Rigetti Computing’s 108-qubit Cepheus-1-108Q superconducting quantum processor is now generally available for cloud access, the company confirmed in April 2026. The milestone marks the first time a quantum chip of this scale has been exposed across both Rigetti’s own cloud service and third‑party gateways such as Amazon Braket. Organizations and independent developers can now queue jobs to the new hardware, pushing the boundaries of what’s possible in variational algorithms, optimization workloads, and materials simulation without owning a single dilution refrigerator.

The Cepheus‑1‑108Q sits at the center of a swiftly maturing quantum‑as‑a‑service ecosystem. For Windows‑oriented teams that manage hybrid infrastructure, the availability of larger, more coherent quantum backends signals a shift from experimentation toward pragmatic integration. This article unpacks the architecture, the cloud delivery model, early developer experiences, and the competitive landscape that Cepheus‑1‑108Q enters.

From Aspen to Cepheus: A modular leap

Rigetti’s journey to 108 qubits did not happen overnight. The company’s earlier Aspen‑M series topped out at 80 qubits and introduced multi‑chip scaling via a modular wafer‑scale approach. Cepheus‑1‑108Q pushes that philosophy further by linking multiple smaller processor tiles into a single addressable unit. Each tile carries a handful of tunable transmon qubits connected through cryogenic signal lines, with inter‑tile couplers that preserve coherence without the crosstalk that typically plagues monolithic designs.

The modular architecture is not just a marketing label. It lets Rigetti fabricate, test, and replace individual tiles rather than scrapping an entire chip when a few qubits underperform. In a field where yield is still measured in single percentages, this matters. The 108‑qubit aggregate is built from 3×3 tiles, each housing 12 qubits in a grid. Quantum engineers can address the full array as a single computational resource, but the underlying isolation improves gate fidelities and simplifies calibration. Early benchmarking data suggests single‑qubit gate fidelities above 99.5 % and two‑qubit gate fidelities approaching 98 %, though official numbers remain under wraps pending peer review.

For developers accustomed to IBM’s heavy‑hex lattice or Google’s Sycamore grid, the connectivity map of Cepheus‑1‑108Q may feel different. The intra‑tile nearest‑neighbor coupling follows a square lattice, while inter‑tile links provide longer‑range connections at the expense of slightly lower fidelity. Compilers and transpilers will need to account for this hierarchy, but the net effect is a clean path to practical qubit counts beyond the NISQ threshold.

Cloud delivery: Rigetti QCS, Amazon Braket, and beyond

General availability means that any registered user can book time on the processor. Rigetti’s own Quantum Cloud Services (QCS) platform, launched in 2018, remains the primary conduit. Here, Quil and pyQuil provide a native programming model that exposes pulse‑level control, parametric gates, and active qubit reset—capabilities that become essential when error mitigation is top of mind.

Simultaneously, the chip appears on Amazon Braket, a move that places Rigetti beside IonQ, D‑Wave, and Xanadu in one of the largest quantum marketplaces. Braket users gain access through a unified API that supports Qiskit, Cirq, and the OpenQASM standard, removing the friction of learning a new assembly language just to experiment with a new backend. For Windows shops already running in AWS, the integration means that a quantum job can be orchestrated from the same environment that hosts classical HPC, containerized applications, and machine‑learning models.

Other cloud partnerships have not been explicitly announced, but Rigetti’s leadership hinted at “additional platform integrations before mid‑2026” during a closed‑door analyst briefing. If history is any guide, Microsoft Azure Quantum would be a natural fit. Azure already showcases Rigetti’s lower‑qubit processors, and the Cepheus‑1‑108Q aligns with Microsoft’s push toward resilient, high‑qubit‑count backends. Government customers, particularly those in the national labs who run hybrid classical‑quantum pipelines under Linux and Windows, are also expected to receive dedicated bare‑metal access through Rigetti’s “Novera” on‑premise appliances, though the company has not confirmed a timeline.

What 108 qubits mean for real workloads

A 108‑qubit device is not a cryptographically relevant quantum computer. It remains squarely in the noisy‑intermediate‑scale quantum (NISQ) regime, where gate and readout errors accumulate rapidly. Error correction is not yet active on Cepheus‑1‑108Q, though the chip’s qubit layout was designed with surface‑code patches in mind. The true value today lies in variational quantum algorithms (VQAs), which offload part of the computation to a classical optimizer and tolerate some noise.

Quantum chemistry is an early beneficiary. With 108 qubits, researchers can simulate lithium‑sulphur battery cells, small proteins, or catalytic surfaces that saturate a 50‑qubit enumeration. Variational quantum eigensolvers (VQEs) and unitary coupled‑cluster methods scale roughly linearly in circuit depth with the number of qubits, and the added width of Cepheus‑1‑108Q directly expands the accessible basis set. Preliminary results from a pharmaceutical partner, disclosed during the Braket launch event, showed a 30‑fold reduction in classical pre‑processing time for a ligand‑docking simulation compared to the same algorithm on a perfect emulator of a 80‑qubit chip. Such speed‑ups are typical of algorithmic phase transitions that kick in when more qubits become available.

In finance, Monte Carlo option pricing, portfolio optimization, and fraud detection continue to be intense areas of exploration. Rigetti and AWS jointly released an open‑source toolchain that translates high‑level financial models into hybrid circuits for Braket backends. The toolchain auto‑selects Cepheus‑1‑108Q when the problem size exceeds 60 variational parameters—a threshold that previously forced users to larger‑scale emulation or to simplifications of the underlying model.

Machine learning workloads also benefit. Quantum kernel methods, where data is mapped into a high‑dimensional Hilbert space, rely on the expressibility of the quantum circuit. More qubits allow higher‑order feature maps, and the 108‑qubit lattice can encode data that would require billions of classical features. Early adopters are training quantum generative adversarial networks (QGANs) to produce synthetic financial time series, though the quality of the generated data still lags behind classical methods on all but the smallest datasets.

Developer experience: Quil, Qiskit, and cross‑compilation

One of the quiet improvements in the Cepheus‑1‑108Q rollout is the maturity of the software stack. Rigetti’s forest‑sdk 4.0, released alongside the hardware, introduces a feature called “Qubit Placer” that optimizes qubit assignment based on real‑time calibration data. Developers write high‑level circuits, and the placer automatically routes two‑qubit gates around lossy or poorly performing links. The placer also handles inter‑tile boundaries gracefully, swapping logical qubits to avoid costly inter‑tile two‑qubit gates when local equivalents are available.

For those more comfortable in the IBM ecosystem, the Braket interface translates Qiskit circuits into Quil using a layered transpiler. A quick test by the Quantum Open Source Foundation showed that a 100‑qubit Greenberger–Horne–Zeilinger state compiled on Qiskit ran on Cepheus‑1‑108Q with only three added SWAP gates per logical layer, preserving state fidelity above 85 % after readout error mitigation. That is a meaningful improvement over earlier cross‑compilation attempts on 80‑qubit hardware.

The Rigetti QCS platform goes deeper, offering parametric circuit execution, where gate angles are streamed mid‑sequence, and active qubit reset that resets qubits without thermal cycling. These are niche features for algorithm developers chasing the last decibel of performance, but they signal that the hardware is no longer a fragile physics experiment. It is a machine designed for repeated production use.

The competitive landscape: 100‑qubit class becomes the norm

Rigetti is not alone in the march toward 100‑plus qubits. IBM’s Condor processor hit 1,121 qubits in late 2023, though it used a monolithic design that has since been superseded by the smaller, modular Heron chips. Google’s Willow, also modular, reached 105 qubits earlier in 2025, while Chinese player Origin Quantum demonstrated a 128‑qubit chip in late 2025. The 100‑qubit threshold, once science fiction, is now table stakes.

What sets Cepheus‑1‑108Q apart is its dual‑channel cloud strategy. By being simultaneously available on Rigetti QCS and Amazon Braket, it lowers the barrier to entry more aggressively than competitors that rely on a single platform. IBM’s hardware remains exclusive to IBM Quantum Platform, and Google’s Sandbox Division limits access to select partners. IonQ’s trapped‑ion systems are also on Braket, but they scale differently, with algorithm‑relevant gate speeds that remain an order of magnitude slower than superconducting qubits. That makes them less suited for workloads that require numerous variational iterations.

Rigetti’s pricing model mirrors the industry trend toward per‑task billing. On QCS, a standard compute hour on Cepheus‑1‑108Q costs approximately $2,000, with volume discounts bringing it below $1,500. Amazon Braket adds a small overhead but offers flexible billing with no subscription commitment. For comparison, IBM’s 127‑qubit Eagle processor costs around $1,600 per compute hour while the more advanced Heron chips hover near $2,400. Price parity with the competition, combined with easier access, could tilt procurement decisions toward Rigetti, particularly for teams that already operate inside AWS.

Workload scheduling, queue times, and hybrid execution

Cloud availability does not guarantee instant access. The Cepheus‑1‑108Q is a single physical machine sitting in a cryostat at Rigetti’s Fab‑1 facility in Fremont, California. The queue on Braket can stretch to several hours during peak US and European working times. Rigetti’s reservation system, available on QCS, allows advanced scheduling, but slots fill quickly. Windows‑based high‑performance computing engineers who orchestrate overnight simulation runs should plan accordingly and fall back on the built‑in circuit emulators for development.

Rigetti’s software now includes a “Hybrid Orchestrator” that interleaves quantum jobs with classical processing stages managed by Azure or AWS Batch. The Orchestrator holds a persistent session on Cepheus‑1‑108Q to avoid re‑initialization delays between iterations. This feature is a quiet productivity enhancer; it reduces total wall‑clock time by up to 40 % for VQA loops that require hundreds of iterations. Sessions are stateful, meaning that calibration data is cached and not re‑acquired unless a sudden drift in coherence is detected.

Quantum forum reactions: praise, caution, and wishlists

Community feedback on the Cepheus‑1‑108Q announcement has been cautiously optimistic. On WindowsForum, threads quickly filled with benchmarks, compiler tips, and the occasional gripe. One poster noted that the inter‑tile two‑qubit gate fidelity is still a bottleneck for deep circuits, showing a drop from 98 % to 91 % when crossing tiles. Another shared a workaround that uses SWAP networks to keep entangling gates within tiles, improving the success rate of a 20‑layer variational circuit by 15 %. These real‑world exchanges highlight the gap between spec‑sheet promise and actual application, but they also show that the community is rapidly building the institutional knowledge required to wring performance out of the hardware.

A recurring wish in the discussion was for tighter integration with Windows‑based classical computing environments. “I can run my entire optimization loop in Azure Machine Learning, but the quantum step feels like a detour into Linux land,” wrote one enterprise architect. Rigetti’s Forest SDK currently requires POSIX‑compatible systems, though it can be accessed through Windows Subsystem for Linux. Native Windows binaries are not on the roadmap. Amazon Braket’s SDK is Python‑based and runs happily on Windows natively, which partially addresses the concern, but pulse‑level control still requires Quil and pyQuil, which are only supported under WSL2. Microsoft users who desire the lowest‑level access will need to provision a Linux jump box.

Security and compliance in a quantum‑cloud world

As quantum hardware reaches this scale, security postures must adapt. Rigetti’s QCS encrypts all jobs at rest and in transit using AES‑256 and TLS 1.3, and offers dedicated isolated cryostats for government contractors. On Braket, customers benefit from AWS’s SOC 2 and FedRAMP certifications, although the qubit‑level channel is not yet built into AWS Nitro Enclaves. Data residency is still limited to US‑West‑2, a concern for European financial institutions governed by GDPR. Rigetti has stated that a European instance, likely hosted in a Dublin AWS region, is under evaluation for late 2026.

From a cryptographic standpoint, the 108‑qubit Cepheus‑1‑108Q cannot break any widely used public‑key scheme. Shor’s algorithm would require millions of error‑corrected qubits. However, security architects are urged to begin inventorying their long‑lived secrets because the mere existence of high‑qubit processors accelerates the timeline to quantum‑safe migration, even if the breaking point remains years away.

Looking ahead: the path to fault tolerance

The Cepheus‑1‑108Q is a stepping stone. Rigetti’s public roadmap outlines a 500‑qubit modular processor, tentatively called “Observatory,” slated for early 2027. That chip will introduce real‑time error correction on a subset of qubits, demonstrating a logical qubit with coherence exceeding that of any physical qubit. If successful, it would be one of the first cloud‑accessible devices with an embedded error‑correction cycle, a psychological milestone for the industry.

In the meantime, the 108‑qubit processor serves as a high‑visibility sandbox. Algorithm researchers can stress‑test error mitigation techniques at a scale that was previously only possible on expensive classical emulators. Hardware engineers can gather statistics on failure modes across thousands of user circuits. And the broader Windows‑centric developer ecosystem can begin to treat quantum computing as just another API endpoint in a modern cloud‑native architecture—a reality that, until this April, felt several years away.

The modular approach that Rigetti has championed seems to be paying dividends. By making the Cepheus‑1‑108Q widely available and embracing multi‑platform distribution, the company is betting that ubiquity will breed innovation. Whether that bet pays off depends on how quickly the quantum‑computing community can turn a 108‑qubit, NISQ‑era machine into something that delivers tangible value beyond the lab. The hardware is here. The software is maturing. Now it’s the users’ turn.