How the two leading quantum hardware roadmaps differ in approach, scale, and what they're targeting.
IBM and Google both build superconducting quantum processors, but their strategies diverge sharply. IBM pursues brute-force scaling with a modular, networked architecture. Google bets on error correction breakthroughs that compound with each generation of chip. Both approaches are scientifically credible. The race is open.
For the record: IBM had 1,121-qubit Condor in 2023, Google's Willow chip in late 2024 demonstrated the first error correction result that beats the theoretical threshold. Numbers matter less than how the qubits behave together.
IBM's roadmap centers on three product lines:
Condor demonstrated that IBM could fabricate and operate a chip with over 1,000 qubits. The catch: 1,000 noisy qubits is less useful than 100 high-fidelity qubits with error correction. Condor's value was proving the fabrication works; the algorithmic value is in their smaller, higher-quality processors.
Heron (156 qubits, late 2023) and its successors are tuned for higher gate fidelity and tunable couplers. IBM's actual scaling plan is to network multiple Heron-class chips via classical communication links — what they call "quantum-centric supercomputing." The model: lots of medium-sized, high-quality chips wired together, rather than one giant chip.
IBM has published a target of 200 logical qubits by 2029 and 2,000 logical qubits by 2033. Whether they hit these depends on reducing physical error rates and increasing the number of physical qubits per logical qubit. As of 2025, their published logical qubit demonstrations were in the low single digits, with one logical qubit supported by hundreds of physical qubits.
Google Quantum AI was founded on the bet that the path to useful quantum computing runs through reducing error rates below the threshold needed for error correction to scale. They published the threshold theorem for the surface code (the leading error correction scheme) and have been chasing that goal since.
Sycamore (53 qubits) ran a sampling task in 200 seconds that Google estimated would take 10,000 years on the world's top classical supercomputer. IBM countered with a claim that classical optimization could do it in days, and the dispute continues. Either way, the experiment demonstrated scaling beyond what classical computers could easily simulate for specific tasks.
Google's Willow chip demonstrated the first result below the surface code threshold — meaning that as you add more physical qubits to encode a single logical qubit, the error rate actually decreases. This is the property error correction schemes need to scale, and it's the result that makes fault-tolerant quantum computing plausible rather than theoretical.
Willow has 105 qubits arranged in a square lattice, with gate fidelities in the 99.7-99.9% range. These are better than any previous superconducting processor. The result was published in Nature in December 2024.
Google's strategy is that error correction at scale requires the physical error rate to be below threshold. Once below, adding more qubits makes the logical qubit better rather than worse. Willow demonstrated this works. The remaining work is engineering: produce larger chips with the same fidelities, or improve fidelities further while scaling qubit counts.
IBM has more qubits. Google has lower error rates. Both are superconducting. The architectures are similar at the hardware level (transmon qubits, microwave control, dilution refrigerator cooling).
The bigger difference is strategic. IBM builds a platform — cloud access, Qiskit software, multiple processor sizes for different use cases — and treats quantum computing as infrastructure. Google treats it as a research problem that needs one or two major breakthroughs before commercial focus makes sense.
For users who want to run experiments today, IBM Quantum Platform gives easier access. Google Quantum AI offers research partnerships and selective cloud access. The cost of either is below the cost of building your own dilution refrigerator, which is the relevant comparison.
China's efforts are largely unpublished in detail but include photonic (Jiuzhang) and superconducting (Zuchongzhi) approaches. IonQ, Quantinuum, and Rigetti focus on trapped ion and other architectures. Microsoft's topological qubit program remains unproven. PsiQuantum is betting everything on photonic quantum computing at scale. None have shipped a system that beats IBM and Google on the combined metrics of qubit count, fidelity, and connectivity.