• Exponential error correction & quantum advantage. Google’s 105‑qubit Willow chip implemented 3×3–7×7 surface‑code grids; each increase in code distance halved the logical error rate (a 2.14× error reduction over earlier devices) and doubled the lifetime of a logical qubit . It ran a random‑circuit‑sampling benchmark in under five minutes, while a classical supercomputer would need 10²⁵ years, demonstrating below‑threshold error correction and real‑time decoding

• IBM’s fault‑tolerant roadmap. IBM’s Starling quantum computer (planned for 2029) will deliver about 200 logical qubits capable of 100 million operations . The company’s new quantum low‑density parity‑check (qLDPC) codes reduce the physical‑qubit overhead by ~90 % . Their modular Starling/Blue Jay systems aim for more than 1 000 logical qubits and incorporate fault‑tolerant, universal and adaptive error correction

• Accelerated qubit scaling. IonQ plans to grow from ~100 physical qubits in 2025 to 10,000 qubits by 2027, 20,000 by 2028 and more than 2 million qubits (≈40 K–80 K logical qubits) by 2030 . Strategic acquisitions provide photonic interconnects and 2‑D ion‑trap chips that increase entanglement rates by 50× and trap density by 300× , while early applications like a blood‑pump simulation already show ~12 % speed‑ups over classical HPC

• Continuous operation. Harvard physicists built the first quantum computer that runs continuously, not just for milliseconds. Their neutral‑atom machine uses an “optical lattice conveyor belt” and optical tweezers to replenish lost atoms, maintaining about 3 000 qubits and running for more than two hours (in principle indefinitely) . This method removes atom‑loss as a bottleneck and clarifies a path to scalable, always‑on quantum devices.

• Neutral‑atom mega‑array. At Caltech, researchers created a neutral‑atom array of 6 139 cesium atoms using optical tweezers, shattering the previous record of 1 180 qubits. Despite the huge array, the qubits maintained superposition for an average of 13 seconds with 99.98 % control fidelity . The team envisions scaling to >10 000 qubits soon and millions within a decade.

• Hour‑long “cat” qubits. French startup Alice & Bob unveiled “Galvanic Cat” qubits where bit‑flip errors are suppressed for 33–60 minutes, millions of times longer than typical superconducting qubits . With 94.2 % operational fidelity, the design embeds error correction into the qubit itself, reducing overhead by ~200× and aiming for 100 logical qubits by 2030.

• Real‑time error‑correction stack. UK company Riverlane will deploy Deltaflow 2 — the world’s first dedicated real‑time quantum‑error‑correction system — at Oak Ridge National Laboratory. It decodes quantum data with ultra‑low latency to support millions to trillions of reliable operations, embedding error correction as a core layer in hybrid quantum‑HPC systems .

• 3‑D superconducting architecture. Fujitsu and RIKEN unveiled a 256‑qubit superconducting quantum computer that uses a scalable 3‑D architecture with 4‑qubit unit cells, quadrupling qubit density inside the dilution refrigerator . The machine will be made available to global users in fiscal 2025, with plans to scale to 1000 qubits by 2026 .

• European star‑topology processor. The LUMI‑Q consortium inaugurated VLQ, a 24‑qubit superconducting quantum computer arranged in a star‑shaped topology. This connectivity minimizes SWAP operations and is integrated with Europe’s high‑performance‑computing infrastructure, giving researchers across Europe access to hybrid classical‑quantum experimentation

• Classical acceleration & AI for quantum error correction. GPU‑accelerated CUDA‑Q decoders developed at the University of Edinburgh double the speed of qLDPC decoding . An AI‑based decoder built with NVIDIA PhysicsNeMo achieved 50× faster decoding , while GPU‑accelerated simulation of open quantum systems delivered up to 4 000× performance boosts . Such integration of accelerated computing and machine‑learning tools is vital for designing better qubits and managing error correction at scale.