Rigetti's Multi-Chip Bet and the 99.1% Fidelity Question

Rigetti Computing now operates the industry's largest modular quantum computer — a 108-qubit system called Cepheus-1-108Q, twelve interconnected nine-qubit superconducting chiplets tiled together into a single quantum machine and made generally available through Rigetti's cloud platform and Amazon Braket. It is also currently running at 99.1% median two-qubit gate fidelity. That number is the entire investment thesis on the stock.

A four-tenths-of-a-percentage-point gap separates Cepheus-1-108Q from the 99.5% threshold Rigetti itself has set as the next milestone, and from the fidelity its earlier 36-qubit multi-chip system already cleared. Closing that gap at twelve chiplets — and then proving the same fidelity at the 200-plus chiplet scale Rigetti's roadmap projects for the next two years — is what determines whether the chiplet architecture is the dominant path to fault-tolerant quantum computing or an interesting research result that gets out-scaled by IBM's superconducting roadmap and Google's error-corrected logical-qubit work.

The fidelity number is doing more work than most coverage of Rigetti acknowledges.

Why 99.5% is the line that matters

Two-qubit gate fidelity is the bottleneck in superconducting quantum computing because every meaningful quantum algorithm requires deep circuits — chains of hundreds or thousands of consecutive two-qubit operations. Error compounds multiplicatively at depth. A circuit of 100 two-qubit gates at 99.0% per-gate fidelity yields a roughly 37% probability of a clean run end-to-end. The same circuit at 99.5% yields about 61%. At 99.9% — the threshold companies like Google and Quantinuum are pushing toward — that same circuit lands around 90%.

Surface-code error correction, the leading path to fault-tolerant quantum computing, requires per-gate fidelity comfortably above roughly 99% just to break even and well above 99.5% before the logical-qubit overhead becomes anything other than catastrophic. Hitting 99.5% on a usefully large processor is not a marketing milestone. It is the line where the math starts to favor scaling up rather than burning compute on error correction.

For a chiplet architecture specifically, the relevant fidelity is the median across both within-chiplet and cross-chiplet two-qubit gates. Cross-chiplet gates are physically harder because they cross a deliberate package boundary; within-chiplet gates use shorter, better-characterized connections. Quoting a chip-wide median that aggregates both is correct, but it does aggregate.

The fidelity progression tells a story

The arc is best read in sequence.

The 84-qubit single-chip Ankaa-3 hit 99.5% median fSim two-qubit fidelity and 99.0% median iSWAP fidelity as a monolithic device. That was the original proof point for Rigetti's qubit design and its in-house ABAA Josephson-junction process at Fab-1.

The first multi-chip system, Cepheus-1-36Q, then hit 99.5% median two-qubit gate fidelity across four nine-qubit chiplets tiled together — a 2× reduction in median two-qubit gate error rate versus the previous 84-qubit single-chip result. The press release does not break out cross-chiplet versus within-chiplet gate fidelity, but the chip-wide median crossing the 99.5% threshold is the data point investors keyed on.

Tripling chiplet count to twelve produced Cepheus-1-108Q, currently at 99.1% median two-qubit gate fidelity. Rigetti's stated plan is to push that figure back to 99.5% later this year through continued chiplet-level tuning and ABAA process improvements, with a roadmap update to follow once results are in. Management has framed the same path as one that reaches quantum advantage in approximately three years.

The pattern that matters is the slip between 36-qubit and 108-qubit systems. Median fidelity dropped by 40 basis points as chiplet count tripled. There are several plausible reasons — yield variance across more chiplets, additional cross-chiplet coupling burden, calibration overhead — and each of them is a real engineering problem rather than a measurement quirk. None is fatal, but none is solved.

Why the latest quarter made the math more honest

Rigetti's first-quarter 2026 financial results put quarterly revenue at $4.4 million against $1.5 million in the same quarter a year prior, a roughly 193% year-on-year increase driven largely by on-premises Novera QPU deliveries and government and research contracts. GAAP net income of $33.1 million for the quarter was boosted by non-cash fair-value adjustments rather than operating performance; non-GAAP net loss was $14.7 million.

Two structural numbers matter more than the GAAP headline.

First, the cash position. As of the end of the most recent quarter, Rigetti held $569 million in cash, cash equivalents, and available-for-sale investments, with no debt. Against a roughly $26 million quarterly operating loss, that implies a runway of more than five years at the current burn — a position that distinguishes Rigetti from quantum-computing peers facing imminent dilution.

Second, the revenue mix. Rigetti's commercial revenue is concentrated in Quantum Foundry Services from Fab-1 and on-premises QPU sales rather than cloud-access fees. The cloud-revenue line is real but is not the company's primary monetization vector today. That detail matters because it changes how Rigetti should be valued against pure cloud-quantum competitors.

The DARPA signal nobody wants to discuss

DARPA's Quantum Benchmarking Initiative is the closest thing the US government has produced to a public credibility scoreboard for utility-scale quantum computing by 2033. Rigetti was selected for Stage A. The agency's Stage B announcement of eleven advancing teams did not include Rigetti, alongside other Stage A participants such as Alice & Bob, Atlantic Quantum, Google, HPE, and Oxford Ionics.

DARPA explicitly noted that additional teams may advance from earlier stages as staggered timelines complete their evaluation cycles, so the non-advancement is not a verdict. It is a data point. Markets that mostly ignore DARPA's procurement-track signals here are probably underweighting one of the few independent technical evaluations of competing quantum architectures in the public record.

If Rigetti hits 99.5% on the 108-qubit system this year and demonstrates clean scaling beyond it, the DARPA evaluation is more likely to come around. If 99.1% is the ceiling at twelve chiplets, the agency's signal becomes harder to read as anything other than a discount.

Fab-1 is a strategic option the market keeps mispricing

Rigetti operates Fab-1 in Fremont, California — the only US-owned superconducting quantum fabrication facility in commercial operation. The site is the only place outside of IBM's internal fab where superconducting QPUs are designed, built, and tested under one roof. Quantum Foundry Services on Fab-1 capacity is a small but real revenue line today, and the strategic logic of that asset is independent of whether Rigetti's own computer business wins.

Two scenarios reward Fab-1 ownership specifically. The first is a Department of Defense or Department of Energy sole-source procurement that wants US-onshore superconducting QPU capacity for sovereign-supply-chain reasons. The second is a foundry-service deal with a larger quantum competitor that does not want to operate its own fab — IonQ's pending SkyWater deal answers that question for trapped-ion systems but does not change the superconducting-fab gap.

Even in a bear case where Rigetti's flagship computer business does not dominate, the fab itself carries optionality that does not show up in a single discounted-cash-flow line for the cloud-quantum business. Sum-of-parts is the analytical frame the market still mostly skips.

Where the competitive ground is shifting

Three pressures bound Rigetti's window.

IBM's roadmap, with its Heron and System Two architectures, is reportedly approaching superconducting scales above 1,000 qubits and is now publishing logical-qubit projections rather than physical-qubit milestones. Google's Willow error-correction work has demonstrated below-threshold scaling on a single chip. IonQ's networking-stack pivot is reframing the competitive set away from raw qubit counts entirely. Rigetti's chiplet bet works in a world where none of those three pulls dominate before the modular approach produces a fault-tolerant logical qubit. The fidelity number gates that race, and the broader public quantum-computing investment cohort is repositioning around it.

Scenarios to watch

The next twelve to eighteen months will resolve most of the question this stock is currently asking.

In a constructive case, Cepheus-1-108Q hits the 99.5% median target this year, Rigetti delivers a roadmap update putting a 200-plus chiplet system on a credible timeline, DARPA reinstates the company into Stage B at the next evaluation window, and Fab-1 picks up at least one major foundry customer outside the Rigetti house. In that world, the chiplet thesis is the modular template.

In a base case, fidelity improves to 99.3% by year-end on the 108-qubit system, Rigetti retains its cash position, and the company sells additional Novera units into government and research while IBM and Google's monolithic and error-corrected paths absorb most of the commercial cloud-quantum demand. The investment becomes a strategic-optionality play around Fab-1 with the computer business as supporting revenue.

In a bear case, fidelity holds near 99.1%, chiplet count beyond 200 produces error compounding that the modular architecture cannot offset, and the path to fault-tolerant quantum computing routes through Google's surface-code work and IBM's larger monolithic chips. Rigetti's multiple compresses toward the fab's standalone value plus a residual research-tools business.

The cleanest analytical question is whether the 40-basis-point fidelity slip between 36-qubit and 108-qubit systems is a transient yield problem or a structural ceiling on chiplet scaling. Rigetti has the cash to keep iterating. The market is pricing roughly even odds.