Quantum computing is entering a testing phase. On 7 October 2026 the US defence research agency DARPA moved four companies into the final stage of its programme to check whether anyone can build a genuinely useful quantum computer [1][2]. Researchers in Germany proposed a cheaper way to measure how ready today’s machines are for error correction [3]. And Infineon said it will help ZuriQ turn a new trapped-ion design into chips that can be manufactured [4]. In the same few days, the UK’s Universal Quantum announced a funding round of more than $100 million [5], IonQ claimed large speed-ups for error-corrected operations [6][7], and Cornell researchers published a peer-reviewed gate result [8].
DARPA’s move, the Infineon partnership and Universal Quantum’s funding are confirmed announcements, though the funding is money, not a technical result. The Jülich and IonQ papers are preprints that have not been peer reviewed, so TSN reports only their abstracts; IonQ’s figures are company-reported simulations. The Cornell paper is peer reviewed.
What did DARPA announce?
DARPA’s Quantum Benchmarking Initiative (QBI) has moved Atom Computing, Diraq, IBM and IonQ to Stage C, its final phase [1]. Each uses a different approach: neutral atoms (Atom Computing), silicon spin qubits made with standard chip processes (Diraq), superconducting circuits (IBM) and trapped ions (IonQ) [1]. All four were among 11 companies in Stage B, announced in November 2025 [1].
They join Microsoft and PsiQuantum, which reached the equivalent final phase through an earlier DARPA programme [1].
In Stage C, DARPA’s own independent team tests the hardware as it is built, to judge whether each system “can be built as specified and perform as designed” [1]. “Stage C is where we begin moving from plans to proof,” said QBI’s managing director, Micah Stoutimore [1]. DARPA stresses that QBI is not a competition: each approach is judged on its own merits, and more companies may reach Stage C [1].
What does “by 2033” mean?
QBI was launched in July 2024 to find out whether any approach can reach “utility scale” by 2033, meaning a machine whose computing value exceeds its cost [1].
Stoutimore now says: “we increasingly expect that someone will build a utility-scale quantum computer by 2033” [1]. That is DARPA’s forecast, not a commitment, and it names no company. IBM separately says it expects to deliver its first fault-tolerant system in 2029 [2], which is IBM’s own target.
A “utility-scale” machine is also not the same as one able to break today’s encryption. Neither DARPA nor IBM makes that claim here.
What does the Jülich preprint propose?
Error correction is how quantum computers are expected to become reliable: extra qubits repeatedly check for errors without destroying the calculation. That depends on operations such as measuring some qubits partway through a calculation (“mid-circuit measurement”), resetting them and acting on the result [3].
Researchers at the Jülich Supercomputing Centre say these steps are usually tested either one at a time or through costly full experiments. Their abstract proposes a benchmark that needs only a few runs: an algorithm built to mimic the pattern of checks used by a given error-correcting code, with the score’s decline over repeated rounds turned into a single effective error rate [3].
According to the abstract, they [3]:
- compared versions with and without mid-circuit measurements on 10 processors from IBM, IQM and Quantinuum, using circuits with up to 2,950 such measurements;
- ran code-shaped tests on Quantinuum’s Helios-1 and H2-1 machines for three code families, up to 81, 91 and 48 data qubits;
- on IBM’s ibm_phoenix, found that the benchmark’s results tracked a full error-correction memory experiment across different regions of the chip.
The paper’s body includes vendor-by-vendor comparisons. Because it has not been peer reviewed, TSN is not reporting them; the abstract does not rank the vendors.
What are Infineon and ZuriQ doing?
ZuriQ, a 2024 spin-off from ETH Zurich, builds trapped-ion processors. Instead of the usual single line of ions, its design holds ions in a flat, two-dimensional grid and moves them across the chip with electric and magnetic fields (a “Penning micro-trap”), avoiding complex junctions [4].
Earlier joint work trapped a 3×3 grid of nine individually controlled ions, which the companies call the largest two-dimensional array of its kind to date [4]. The expanded partnership aims to scale to “significantly larger qubit counts”, using Infineon’s semiconductor manufacturing, packaging and integrated photonics [4]. The release talks of making the hardware “manufacturable” and “industrializing” it; it gives no timeline, volumes or financial terms [4].
What did Universal Quantum raise?
Universal Quantum, a UK-headquartered trapped-ion company founded by two University of Sussex physicists, said on 8 October that it had raised more than $100 million in a Series A round co-led by DCVC and Firgun Ventures [5]. Other backers include EDBI, a Singapore sovereign investor; Integral GlobalTech of Japan; the Australian pension funds Hostplus and NGS Super; and Roblox founder Dave Baszucki [5]. The company calls it the largest Series A ever raised by a UK-headquartered quantum firm; that is its own claim [5].
Its design tiles many small trapped-ion chips together, which it says avoids relying on photonic links or large lasers and cooling systems [5]. The money will fund that work, its first research centre outside Europe, in Singapore, and expansion in the US, Japan and Hamburg [5]. The company also says it is delivering a $77 million contract with the German Aerospace Center (DLR) [5]. This is funding news, not a technical result.
What is IonQ claiming about error correction?
On 6 October IonQ, one of DARPA’s new Stage C firms, described work by its researchers Mark Webster and Nicolas Delfosse [6]. Their paper is a preprint, not peer reviewed, so TSN reports only its abstract [7].
The problem is one of trade-offs. Quantum LDPC codes, a family of error-correcting codes, need far fewer physical qubits per reliable “logical” qubit than the standard surface code, but working on several logical qubits stored in the same block is harder and can slow operations down [7]. The abstract proposes measuring several logical quantities together using only simple “cat” states, with a “scheduler code” setting the order of measurements [7].
In numerical simulations of two codes from IonQ’s planned “walking cat” trapped-ion architecture, the abstract reports a speed-up of nearly 3x for measuring 20 logical quantities, and, combined with a new version of a noise-reduction scheme called CliNR, up to 74x for random Clifford circuits and up to 5x for Toffoli gates [7]. The comparison is with an earlier measurement method, not with a working machine. These are design estimates for a planned architecture, not hardware measurements.
IonQ’s blog adds that the method has already been used in a separate IonQ design study for a machine aimed at 256-bit elliptic-curve cryptography [6]. That, too, is a paper design; no such machine exists.
What did the Nature Communications paper show?
Some quantum computers store a qubit in the many possible states of a microwave oscillator, a “bosonic” code, so that common errors can be spotted and fixed. The main error is photon loss. Such codes have already kept stored information longer than the hardware alone could, but performing operations without spreading errors remains a bottleneck, and gates that tolerate errors had been shown only for one simple type of gate [8].
In a peer-reviewed paper published on 8 October, Saswata Roy, Owen Wetherbee and Valla Fatemi of Cornell University report a full set of logical gates, X, H and T, that are “semi-transparent” to photon loss [8]. When a photon was lost during a gate, the error rate fell five-fold compared with ordinary gates, and the team chained eight gates into a more complex operation [8].
The limits matter. This is one logical qubit in a lab, and “semi-transparent” means errors are reduced, not eliminated [8].
What links them?
Each is about evidence. DARPA is moving from paper plans to testing hardware. Jülich proposes a cheap yardstick for comparing hardware generations before full error-correction experiments are run [3]. Infineon and ZuriQ, and now Universal Quantum’s investors, are betting that trapped-ion designs can be built at scale [4][5]. IonQ’s claims are still simulations [7]; Cornell’s result has passed peer review but covers a single qubit [8].
What this does not prove
- That a useful quantum computer will exist by 2033. That is DARPA’s expectation, not a promise [1].
- IBM’s 2029 target. It is IBM’s own claim [2].
- The Jülich results. They are in a preprint that has not been peer reviewed; only the abstract is reported here [3].
- That ZuriQ’s chips will be mass-produced. The partnership is about scaling and manufacturability; no production plan is announced [4].
- That Universal Quantum’s machines work at scale. The round is funding; “largest UK Series A” is the company’s claim [5].
- IonQ’s speed-ups in hardware. They come from simulations in a preprint that has not been peer reviewed [6][7].
- That bosonic qubits are ready to scale. The Cornell result covers one logical qubit and reduces, rather than removes, loss errors [8].
- Any threat to today’s encryption. IonQ mentions a paper design aimed at elliptic-curve cryptography [6], but no source here claims a machine that can break today’s encryption exists.
The Bottom Line
DARPA now has six companies at its final stage, the four new entrants plus Microsoft and PsiQuantum, and says it increasingly expects someone to reach a useful machine by 2033 [1]. That is a forecast. The Jülich preprint offers a cheap way to check progress on error correction, pending peer review [3], and Infineon is betting its chip-making skills can help trapped ions scale [4]. Universal Quantum’s $100m-plus round shows investors making the same bet [5]. IonQ’s 74x is a simulated design figure [7]; Cornell’s five-fold gain is a peer-reviewed lab result on one qubit [8]. The direction is clear; the proof is what Stage C is for.
Sources
- Drew Jolly, “Atom Computing, Diraq, IBM and IonQ Advance to DARPA QBI Stage C,” HPCwire, 7 October 2026. https://www.hpcwire.com/2026/10/07/atom-computing-diraq-ibm-and-ionq-advance-to-darpa-qbi-stage-c/
- IBM, “IBM Advances to Stage C of DARPA Quantum Benchmarking Initiative,” IBM Newsroom, 7 October 2026. https://newsroom.ibm.com/2026-10-07-ibm-advances-to-stage-c-of-darpa-quantum-benchmarking-initiative
- J. A. Montañez-Barrera and Kristel Michielsen, “Evaluating the performance of QEC primitives on quantum processors at large width and depth,” arXiv:2610.05928 (preprint, not peer reviewed), submitted 5 October 2026. Abstract only. https://arxiv.org/abs/2610.05928
- Infineon Technologies, “Infineon and ZuriQ deepen partnership to advance scalable quantum chips,” press release, 7 October 2026. https://www.infineon.com/press-release/2026/infpr202610-004
- Universal Quantum, “Universal Quantum raises over $100m Series A to scale quantum computing globally,” press release, 8 October 2026. https://universalquantum.com/knowledge-hub/universal-quantum-raises-over-100m-series-a-to-scale-quantum-computing-globally
- IonQ, “How qLDPC with Scheduled Error Correction Unlocks Faster Fault-Tolerant Quantum Computers,” IonQ blog, 6 October 2026 (company source; includes forward-looking statements). https://ionq.com/blog/how-qldpc-with-scheduled-error-correction-unlocks-faster-fault-tolerant-quantum-computers
- Mark Webster and Nicolas Delfosse (IonQ), “Fast logical operations in quantum LDPC codes using simple resource states,” arXiv:2607.16166 (preprint, not peer reviewed). Abstract only. https://arxiv.org/abs/2607.16166
- Saswata Roy, Owen C. Wetherbee and Valla Fatemi, “Error semitransparent universal control of a bosonic logical qubit,” Nature Communications, published 8 October 2026 (peer reviewed, open access). https://www.nature.com/articles/s41467-026-78317-y
Source note: DARPA’s own release was not opened directly (a syndicated copy timed out and returned 403); the DARPA quotes are as reported by HPCwire and are consistent with IBM’s release. For the Jülich paper, only the abstract and author affiliations were used, per TSN’s preprint rule; its vendor-by-vendor comparisons are deliberately not reported. IonQ’s figures come from the preprint’s abstract, which matches the blog’s numbers; the blog’s commercial claims (“higher value per machine”) are not repeated. For the Nature Communications paper, the abstract, author list and affiliations were read; the description of a microwave oscillator draws on the abstract’s “oscillator photon loss” and the paper’s acknowledgements (superconducting cavity).
