

Current systems:

• Multiple deployed systems at GENCI (French national supercomputing), Forschungszentrum Jülich (Germany), and other European supercomputing centers. The deployments are integrated with classical HPC infrastructure for hybrid quantum-classical workloads.
• Pasqal Orion Alpha and Beta — research systems with hundreds of atoms.

Strategic positioning: Pasqal is the leading European quantum hardware company and is central to the European Union’s quantum computing strategy. The systems are accessible primarily through European HPC centers rather than through commercial cloud platforms.

PsiQuantum

PsiQuantum is pursuing a photonic fault-tolerant million-qubit architecture from the outset, with no commercial intermediate-scale system deployment.

Current status: Pre-production. PsiQuantum has not deployed a publicly accessible quantum computer. The strategic bet is that the architecture they’re building is the right one for scale, and that intermediate demonstrations are not the goal.

Notable developments:

• Site selection in Brisbane, Australia (announced April 2024) — partnership with Australian Commonwealth and Queensland state governments, committing approximately AUD $1 billion to building a million-qubit photonic quantum computer in Brisbane by the end of the decade.
• Site selection in Chicago (announced 2024) — additional facility commitment.

Technology: Photonic qubits using silicon photonics fabrication, with single-photon detectors and the fusion-based quantum computing architecture. Manufacturing partnership with GlobalFoundries leverages standard semiconductor fabrication processes.

Access: None currently. PsiQuantum’s strategic position is fundamentally different from other vendors in this section — the company is building toward fault tolerance directly without intermediate commercial systems.

Xanadu

Xanadu operates photonic quantum computing systems with continuous-variable encoding, distinguished from PsiQuantum’s discrete-variable approach.

Current production systems:

• Borealis (216 squeezed-light modes, 2022) — the system that demonstrated quantum advantage on Gaussian boson sampling in 2022. Available through Xanadu Cloud.
• Aurora (next generation, 2024) — extended system with networked photonic processors.

Strategic positioning: Xanadu has focused on continuous-variable photonic computing and Gaussian boson sampling as both a research direction and a near-term commercial application. The approach is technologically distinct from PsiQuantum’s discrete-photon approach.

Access: Xanadu Cloud direct access. Programming through PennyLane, Xanadu’s open-source quantum machine learning framework.

Rigetti

Rigetti Computing operates smaller-scale superconducting systems with a commercial cloud focus.

Current production systems:

• Ankaa-2 (84 qubits) — production system with multi-chip architecture.
• Ankaa-3 (84 qubits) — improved multi-chip system with better connectivity.

Strategic positioning: Rigetti is publicly traded but operates at smaller scale than IBM, Google, or the leading trapped-ion vendors. The strategic bet is on continued algorithmic improvements making modest-qubit-count systems commercially useful.

Access: Rigetti Quantum Cloud Services, plus AWS Braket. Programming through Forest SDK / pyQuil, plus Qiskit compatibility.

Microsoft

Microsoft’s quantum hardware program has been historically focused on topological qubits, with a fundamentally different approach than every other vendor in this section.

Current status: Contested. The Majorana 1 processor announcement (February 2025) claimed the first topological qubit prototype based on a new topoconductor device structure. The claims have been received with significant skepticism by parts of the physics community, with several researchers publishing technical critiques arguing that the announced measurements do not unambiguously establish the presence of Majorana modes. As of mid-2026, the status remains contested — Microsoft maintains the claims, several independent researchers dispute them, and the broader community is waiting for additional results to settle the question.

Strategic positioning: Microsoft simultaneously operates the Azure Quantum cloud platform, which provides access to third-party hardware (Quantinuum, IonQ, Atom Computing, Pasqal, Rigetti, and others). The Azure Quantum platform is the largest multi-vendor quantum cloud and is a significant business regardless of the outcome of the topological qubit program.

Access (Microsoft hardware): None publicly available.
Access (Azure Quantum third-party): Multiple hardware vendors through a unified API and the Q# programming language plus Qiskit/Cirq compatibility.

AWS Quantum

Amazon Web Services entered direct quantum hardware development with the Ocelot announcement in February 2025, after years of operating only as a multi-vendor cloud platform through AWS Braket.

Ocelot: Cat qubit architecture using superconducting cavities, with hybrid bosonic-plus-discrete-qubit error correction. The strategic argument is that cat qubits provide intrinsic protection against bit-flip errors at the hardware layer, reducing the error correction overhead needed at the code layer.

Current status: Research phase. Ocelot is not yet deployed for external use; AWS has published technical details but has not provided access to the hardware.

AWS Braket: AWS’s multi-vendor quantum cloud platform, providing access to IonQ, Rigetti, QuEra, Pasqal, Oxford Quantum Circuits, and IQM systems through a unified API. Programming through the Amazon Braket SDK, plus integration with Qiskit, Cirq, and other major frameworks.

Chinese quantum programs

Chinese quantum computing efforts operate under different visibility constraints than Western programs, with public disclosures concentrated around major academic publications and state-affiliated research institutions. The major efforts:

USTC (University of Science and Technology of China) operates both superconducting and photonic quantum programs:

• Zuchongzhi superconducting line — Zuchongzhi 1.0 (62 qubits, 2021), Zuchongzhi 2.0 (66 qubits, 2021), Zuchongzhi 3.0 (105 qubits, 2024). The Zuchongzhi 3.0 system demonstrated quantum advantage results comparable to Google’s, with published 2024 results extending the random circuit sampling demonstrations.
• Jiuzhang photonic line — Jiuzhang 1.0 (76 photons, 2020), Jiuzhang 2.0 (113 photons, 2021), Jiuzhang 3.0 (255 photons, 2023). The Jiuzhang series has demonstrated Gaussian boson sampling at scales that establish photonic quantum advantage.

Origin Quantum is the commercial Chinese quantum hardware company, operating superconducting systems available through the Origin Quantum Cloud platform. International visibility is limited but the company has been the most commercially active of the Chinese quantum hardware efforts.

Other Chinese efforts include programs at the Chinese Academy of Sciences, Tsinghua University, Tencent Quantum Lab, and Baidu’s quantum computing efforts. The total scale of Chinese quantum investment is substantial — comparable to US efforts — but with less public disclosure of operational metrics.

The cloud platforms

For practitioners wanting actual hands-on access to quantum hardware in 2026, the cloud platforms are the relevant interface:

IBM Quantum is the largest by both qubit count accessible and by user base. Free open-access tier provides limited compute time on smaller systems; paid tiers (Premium, Enterprise) provide access to the latest Heron-generation hardware. Programming through Qiskit.

Microsoft Azure Quantum is the largest multi-vendor platform, offering access to Quantinuum, IonQ, Atom Computing, Pasqal, Rigetti, and others through a unified API. Programming through Q#, plus Qiskit and Cirq compatibility.

AWS Braket is Amazon’s multi-vendor platform with IonQ, Rigetti, QuEra, Pasqal, Oxford Quantum Circuits, and IQM. Programming through the Amazon Braket SDK plus framework integrations.

Google Quantum AI provides limited external access to Google’s hardware, primarily through research partnerships. Programming through Cirq.

Xanadu Cloud provides access to Xanadu’s photonic systems with continuous-variable programming through PennyLane.

Quantinuum provides direct cloud access to H-series systems with their TKET compiler.

For most practitioners getting started, IBM Quantum (free tier, broadest qubit-count access) or Azure Quantum (broadest hardware diversity) are the natural entry points. For specific architectural experiments, the direct vendor platforms make sense.

The capability frontier in mid-2026

Bringing the vendor-by-vendor picture together into a coherent frontier assessment:

Largest physical qubit count in a production-accessible system: Atom Computing’s neutral atom system exceeding 1,000 qubits. IBM Quantum System Two architectures connecting multiple Heron processors approach this scale through modular design.

Highest two-qubit gate fidelity in production: Quantinuum H2 at 99.9%+ on the best operations.

Most advanced error correction demonstration: Google Willow’s distance-7 surface code below-threshold result (December 2024).

Most advanced logical qubit operations: Quantinuum H2’s fault-tolerant operations with logical fidelity exceeding physical fidelity.

Most extensive quantum advantage demonstrations: USTC Jiuzhang and Zuchongzhi lines on contrived sampling problems.

Highest published Algorithmic Qubits: IonQ’s AQ 35 on Forte systems.

Most ambitious roadmap: PsiQuantum’s million-qubit photonic fault-tolerant target by end of decade, IBM’s Blue Jay 10,000 logical qubit target for 2033.

Most contested headline result: Microsoft’s Majorana 1 topological qubit claim from February 2025.

None of the current systems can break RSA-2048. None can perform useful quantum chemistry simulation beyond classically-tractable molecule sizes. None have demonstrated complete fault-tolerant algorithm execution. The capability frontier is roughly where the field had been hoping to be by the mid-2020s — substantial progress, no commercially decisive applications, error correction now experimentally validated at meaningful scale.

What needs to happen for cryptographic relevance

For Shor’s algorithm against RSA-2048, the requirements are:

• Logical qubits: approximately 4,000-10,000 (depending on the specific algorithm variant and resource estimation methodology).
• Physical qubits: approximately 5-20 million at current physical error rates of 0.5%, scaling down with better physical error rates.
• Per-gate error rates: physical error rates below 0.1% would substantially reduce the required code distance and total physical qubits.
• Magic state distillation: functional at scale, with the substantial overhead currently estimated.
• System integration: complete fault-tolerant stack from physical hardware through error correction decoders to algorithm execution.

The gap between current capability (~1,500 physical qubits, distance-7 error correction in a memory experiment) and what RSA-breaking requires is roughly four orders of magnitude in qubit count and several layers of system integration that have not been demonstrated. The honest timeline estimates from credible practitioners range from a decade to several decades, with no scientific consensus.

The post-quantum cryptography migration is being driven by these long timelines combined with the “harvest now, decrypt later” threat model and the operational reality that cryptographic systems take years to migrate. The Post-Quantum Cryptography page covers the migration in detail.

What needs to happen for chemistry relevance

For useful quantum chemistry simulation (specifically, simulating systems beyond what classical methods can handle accurately):

• Logical qubits: approximately 1,000-5,000 depending on the molecule and the simulation method.
• Physical qubits: approximately 2-10 million at current physical error rates.
• Hamiltonian simulation primitives: efficient implementations of Trotter decomposition or alternative simulation methods.
• Application-specific algorithms: VQE-style variational approaches at scale or fully fault-tolerant quantum phase estimation.

The chemistry frontier is structurally closer to current capability than the cryptography frontier — fewer logical qubits required, modest fault-tolerant operation depth, well-defined problem instances of clear value. Several research groups are projecting useful quantum chemistry advantage in the late 2020s, conditional on continued hardware progress and the development of practical algorithms that work within the available logical qubit budgets.

National programs

The major national quantum computing programs that shape the broader landscape:

United States — National Quantum Initiative Act (2018, reauthorized 2024). Approximately $1.2 billion in initial funding plus subsequent appropriations. The program covers basic research, hardware development, workforce development, and national laboratory infrastructure. Major participants: NIST, DOE national labs, NSF research programs, academic institutions.

European Union — Quantum Flagship (2018-2028). €1 billion over ten years. Quantum Technologies Flagship covers four pillars: quantum computing, quantum communication, quantum simulation, quantum sensing. Major participants: research institutions across all EU member states, with significant programs in Germany, France, Netherlands, and Austria.

China — National Laboratory for Quantum Information Sciences (Hefei). Reported budget exceeds $10 billion across multiple programs, though precise figures are difficult to verify. Major participants: USTC, Chinese Academy of Sciences, state-affiliated research institutions, major technology companies.

United Kingdom — National Quantum Strategy (2023). £2.5 billion over ten years. Major participants: UK Research and Innovation, NQCC (National Quantum Computing Centre), academic institutions, growing UK quantum industry.

Australia — National Quantum Strategy (2023) plus major commitments to PsiQuantum’s Brisbane facility (AUD $940 million federal plus AUD $470 million Queensland state). Major participants: CSIRO, Sydney Quantum Academy, the Australian universities, and the international companies establishing Australian operations.

Other significant programs — Japan, South Korea, Singapore, Canada, India, and Israel all operate national quantum computing programs at varying scales. The total global investment in quantum computing exceeds $25 billion across public and private sources, with no slowdown in commitment despite the uncertain timeline to practical impact.

Honest assessment in mid-2026

A few framing observations to close.

Hardware progress has been genuinely substantial in 2023-2026. Qubit counts have grown by roughly 10x. Per-qubit fidelities have improved meaningfully across all major platforms. Error correction has moved from theoretical to experimentally demonstrated. The trajectory is positive.

Quantum advantage on useful problems has not been demonstrated. Every “quantum advantage” demonstration to date has been on a contrived problem with no practical application. The leap from “quantum advantage on sampling tasks” to “quantum advantage on problems someone wants the answer to” has not been made, and is genuinely difficult.

The post-NISQ era is beginning. The 2023-2026 period has been the transition from NISQ-era variational algorithms (which produced limited progress over five years) to fault-tolerant-era preparations. The leading vendors are now building toward fault-tolerant operation rather than further refining NISQ approaches.

The cryptographic threat remains structural, not immediate. No system in 2026 is close to breaking RSA-2048. The post-quantum cryptography migration is being driven by future-looking risk management and harvest-now-decrypt-later concerns, not by present capability. The migration will likely complete before the quantum capability that would break the migrated-away-from algorithms is built.

Productive skepticism remains appropriate. Specific predictions in this field have repeatedly failed (Microsoft’s topological qubit timeline most prominently), and “we will achieve X by Y” claims should be discounted accordingly. The general trajectory is real; specific timelines are not reliable.

The field is well-funded and not going away. Whatever the timeline to practical impact turns out to be, the investment levels and national-program commitments mean quantum computing development will continue for the foreseeable future. The relevant question for practitioners is not whether to track the field, but how to allocate attention given the genuinely uncertain timeline.

Where to read next

Adjacent material on this site:

• Quantum Computing — the umbrella overview.
• Qubit Architectures — the architectures behind the systems covered here, with deeper treatment of the physics and engineering tradeoffs.
• Quantum Algorithms — what these systems will eventually be able to compute.
• Quantum Error Correction — the layer between the physical systems described here and the logical qubits algorithms require.
• Quantum Mechanics Fundamentals — the physics underlying all of the above.
• Post-Quantum Cryptography — the cryptographic response to the threat these systems will eventually pose.

The intent of this page is to be a living reference that updates as the field advances. Specific numbers will change, new vendors will emerge, current vendors will release new systems, and the capability frontier will move. The framework — what the metrics mean, how the architectures compare, what the cloud platforms provide, where the application thresholds sit — should remain useful even as the specific snapshot evolves.