Latest / Quickly Quantum / September 3, 2026 — 100,000 Circuits a Second: IBM's Speed Bet
Transcript
- One hundred thousand quantum circuits a second — that's the number IBM wants you to sit with today, twenty-five times faster than its own last-generation chip. But does raw speed actually get quantum computing closer to the useful, verifiable advantage IBM has promised, or is it an engineering flex that leaves the harder problem — error rates — untouched? That's our main story. Before we get there, in the headlines: PsiQuantum teams up with a DOE national lab on fault-tolerant software, Rigetti and Purdue speed up a classical solver with a quantum assist, MIT engineers redesign the qubit itself, and a French chipmaker signs two deals to industrialize silicon spin qubits. Welcome back to Quickly Quantum, your daily brief on the quantum frontier. It's Thursday, September 3, 2026. Let's get into it. Let's start with PsiQuantum, a company betting its future on photonic — light-based — fault-tolerant hardware that doesn't exist yet, and seeding the next generation of users before the machine ships. PsiQuantum announced a partnership with Brookhaven National Laboratory, one of the Department of Energy's own labs, where Brookhaven scientists will use PsiQuantum's Construct software — a free, open-access platform for designing and simulating fault-tolerant quantum algorithms — to build tools for utility-scale, meaning genuinely useful-at-scale, workloads. The deal supports DOE's Quantum Genesis initiative, which aims to deploy the world's first scientifically relevant fault-tolerant quantum computer by 2028. On X, PsiQuantum's own account called it 'preparing together for the quantum future,' and HPCwire, the trade outlet that covers this beat, flagged the announcement to its followers too. Now, here's the catch: this is a software agreement, not a hardware milestone. Nothing's been built or published yet — the value only shows up once Brookhaven actually publishes something using these tools. PsiQuantum still hasn't shipped a working large-scale machine, and today's news doesn't change that; it just widens who's lined up to use one when it arrives. Now, Rigetti's story today isn't about a quantum computer solving anything outright — it's about using one to make a classical computer faster. Rigetti and researchers at Purdue University extended what they call a quantum preconditioning framework to a much harder class of problems: constrained combinatorial optimization, the kind of puzzle you hit in logistics routing or scientific computing where you can't just ignore the rules. Their method runs a shallow quantum circuit — QAOA, a near-term algorithm — to extract correlations between variables, then feeds those correlations into a classical solver, Gurobi, to guide its search. According to Quantum Computing Report, which broke this story and which we haven't independently confirmed elsewhere yet, the preconditioned solver reached near-optimal answers up to a hundred times faster than running unpreconditioned, tested across fifty graph problems. That's a real number. But it's a speedup measured against a classical baseline, not a demonstration that quantum hardware solved something a classical machine couldn't — the quantum part here is doing prep work, not the heavy lifting. Meanwhile at MIT, researchers just redesigned the qubit itself to fix a problem that's quietly capping how big these machines can get. Most qubits today only store data — they rely on separate electronics bolted on to actually interact with each other, and that hand-off is fragile and slow. The MIT team, led by professor Kevin O'Brien with graduate student Jeremy Kline and co-author Alec Yen, built what they call an 'arm qubit' — one part stores information, a second part reaches out, like an arm, to talk to neighboring qubits and readout electronics, linked through a device called a quarton coupler. Their simulations — and it's simulation only so far, MIT's own release calls this 'early days' — suggest the design could allow much faster, higher-fidelity operations than current architectures. Yen put it simply: this is a step toward the kind of qubit codesign the field needs for real error correction. Now, no physical chip has been built yet, but if this scales from simulation to silicon, it attacks the same connectivity bottleneck that limits every large quantum processor you'd want to buy today. Silicon spin qubits — built using the same manufacturing techniques as ordinary computer chips — have spent years earning respect without earning much investment. That's been changing lately, and today adds another data point. Quobly, the French silicon spin qubit developer, signed two agreements in the Netherlands — one with the applied-research institute TNO, one with test-equipment maker OrangeQS — to industrialize how these qubits get measured and screened on 300-millimeter silicon wafers, the same production standard used for phones and laptops. The deals were signed during a state visit by French President Macron and Dutch Prime Minister Rob Jetten. Quobly's roadmap calls for cloud deployment of its Alloy chips by late this year, scaling toward one million physical qubits by 2032, backed by a hundred fifteen million euro Series A closed in June. Now, these are memorandums of understanding, not delivered hardware — the real test is whether you'll actually see Quobly's yield numbers show up in a foundry, not just a press release. And in China, state outlet Global Times is reporting a ninety-eight percent transmission efficiency figure for a single quantum router — hardware that would route entangled signals across a future quantum internet. I'd love to tell you which lab, which paper, or whether it's been peer-reviewed, but Global Times doesn't say, and no independent outlet has corroborated the number yet. Treat this one as a claim, not a result — China has been racing the US and Europe on quantum networking headlines for a while, and state media isn't where I'd look for the fine print. Speed is the theme running through today's whole slate, and nobody made a bigger speed claim this week than IBM. Our main story today: does speed alone move quantum computing toward doing something useful, or is IBM just chasing a bigger speedometer? IBM officially released Nighthawk r2, a 120-qubit processor now live on the IBM Quantum Platform under the name ibm_phoenix. The headline number: it runs more than one hundred thousand quantum circuits per second. That's twenty-five times the throughput of IBM's previous generation, Heron, which tops out around four thousand circuits a second. Now, to understand why that matters, you need the plumbing. A quantum processor runs the same circuit thousands of times — each run called a shot — to build up a statistically reliable answer. Between shots, every qubit has to reset back to its starting state, and that reset has always been slow. IBM's older chips used something called conditional reset: measure the qubit, and if it's in the wrong state, fire a pulse to flip it back. That measurement step is itself imperfect, and it could leave qubits idle for hundreds of microseconds between runs while the system waited to be sure. Nighthawk r2 replaces that wait with what IBM calls a dissipative reset — think of it as opening a drain that pulls excess energy straight out of the qubit, no measurement required. IBM says this drops the qubit's effective energy-retention time, called T1, from about two hundred microseconds down to roughly twenty-five nanoseconds during the reset window, shrinking the idle gap between circuit runs to as little as one microsecond. Multiply that saved time across a hundred thousand circuits, and you get the twenty-five-fold jump. The chip keeps its predecessor's 120 qubits but adds real complexity underneath: 218 dedicated couplers and 120 separate reset elements, for 458 total physical components IBM now has to control without one qubit's reset disturbing its neighbors. IBM says the processor has already hit a 2026 roadmap target — producing accurate estimates from circuits containing more than seventy-five hundred gates — and that early tests show some large, repetitive workloads running up to ten times faster without losing accuracy. So here's the tension, and it's one this show keeps running into with IBM: is this progress toward the thing IBM actually promised — verified quantum advantage sometime this year — or is it progress on a metric that mostly helps IBM's own internal experiments? Here's my read: this is real engineering, and it isn't a free lunch either. IBM's own materials are careful to say the chip shows 'no loss in accuracy' on early tests, which is IBM grading its own work — no independent third party has benchmarked Nighthawk r2 yet. Twenty-five times the throughput is a genuine number, and it's worth taking seriously as an engineering achievement: swapping a slow, measurement-based reset for a dissipative one that runs a qubit's reset window down to twenty-five nanoseconds is a hard problem IBM's competitors are chasing too. But throughput isn't the bottleneck that's actually stopping quantum computers from being useful today. That bottleneck is logical error rate — how many physical, error-prone qubits it takes to build one reliable logical qubit through error correction. Nighthawk r2 doesn't claim to move that number. It claims to run the same class of circuits faster, and more of them per second, which speeds up how quickly IBM can gather statistics for its own advantage-tracking experiments, but doesn't obviously change what a paying customer can run today that they couldn't before. Now, this lands right on top of a question we've been sitting with for a while: IBM needs quantum to be the answer to its long-term bet, and whether the market is actually buying that thesis is still unresolved. A twenty-five-times throughput number is a great headline for that story. It's a much smaller answer to the question IBM's roadmap actually needs answered by year's end — can you point to a specific calculation only a quantum computer can do, verifiably, faster than the best classical alternative? IBM's own framing hedges toward that by flagging the seventy-five-hundred-gate observable-estimation milestone, which is closer to the advantage question than throughput is. That's the number I'd watch, not the hundred thousand circuits per second. Time for the Hype Check. I'm putting this at a 6. The hardware is real, deployed, and the physics behind the dissipative reset is a legitimate advance — that's not nothing. But the twenty-five-fold figure is a throughput metric dressed up to sound like a capability breakthrough, and IBM is the only one who's measured it so far. Until an outside lab runs Nighthawk r2 and confirms the accuracy holds at scale, you shouldn't take 'no loss in accuracy' as settled — I'm filing this as IBM got faster, not yet IBM got closer to advantage. So the assumption today's news undercuts is that raw circuit throughput equals progress toward quantum advantage — the real marker to watch is whether independent labs can reproduce IBM's gate-count milestone, not the speedometer number IBM's leading with. If today's mix of speed claims, software deals, and silicon-spin partnerships got your attention, follow Quickly Quantum wherever you listen so tomorrow's episode finds you automatically. This has been Quickly Quantum, an AI-voiced podcast, created and built by a real human using today's cutting-edge technology. Nothing you heard on this show is financial advice. I'm Brian Lampert, and I'll catch you all tomorrow — take care!