🔗 Career 06 · Google vs IBM Google · Willow, 105 qubits IBM · Nighthawk, 120 qubits

Google vs IBM , Superconducting Qubit Showdown

The two longest-running, most heavily scrutinized superconducting-qubit quantum computing programs, and two very different philosophies about what counts as progress: Google's headline "quantum advantage" benchmarks against classical supercomputers, versus IBM's "utility" claims and an explicit, dated roadmap to large-scale fault tolerance. A thorough, source-checked comparison across architecture, error correction, advantage claims, and business.

Google Quantum AI

Alphabet subsidiary · transmon superconducting · not separately public

Below-threshold QEC (Nature 2024) 105 qubits (Willow) Verifiable advantage claim (2025) Part of Alphabet (NASDAQ: GOOGL)

IBM Quantum

IBM Corp division · transmon superconducting · NYSE: IBM

"Utility" claim (Nature 2023) 120 qubits (Nighthawk) Starling FTQC target: 2029 qLDPC bivariate-bicycle codes

01 Why These Two Companies?

The two research programs that between them have made both of the biggest public claims in superconducting quantum computing: "quantum advantage" and "quantum utility."

Superconducting transmon qubits are tiny LC circuits made from aluminum films on silicon or sapphire, cooled to millikelvin temperatures in a dilution refrigerator, and manipulated with microwave pulses. It is the oldest continuously-funded qubit modality at industrial scale, and it is dominated by two very differently structured organizations: Google Quantum AI (a research division inside Alphabet/Google, not separately publicly traded, that treats each generation of hardware as a physics-demonstration vehicle) and IBM Quantum (a business unit of a NYSE-listed 114-year-old technology company, that has published detailed multi-year commercial roadmaps and committed more than $10B to quantum computing as of June 2026). Both are chasing the same underlying physics — the surface code, or a related error-correcting code, layered on top of a scalable transmon lattice — but from opposite institutional starting points.

The core trade-off: Google has made the two headline-grabbing physics claims in the field: the original (and long-disputed) 2019 "quantum supremacy" result on the 53-qubit Sycamore chip, and a 2025 "verifiable quantum advantage" claim using the Quantum Echoes algorithm on the newer 105-qubit Willow chip, peer-reviewed and published in Nature. IBM, by contrast, has focused on a narrower, more conservative "utility" claim — that a 127-qubit device (Eagle) could produce numerically useful results on certain physics simulations before full error correction — a claim that was itself subsequently challenged by classical tensor-network simulations. IBM has also published by far the most detailed public fault-tolerance roadmap in the industry, naming every intermediate chip through 2029. The tension this page is built around: Google currently holds the stronger, more recent, peer-reviewed advantage claim; IBM has the more transparent, more skeptically stress-tested roadmap and a hardware philosophy (bivariate-bicycle qLDPC codes) that promises roughly 10× less physical-qubit overhead than the surface code Google has used to date. Neither company has yet demonstrated a fault-tolerant, commercially useful quantum computer.

13,000×

Google Quantum Echoes speedup
vs. best classical algorithm (Nature 2025)

105

Willow qubits
(two chip variants, Dec 2024)

120

IBM Nighthawk qubits
(square lattice, Nov 2025)

2029

IBM's public target date
for Starling, its first FTQC system

$10B+

IBM's cumulative quantum
investment commitment (Jun 2026)

Source-confidence rule for this page: peer-reviewed numbers come directly from journal papers; preprint numbers are posted but not yet peer-reviewed; company roadmap numbers are product specs, press releases, or future targets. Both companies' most famous advantage/utility claims were later challenged by independent classical-simulation papers — those challenges are presented alongside the original claims throughout this page rather than omitted, because that back-and-forth is itself the most instructive part of the story.

02 Architecture & Qubit Control

Both use transmon superconducting qubits and microwave control — the difference is in lattice topology and how couplers are wired.

🔵 Google: Fixed-Frequency, Square/Diagonal Lattice

Google's transmons are arranged on a lattice with roughly grid-like nearest-neighbor connectivity (Willow's spec sheet reports an average connectivity of 3.47 — close to, but not exactly, a uniform 4-way square grid, since edge and corner qubits have fewer neighbors). Two-qubit gates use tunable couplers between fixed-frequency qubits, a design first introduced on Sycamore and refined through Willow.

Fixed-frequency transmons + tunable couplers: reduces frequency-crowding-related errors relative to fully tunable-qubit designs, at the cost of needing careful frequency allocation across the chip during fabrication established technique
Two chip variants at the Willow generation: Google fabricates and tunes separate chips optimized for different demonstrations — one tuned for quantum error correction (better two-qubit fidelity, shorter T1), one tuned for random-circuit-sampling/advantage demonstrations (longer T1, even better two-qubit fidelity) — rather than a single chip serving both roles equally well company spec sheet
Packaging: flip-chip bump-bonded architecture with through-silicon vias for wiring, allowing control and readout lines to be routed from beneath the qubit plane rather than crowding the chip edges — a packaging approach Google has used since Sycamore and refined for Willow's denser lattice
Google Quantum AI is a research division inside Alphabet, not a separately reported business unit — there is no public product line comparable to IBM's cloud-accessible System One/System Two hardware tiers

🟣 IBM: Heavy-Hex, Now Pivoting to Square

From Eagle (2021) through Heron R2 (2024), every mainline IBM chip used a heavy-hexagonal lattice — a topology chosen specifically because it is compatible with the surface code while minimizing frequency collisions between neighboring qubits, at the cost of below-4 average connectivity. With Nighthawk (Nov 2025), IBM switched to a denser square lattice with tunable couplers.

Heavy-hex generations (2021–2024): Eagle (127q), Osprey (433q), Condor (1,121q), Heron/Heron R2 (133q/156q) — connectivity capped at 3 for most qubits, chosen to suppress two-qubit crosstalk and frequency collisions company
Nighthawk (2025–): 120 qubits, square lattice, tunable couplers, each qubit connecting to up to 4 neighbors via 218 coupler pairs — IBM reports this is a 20% increase in inter-qubit connections over Heron, letting circuits use ~30% more two-qubit gate layers at comparable fidelity by cutting the number of SWAP gates needed to route non-adjacent qubits together
Two-qubit gate fidelity above 99.9% is reported for over half of tested qubit pairs on Nighthawk-class hardware, with circuit-layer-operations-per-second (CLOPS) reaching 330,000, a 65% improvement over IBM's 2024 hardware company spec
IBM explicitly frames the heavy-hex → square-lattice shift as a deliberate strategic pivot: the earlier 1,121-qubit Condor chip proved raw qubit count could scale, but IBM subsequently prioritized fidelity and connectivity per qubit (the Heron/Nighthawk lineage) over further count-scaling before returning to larger systems

Why this is the real story of this comparison: both companies converged on transmon qubits with tunable couplers, but they made opposite bets on lattice density at different points in time — Google has favored denser near-square connectivity since Sycamore, while IBM spent 2021–2024 deliberately using a lower-connectivity heavy-hex lattice to buy fidelity headroom, only pivoting to a denser square lattice once its two-qubit fidelities were reliably above 99.9%. Nighthawk is, in a real sense, IBM catching up to a topology choice Google made years earlier — but arriving there with a more mature qLDPC error-correction strategy already public (Section 05).

03 Hardware Generations, Side by Side

Very different scaling philosophies: Google iterates slowly with big physics claims; IBM iterates fast with named chips on a public cadence.

Google Quantum AI Chips

2017–2018 · Foxtail / Bristlecone

20-qubit and 72-qubit prototypes establishing the fixed-frequency, tunable-coupler transmon architecture Google has used ever since.

2019 · Sycamore

53 usable qubits (54 fabricated, 1 non-functional). Basis of the original "quantum supremacy" claim (Section 06) — later disputed by classical-simulation papers.

Feb 2024 · Below-threshold QEC

Published in Nature: first experimental demonstration that increasing surface-code distance (from 3 to 5 to 7) decreases logical error rate, the long-sought "below threshold" regime for quantum error correction.

Dec 2024 · Willow

105 qubits, average connectivity 3.47, fabricated as two tuned variants (QEC-optimized and RCS/advantage-optimized — see Section 04 for the differing specs). Random circuit sampling benchmark completed in under 5 minutes, a task Google estimated would take a leading classical supercomputer roughly 10²⁵ years.

Oct 2025 · Quantum Echoes

Peer-reviewed in Nature: a verifiable out-of-time-order-correlator (OTOC) algorithm run on Willow, reported 13,000× faster than the best known classical algorithm on a leading supercomputer, cross-validated against real nuclear magnetic resonance (NMR) measurements on two molecules with UC Berkeley.

IBM Quantum Chips

2021 · Eagle

127 qubits, first IBM chip to break the 100-qubit barrier and the first to use the heavy-hex lattice at scale.

2022 · Osprey

433 qubits, heavy-hex, primarily a packaging/scaling demonstration rather than a fidelity milestone.

Jun 2023 · "Utility" claim (Eagle)

127-qubit Eagle ran a 60-gate-layer kicked-Ising-model simulation with error mitigation, published on the cover of Nature — later challenged by classical tensor-network papers (Section 06).

Dec 2023 · Condor & Heron

Condor: 1,121 qubits, heavy-hex — IBM's largest qubit-count chip to date and, so far, its last "more qubits" flagship. Heron (133 qubits): first chip to drop the fixed-coupler design in favor of tunable couplers, prioritizing fidelity over count.

2024 · Heron R2

156 qubits, refined tunable-coupler heavy-hex design; became IBM's primary cloud-accessible production chip through 2025.

Nov 2025 · Nighthawk & Loon

Nighthawk: 120-qubit square-lattice production chip (Section 02). Loon: a separate, non-production chip introducing long-range "c-couplers" needed for qLDPC codes — first rung of IBM's fault-tolerance roadmap (Section 05).

Qubit count by named generation for each company (log scale). Note IBM's Condor (1,121 qubits, 2023) was a scale-only demonstration; IBM's subsequent chips (Heron, Nighthawk) deliberately used far fewer, higher-fidelity, better-connected qubits instead of continuing to scale count.

04 Gate Fidelity Deep Dive

Willow shipped as two different chips with two different fidelity profiles — a nuance worth getting right before comparing to IBM.

Google's own Willow spec sheet (published alongside the December 2024 announcement) reports different calibration numbers for "Chip 1," tuned for the below-threshold quantum error correction demonstration, and "Chip 2," tuned for random circuit sampling / advantage demonstrations. Conflating the two — quoting Chip 2's better two-qubit fidelity as if it were the QEC chip's number, or vice versa — is a common but avoidable mistake, so both are reported separately below.

$$\varepsilon_{2Q}^{\rm Willow\,Chip1\,(QEC)} = 0.33\%\pm0.18\% \;\Rightarrow\; \mathcal{F}_{2Q} \approx 99.67\%,\qquad T_1 = 68\,\mu s\pm13\,\mu s$$ $$\varepsilon_{2Q}^{\rm Willow\,Chip2\,(RCS)} = 0.14\%\pm0.052\% \;\Rightarrow\; \mathcal{F}_{2Q} \approx 99.86\%,\qquad T_1 = 98\,\mu s\pm32\,\mu s$$ Both figures are randomized-benchmarking averages across the full 105-qubit device. Google's spec sheet attributes the T1/fidelity split between the two chips to "a tradeoff between optimizing qubit geometry for electromagnetic shielding and maximizing coherence" — i.e. the two chips are deliberately different devices, not the same hardware measured twice.

Two-qubit gate fidelity by system. Dashed line marks ~99.9%, a commonly cited practical threshold for useful near-term circuits. IBM's ">99.9% for over half of tested pairs" is a distribution claim, not a single average — shown here as its reported representative figure.

Google, fidelity record

Sycamore (2019): ~99.4% average 2Q fidelity on the RCS benchmark circuits used for the original supremacy claim peer-reviewed, Nature 574 (2019)
Willow Chip 1 (QEC-tuned, 2024): 99.965% 1Q / 99.67% 2Q / 99.23% readout, averaged across the device company spec sheet
Willow Chip 2 (RCS-tuned, 2024): 99.964% 1Q / 99.86% 2Q / 99.33% readout company spec sheet

IBM, fidelity record

Eagle (2021–2023): fidelities in the low-to-mid 99% range for 2Q gates, sufficient for the error-mitigated (not error-corrected) 2023 utility experiment, but not separately publicized as a headline spec company
Heron / Heron R2 (2023–2024): tunable couplers pushed typical 2Q fidelity into the "Heron-class" baseline IBM continues to cite, roughly 99.9% on well-behaved pairs company spec
Nighthawk (2025): >99.9% 2Q fidelity reported for over 50% of tested qubit pairs on the new square lattice, alongside a 65% CLOPS improvement — IBM reports this as a distribution across pairs rather than a single device-wide average company spec

Reading the fidelity race honestly: Google's RCS-tuned Willow chip (99.86% 2Q, company spec) and IBM's Nighthawk (>99.9% for a majority of pairs, company spec) are in the same performance class, and neither company has published a directly comparable, identically-measured, full-device average that would let this page declare a clean winner. What differs more than raw fidelity is what each company chose to build on top of that fidelity: Google used its QEC-tuned chip to cross the below-threshold error correction milestone (Section 05), while IBM used its fidelity headroom to justify committing to a specific, named fault-tolerance roadmap (Section 05) years in advance.

05 Fault Tolerance & Error Correction

Google has demonstrated the below-threshold milestone; IBM has published the more detailed public roadmap and a fundamentally different, lower-overhead code family.

$$p_L(d) \approx A\left(\frac{p}{p_{\rm th}}\right)^{\lfloor (d+1)/2\rfloor}, \qquad \Lambda \equiv \frac{p_L(d)}{p_L(d+2)} \approx 2.14\pm0.02\ \text{(Google, Willow, 2024)}$$ Google's Feb 2024 result measured the error-suppression factor $\Lambda$ directly: each increase in surface-code distance by 2 (from $d=3$ to $d=5$ to $d=7$) reduced the logical error rate by a consistent factor of about 2.14 — the first time this exponential suppression had been shown to continue below the code's error threshold, rather than saturating or reversing due to correlated/leakage errors as in earlier smaller-distance experiments.

Google: Below-Threshold Surface Code

Feb 2024 (Nature): distance-3, -5, and -7 surface codes on Willow's QEC-tuned chip; logical error rate roughly halved with each larger distance, and the best logical qubit outperformed the best constituent physical qubit's lifetime for the first time peer-reviewed
Uses the standard rotated surface code — well-understood theoretically but with a relatively high physical-to-logical qubit overhead (order 100+ physical qubits per logical qubit at useful error rates), compared with qLDPC alternatives
No public Google roadmap names specific future chips or years the way IBM's does; Google's public messaging emphasizes physics milestones (below-threshold, verifiable advantage) over named hardware-generation dates

IBM: qLDPC Bivariate-Bicycle Codes

Bivariate-bicycle "gross code": $[[144,12,12]]$ — 144 physical data qubits plus 144 syndrome qubits (288 total) encode 12 logical qubits; IBM reports roughly 10× fewer physical qubits than a surface code would need for comparable logical error rates, because the code's long-range checks pack more logical information per physical qubit company research
Loon (2025): first chip with the long-range "c-couplers" needed to physically realize qLDPC check operators — a proof-of-concept for the wiring, not yet a full logical-qubit demonstration
Kookaburra (2026, target): first module combining qLDPC memory with an attached logical processing unit (LPU) roadmap
Cockatoo (2027, target): entanglement between separate qLDPC modules via a "universal adapter" — the modular-scaling step IBM's whole roadmap depends on roadmap
Starling (2028–2029, target): 2028 — magic-state injection across multiple modules; 2029 — full system target of 100 million quantum gates across 200 logical qubits, IBM's first claimed fault-tolerant machine roadmap

Logical-qubit figures that are actually quantified in each company's own publications or roadmap statements — not every named milestone has a published qubit count, so only the ones that do are plotted. Google's point is a textbook-formula estimate (a distance-7 rotated surface code encodes 1 logical qubit in 2d²−1 = 97 physical qubits) applied to its 2024 below-threshold demonstration, not a number stated in the paper itself.

Demonstrated result vs. named roadmap: Google's below-threshold result is a completed, peer-reviewed physics demonstration on real hardware today. IBM's qLDPC/bivariate-bicycle overhead advantage is real and independently credible as theory (the code's distance and rate properties are published and reviewed), but the full modular architecture needed to use it at scale — Loon through Starling — is still almost entirely a forward-looking roadmap as of this writing, with only the first rung (Loon's c-couplers) built. Neither company has yet run a real, useful algorithm on a fault-tolerant logical processor.

06 Advantage & Utility Claims — and Their Rebuttals

Both companies' most famous headline results were later contested by classical-computing researchers. That back-and-forth is the most important context for reading either claim.

Google's Claims

2019 "Quantum Supremacy" (Sycamore): a random-circuit-sampling task Google estimated would take the Summit supercomputer ~10,000 years, completed in 200 seconds. Published in Nature peer-reviewed, but the classical-runtime estimate was disputed almost immediately — IBM itself published a rebuttal estimating 2.5 days with better classical algorithm/storage use, and subsequent papers (including a 2022 result using a large GPU cluster) narrowed or reversed the gap further for that specific circuit size.
2025 "Verifiable Quantum Advantage" (Quantum Echoes, Willow): an out-of-time-order-correlator (OTOC) algorithm reported as 13,000× faster than the best known classical algorithm on a leading supercomputer, published in Nature peer-reviewed. Google emphasizes this is the first verifiable advantage claim — the same algorithm was cross-validated on a second, independent Willow-class device and against real NMR spectroscopy measurements on two molecules (15 and 28 atoms) with UC Berkeley, rather than only checked against a classical simulation that becomes intractable at scale.
As of this writing, no classical-simulation rebuttal of the October 2025 Quantum Echoes claim has been published — a meaningfully different situation from 2019, though the field's track record (below) suggests such attempts are likely and worth watching for.

IBM's Claim

2023 "Utility" (Eagle, kicked Ising model): a 60-gate-layer, 127-qubit circuit, run with zero-noise-extrapolation error mitigation, reported as producing results "more accurate than leading classical approximation methods," published on the cover of Nature peer-reviewed. IBM was explicit at the time that this was not a claim of outright classical-impossibility — only that the quantum device beat the classical approximation methods tested at that point.
Rebuttal: within months, multiple independent groups showed classical tensor-network methods could match or beat the quantum device's accuracy on the same circuit. The most cited, Tindall, Fishman et al. (arXiv:2306.14887, later published in PRX Quantum 5, 010308, 2024), used belief-propagation tensor networks exploiting the heavy-hex lattice's tree-like correlation structure to produce results the authors report as "significantly more accurate and precise" than IBM's noise-mitigated quantum output — and argued their method scales to simulate the same physics in the thermodynamic (infinite-qubit) limit classically.
IBM's own framing anticipated this: the paper stated the team "fully expect[ed] that the classical computing community will develop methods that verify the results" — which is exactly what happened. The net effect: the 2023 "utility" claim is best read as a genuine, peer-reviewed milestone in useful near-term computation, not as a surviving claim of classical intractability.

How to read both claims fairly: "peer-reviewed" is not the same as "unchallenged" — both of these headline results passed peer review and were subsequently contested by other peer-reviewed or arXiv-posted classical-simulation work, which is normal, healthy scientific practice rather than evidence of fraud on either side. Google's 2025 claim differs from both companies' earlier ones mainly in that it was independently cross-checked against a real physical measurement (NMR) rather than solely against classical simulation — a stronger form of verification, though not immune in principle to a future, more efficient classical algorithm for the same task.

07 Full Head-to-Head Comparison

Every major dimension, side by side. Advantage marked in bold color where the evidence supports a clear edge.

Category	Metric	Google Quantum AI	IBM Quantum	Edge
Organization	Program founded	~2012 (Google Quantum AI, initially the Quantum Artificial Intelligence Lab with NASA/USRA)	IBM's quantum computing research dates to the 1980s-90s theoretically; IBM Q commercial cloud access launched 2016	IBMlonger institutional history
	Structure	Research division within Alphabet/Google; no separate stock or public financials	Business unit of NYSE-listed IBM Corp (ticker: IBM)	IBMtransparency
	Public roadmap detail	Physics-milestone driven; no named future chips or dated commercial targets published	Detailed, named-chip roadmap (Loon→Kookaburra→Cockatoo→Starling) through 2029	IBMroadmap clarity
	Investment disclosed	Not separately disclosed (folded into Alphabet's broader R&D)	$10B+ cumulative commitment announced June 2026	IBMdisclosed
Hardware	Flagship chip / qubit count	Willow, 105 qubits (two tuned variants)	Nighthawk, 120 qubits	IBMqubit count
	Lattice topology	Near-square, avg. connectivity 3.47, used consistently since Sycamore	Heavy-hex 2021–2024, pivoted to square lattice with Nighthawk (2025)	Tieconverged
	Best reported 2Q fidelity	99.86% (Willow Chip 2, RCS-tuned, company spec)	>99.9% for >50% of pairs (Nighthawk, company spec)	Too close to callnot directly comparable
Headline Claims	Marquee physics result	Below-threshold QEC (2024) + verifiable quantum advantage (2025), both Nature, both currently unrebutted	"Utility" claim (2023), Nature, subsequently matched/beaten by classical tensor-network simulation	Googlecurrently standing
	Independent verification method	Cross-checked against real NMR spectroscopy data, not just classical simulation	Checked only against classical simulation, which subsequently caught up	Googlestronger verification
	Error-correction code family	Standard surface code (higher physical-qubit overhead, mature theory)	Bivariate-bicycle qLDPC (≈10× lower overhead, theory published, hardware unproven at scale)	IBMoverhead theory
Fault Tolerance	Demonstrated milestone	Below-threshold logical error suppression, distance 3→5→7, achieved and peer-reviewed	Long-range c-couplers for qLDPC demonstrated (Loon); full logical-qubit system not yet shown	Googleahead, demonstrated
Fault Tolerance	Target date, full FTQC system	No public date committed	Starling, 2029 (100M gates, 200 logical qubits) — company target	IBMonly company with a date
Facilities, Software & Access	Primary lab / campus	Quantum AI campus, Santa Barbara, CA — integrated hub combining a quantum data center, chip fabrication facility, and R&D under one roof	Distributed across multiple IBM Quantum System Two sites (including a dedicated Quantum Data Center) rather than one integrated campus	Tiedifferent models
	Primary open-source SDK	Cirq — used internally and by researchers, smaller ecosystem than Qiskit	Qiskit — widely regarded as the most broadly adopted open-source quantum SDK, with the largest third-party tooling and tutorial ecosystem	IBMecosystem size
	External hardware access	Limited; primarily research collaborators rather than open self-serve cloud access	IBM Quantum Network: a large, decade-old (since 2016) consortium of enterprise, government, and academic partners with broad cloud access	IBMaccess breadth
	Standardized throughput metric published	Not published as a standing benchmark	CLOPS (circuit layer operations per second): 330,000 on Nighthawk, tracked release-over-release	IBMonly one with a tracked metric

08 Business, Investment & Roadmaps

One company reports quantum results inside a trillion-dollar conglomerate's research budget; the other reports them to public shareholders.

Google / Alphabet

Google Quantum AI is organized as a research division within Alphabet (parent of Google), alongside Google Research and DeepMind — it has never been reported as, or spun out into, a separately-traded entity or disclosed business segment company structure
Alphabet does not break out Quantum AI revenue or spending in its quarterly SEC filings; quantum computing costs are absorbed within Alphabet's overall "Other Bets"/core-research R&D spending, which is not itemized by project
No public target date for commercial quantum revenue has been stated by Google, in contrast to IBM's explicit 2028–2029 guidance (below) — consistent with Google's general public messaging emphasis on scientific milestones over commercialization timelines
Alphabet's overall scale (quarterly revenue in the tens of billions of dollars) means Quantum AI's budget, whatever its size, is a research-and-development line item rather than a business the company is under investor pressure to monetize on any fixed near-term schedule inferred from public filings structure

IBM

IBM Quantum is a unit within IBM's broader Technology segment; as with Google, quantum-specific revenue is not yet broken out as its own reported line in IBM's quarterly filings company filings
June 2026: IBM announced a cumulative commitment of more than $10B to quantum computing, funding its roadmap from current commercial systems through the targeted first fault-tolerant machines company press release
IBM's CEO stated in a July 2026 earnings call that the company expects quantum computing to have a "measurable impact" on IBM's earnings by 2028–2029, tied to the Starling launch target, and separately projected a trillion-dollar addressable market for quantum computing by the late 2030s — both are forward-looking company statements, not realized revenue company guidance
IBM offers cloud-accessible commercial hardware today (IBM Quantum Network, System Two) with a large existing enterprise/government customer base built over a decade, a more mature commercialization posture than Google currently discloses publicly

Reading the business picture honestly: this is not a fair apples-to-apples financial comparison the way the Quantinuum-vs-IonQ page can offer, because neither Google nor IBM discloses standalone quantum-computing financials — Alphabet doesn't break out Quantum AI at all, and IBM's quantum unit sits inside a much larger technology segment. What can be compared is public communication style: IBM has chosen to publish detailed, dated, named-chip commercial commitments (and a specific revenue-impact year), while Google has chosen to publish detailed, dated physics milestones and left commercialization timing unstated. Both are legitimate strategies for a research program embedded inside a much larger company; neither is inherently more credible than the other on business grounds alone.

09 Companies & Ecosystem

Academic roots, partners, and how each program maps onto the AMO/condensed-matter physics job market.

Google Quantum AI Ecosystem

Alphabet Inc. ↗

Parent company (NASDAQ: GOOGL). Quantum AI is funded as part of Alphabet's broader research investment, alongside DeepMind and Google Research, without a separately reported budget.

UC Berkeley (Pines Magnetic Resonance Center)

Academic partner on the 2025 Quantum Echoes verification work, providing the independent nuclear magnetic resonance measurements used to cross-check the algorithm's output on real molecules.

Quantum AI campus, Santa Barbara

Google's integrated hardware hub (built substantially from the UC Santa Barbara group led by John Martinis, who ran the original Sycamore effort) houses chip fabrication, a quantum data center, and R&D under one roof — Google Quantum AI traces to around 2012, when the effort began as the Quantum Artificial Intelligence Lab in partnership with NASA and USRA.

Cirq (open source)

Google's open-source quantum SDK, used internally and by researchers. It has a smaller third-party tooling and tutorial ecosystem than IBM's Qiskit, which is more commonly cited as the field's default entry point for new quantum-software developers.

Google Cloud

Provides limited external/API access to Google's quantum processors for select research partners; far less broadly commercialized than IBM's decade-old public cloud-access model.

IBM Quantum Ecosystem

IBM Quantum Network ↗

A large, long-running (since 2016) consortium of enterprise, government, and academic partners with cloud access to IBM hardware — one of the broadest and oldest quantum-cloud-access programs in the industry.

US national labs & universities

IBM has published joint physics results with numerous academic and national-lab partners across its Eagle/Heron/Nighthawk generations, and its qLDPC code work builds on published academic error-correction theory (bivariate-bicycle codes).

Enterprise customers

IBM markets quantum access bundled with its broader enterprise software/consulting relationships (finance, materials science, logistics use cases), reflecting its business-unit structure versus Google's research-first posture.

Open-source: Qiskit

IBM's Qiskit SDK is one of the most widely used open-source quantum programming frameworks, a significant non-hardware contribution to the broader field's tooling and pedagogy.

Career angle: superconducting-qubit hardware roles draw heavily from condensed-matter and AMO physics, cryogenic engineering, and microwave/RF electronics — dilution-refrigerator experience and Josephson-junction fabrication are directly transferable skills at either company. Google's research-division structure tends to favor more academic-style hiring (PhD-heavy, publication-oriented); IBM's larger, longer-running commercial program offers a wider range of roles spanning hardware, software (Qiskit), and customer-facing quantum consulting. See the Quantum Industry Map for how superconducting roles compare to trapped-ion, neutral-atom, and photonic quantum computing careers, and Quantinuum vs IonQ for the equivalent comparison in the trapped-ion sector.

10 Key Papers & References

Primary sources for every claim above: peer-reviewed papers, preprints, and company/press material, clearly separated.

Google, Landmark Papers & Releases

Quantum supremacy using a programmable superconducting processor

Arute, Arya, Babbush et al., Nature 574, 505–510 (2019)

The original Sycamore "quantum supremacy" claim: 53-qubit random circuit sampling completed in ~200 seconds, estimated at ~10,000 years on a classical supercomputer. Subsequently disputed by IBM and later classical-simulation papers (Section 06).

nature.com ↗

Quantum error correction below the surface code threshold

Google Quantum AI & Collaborators, Nature 638, 920–926 (2024)

The below-threshold demonstration on Willow's QEC-tuned chip: distance-3/5/7 surface codes, consistent ~2.14× logical error suppression per distance increase.

nature.com ↗

Meet Willow, our state-of-the-art quantum chip

Google Quantum AI, company blog (Dec 9, 2024)

Launch announcement and accompanying spec sheet distinguishing Willow's two tuned chip variants (QEC vs RCS), the primary source for the fidelity figures used in Section 04.

blog.google ↗

Our Quantum Echoes algorithm is a big step toward real-world applications for quantum computing

Google Quantum AI, company blog, with Nature publication (Oct 22, 2025)

The 2025 verifiable quantum advantage claim: 13,000× speedup on an OTOC algorithm, cross-validated against real NMR spectroscopy data with UC Berkeley on two molecules (15 and 28 atoms).

blog.google ↗

Discover the Quantum AI campus

Google Quantum AI, company page (accessed Sep 2026)

Source for the Santa Barbara, CA campus description (integrated fabrication, quantum data center, and R&D hub) used in Sections 02 and 09.

quantumai.google ↗

IBM, Landmark Papers & Releases

Evidence for the utility of quantum computing before fault tolerance

Kim, Eddins, Anand et al., Nature 618, 500–505 (2023)

IBM's 2023 "utility" claim: a 127-qubit, 60-gate-layer kicked-Ising-model simulation on Eagle with error mitigation, reported cover story of the June 15, 2023 issue.

nature.com ↗

Efficient tensor network simulation of IBM's Eagle kicked Ising experiment

Tindall, Fishman, White, Stoudenmire, PRX Quantum 5, 010308 (2024); arXiv:2306.14887 (2023)

The primary rebuttal: belief-propagation tensor networks classically reproduce IBM's kicked-Ising results with higher accuracy, and the authors argue the method scales to the thermodynamic limit — directly challenging the classical-intractability implication of the 2023 utility claim.

arXiv:2306.14887 ↗

IBM Unveils Nighthawk and Loon: milestones toward quantum advantage and fault tolerance

IBM, company press release / blog (Nov 12, 2025)

Launch of the 120-qubit square-lattice Nighthawk processor and the Loon c-coupler proof-of-concept chip; source for the connectivity and CLOPS figures in Sections 02–04.

newsroom.ibm.com ↗

IBM lays out clear path to fault-tolerant quantum computing

IBM Quantum Computing Blog (Jun 10, 2025)

The full named-chip roadmap (Loon → Kookaburra → Cockatoo → Starling) and bivariate-bicycle qLDPC code details (the $[[144,12,12]]$ gross code) used throughout Section 05.

ibm.com ↗

IBM Commits More Than $10 Billion to Quantum Computing

IBM Newsroom (Jun 2, 2026)

The investment commitment and framing used in Section 08, funding IBM's roadmap from current commercial systems through the targeted Starling fault-tolerant system.

newsroom.ibm.com ↗

Independent Surveys & Reviews

Questions and Concerns About Google's Quantum Supremacy Claim

Various authors, arXiv:2305.01064

A survey of the technical objections raised against the 2019 Sycamore supremacy claim, useful independent context alongside Google's own paper.

arxiv.org ↗

IBM CEO Expects Quantum Computing to Drive Revenue by 2020s, Trillion-Dollar Value by End of 2030s

The Quantum Insider (Jul 31, 2026)

Independent reporting on IBM's 2028–2029 revenue-impact guidance and long-term market-size projection, cross-referenced against IBM's own statements in Section 08.

thequantuminsider.com ↗