IBM and three others enter DARPA's toughest quantum test stage

What happened

The Defense Advanced Research Projects Agency, or DARPA, is the research arm of the United States Department of Defense. It launched the Quantum Benchmarking Initiative (QBI) in 2024. The programme asks whether an industrially useful, fault-tolerant quantum computer can be built by 2033. DARPA defines useful as a machine whose computing value is greater than its total cost to build and run, according to IBM and Quantum Computing Report.

On 7 October 2026, DARPA named four companies for Stage C: Atom Computing, Diraq, IBM and IonQ. Data Center Dynamics reports that they join Microsoft and PsiQuantum, which reached the same level through an earlier DARPA programme called US2QC. That makes six companies publicly known to be in the final stage. DARPA has said that QBI is not a competition between companies but an attempt to find every company on a plausible path.

The programme has three stages so far. In Stage A, companies sent an initial technical concept. In Stage B, they sent a full research and development plan, with risks and ways to reduce them. In Stage C, according to IBM, DARPA's independent verification and validation team moves beyond plans and tests the companies' hardware. DARPA's QBI director, Micah Stoutimore, described Stage C to Data Center Dynamics as the move from plans to proof.

Funding details differ by company. PostQuantum reports that Atom Computing, Diraq and IonQ each describe agreements worth up to 300 million US dollars. Diraq disclosed a first payment of 51 million dollars, and IonQ says its agreement runs through 2029. IBM did not state an amount or a duration. Several Stage B companies, including Quantinuum, QuEra and Xanadu, were not on this list, but DARPA says more companies may advance soon.

The engineering behind it

A normal computer bit is either 0 or 1. A qubit can be in a combination of both, and qubits can be linked so their states depend on each other. This lets a quantum computer handle some problems in ways a normal computer cannot. But qubits are fragile. Heat, vibration and stray electric or magnetic fields disturb them and cause errors. This is general background, not specific to any one company.

Fault tolerance is the answer engineers are working toward. The idea is to combine many physical qubits into one logical qubit, using error-correcting codes, and to detect and fix errors faster than they build up. This needs very good qubits, a large number of them, and fast classical electronics that read the errors and decide on corrections in real time. That is why the cost of a whole machine is hard to estimate.

The four new companies use four different kinds of qubit. Quantum Computing Report describes them as follows. IBM uses superconducting circuits, cooled in multi-part refrigerators below 15 millikelvin, with error-decoding done by programmable chips called FPGAs. IonQ uses trapped ions, single charged atoms held by electric fields. Atom Computing uses neutral atoms held by focused laser beams called optical tweezers. Diraq uses spin qubits in silicon chips made in standard 300 mm chip factories.

The company targets

IBM repeated its plan to deliver its first fault-tolerant system, called Starling, in 2029. PostQuantum notes that IBM's published roadmap describes Starling as a modular design aiming at 200 logical qubits and circuits of 100 million quantum operations. IBM also says its quantum computers are used by more than 340 organisations. Diraq says it aims for 150,000 physical qubits in 2029 and more than two million on a single chip in 2031.

The other two companies also gave details. IonQ says several generations of its Superion systems will be tested through 2029. PostQuantum reports that IonQ has made its first 256-qubit chips at the SkyWater factory and is aiming for customer deliveries in 2027. Atom Computing says Microsoft will provide algorithm support and error-correction codes for its neutral-atom machines during the evaluation.

These are company targets, not results. PostQuantum points out that Diraq has shown control of an eight-qubit array made in a standard factory process, but reliable linking across the whole array is still to be shown. Testing such claims against real hardware is exactly the job DARPA has given its Stage C team. For readers, that makes the programme a useful check on company announcements.

What it means in Nepal

None of the sources mention Nepal, so this section is about the skills involved. A quantum computer is not only a physics experiment. Each of the four approaches needs control electronics, signal generation, cooling systems, lasers or microwave hardware, fast digital processing and software. Quantum Computing Report lists control electronics, cryogenic packaging and software decoders among the things DARPA will test. Much of that is electronics and computer engineering.

For students, the better lesson may be how to judge progress. Quantum computing news often mixes laboratory results with marketing claims. A programme that sets a clear goal, a computer worth more than it costs, and then tests hardware independently, gives a better standard. Asking what was measured, by whom and against what target is a habit that works for any technology story.