Quantum Error Correction and the Road to Fault Tolerance with Benjamin Bloom

Ben Bloom has an unusual vantage point on quantum computing: he built one of the world's most precise atomic clocks at JILA, learned semiconductor manufacturing discipline at Intel, watched a superconducting startup scale from the inside at Rigetti, and then founded Atom Computing in 2018 to apply alkaline-earth atomic physics to quantum computation. He returned as CEO in 2024 to push the company through what he describes as the transition from binary technical challenges to engineering execution. The conversation lands at a pivotal moment. In June 2026, Atom published a paper demonstrating sustained, many-round toric-code quantum error correction on neutral atoms — including mid-circuit measurement and mid-computation atom replacement from a reservoir — the first time this has been done on this platform. The same month, the company closed a $100M Series C, signed a $100M Letter of Intent with the Department of Commerce, and announced a photonic networking partnership with Nu Quantum. The Magne system, built jointly with Microsoft for Denmark's QuNorth, is due around the turn of the year. If you want to understand what "fault-tolerant quantum computing" looks like when it stops being a research milestone and starts being an engineering program, this is the episode. What We Get Into Why the toric-code QEC result is different from prior milestones — Bloom explains that the June paper was designed to demonstrate every component exactly as it would be used in a utility-scale system: mid-circuit measurement, atom loss, and live atom replacement from a reservoir, all running for ~90 error-correction cycles. How atom loss goes from fatal flaw to managed engineering problem — Neutral atoms can literally escape the trap mid-computation. Bloom describes how Atom now cycles atoms between data-qubit and ancilla roles and reloads from a reservoir, removing what he calls the last "binary challenge" for the platform. Why the number of qubits in a neutral atom system is a software question as much as a hardware one — Because the array is fully reconfigurable, the same physical system can run a shallow circuit with many logical qubits or a deep circuit with fewer, higher-distance logical qubits. Bloom explains why Atom gives approximate qubit numbers on its roadmap. What "Level 2" means for the Magne system — The ~50-logical-qubit machine being delivered to QuNorth in Denmark is described as the first commercially available system with enough logical qubits and error-correction distance to go meaningfully beyond what's possible with a handful of logical qubits today. Bloom explains how the spec was driven by use cases, not raw qubit counts. How the Microsoft partnership actually works — Bloom describes the relationship as starting in R&D and expanding to go-to-market, with Microsoft contributing error-correction software and stack expertise above the hardware layer, while Atom owns the physical system. Why Atom's real competitive differentiation is invisible — Bloom argues that while all neutral atom companies work with the same atoms, Atom's proprietary work on fast laser steering, pipelining around atom movement, and system integration is where the competitive moat actually lives — and none of it is in public papers. The engineering ceiling on a single neutral atom module — Bloom estimates tens of thousands to hundreds of thousands of qubits may be achievable in a single chamber before networking becomes necessary, and explains why Atom has been building photonic interconnect capability for nearly two years. What utility-scale looks like in 2030 and 2031-32 — Bloom sketches a near-term picture of single systems with tens of thousands of qubits serving national lab and government customers, followed by networked