By 2027, IBM Plans to Link Two 520-Qubit Chips Instead of Nine 120-Qubit Ones

Quantum computers are usually discussed through the lens of big promises, but the most useful thing in such news is what a company quietly changes in its plan. At the Quantum Developer's Conference, IBM updated its roadmap, and the main change looks like a shift in scaling strategy: instead of many small chips, the bet is on a smaller number of large ones.

What Changed in the Plan

IBM runs two parallel tracks. The first is technical: the company is trialing technologies for future error-corrected machines, and much of it is useless outside that context. The second is public: the Nighthawk processors, on which researchers and developers can already test their algorithms today.

The shift happened in the public track. The previous roadmap promised a 1,080-qubit system assembled from nine 120-qubit chips. Now IBM plans a 1,040-qubit machine built from two Nighthawk r3 chips with 520 programmable qubits each. The number is almost the same, but the path to it is different.

There is also a negative signal. The previous plan assumed that as early as 2026, three Nighthawk chips would be joined into a 360-qubit machine. That did not happen: the current Nighthawk r2, released in September, is sold as a single chip. Ars Technica journalists write bluntly that this may hint at difficulties with chip-to-chip connectivity. IBM representatives, however, did not complain of any problems.

Why the Chip Has "Extra"

Qubits

The main technical news concerns r2. For every programmable qubit, it adds one more physical qubit that does not take part in computation. Its job is to quickly draw energy away from the working qubit and return it to its initial state. That is why the chip has 458 physical qubits but only 120 programmable ones.

Imagine an assembly line where, after every part, the machine needs a long time to be put back in order. Hardware reset removes that pause. IBM claims a speedup of up to 19x for some operations. Not every circuit gets such a gain, but for those that need frequent resets, the difference is substantial.

According to Ars Technica's estimate, with the same proportions r3 could contain around 2,000 qubit-like elements. That is roughly twice as many as IBM's largest processor today.

Chip Stitching: What Exists Already

The main risk in the plan is connecting the chips to one another. According to Oliver Dial of IBM, four L-couplers between two Nighthawk r1 processors in a single cryostat have already demonstrated state transfer fidelity of up to 99.4%. The long-term target is 99.9%. Jerry Chow, IBM's CTO for quantum-centric supercomputing, added that the speed of such a link is comparable to that of qubit connections within a single chip. The company has already shown a modular refrigerator for multiple chips.

Error correction requires a much larger scale. IBM expects to use roughly 300 physical qubits per block of 12 logical ones. One chip will most likely hold one or two such blocks. To reach 100-plus logical qubits, when interesting algorithms can finally be run, the chip must both grow relative to r2 and be linked with at least five such chips. According to the schedule, this is to be done by the end of 2027.

What to Do With This

If you are a developer or product manager looking at quantum computing, the practical takeaways are as follows.

• Do not plan a product around logical qubits in the next couple of years. Throughout 2028, IBM's public track consists of Nighthawk revisions with no new generation.
• Focus on real limitations: r1 manages about 5,000 operations before errors become critical, and the 10,000 mark for the next version remains a goal, not a fact.
• Watch the chip-link metric. If state transfer fidelity approaches 99.9%, the stitching schedule will become more realistic.

The caveat to this news is simple: it is a roadmap, not a result. IBM has already pushed back the timeline for connecting chips once, and repeating such a slip is easier than building a machine. Moreover, 2027 promises both a larger chip and a link of several of them at once. For now, it is better to treat the numbers as stated goals.