Level 1 - Absolute Beginner

A quantum computer is a very special kind of computer. It is very hard to make a big one.

One idea is to join small chips together. Then many small chips can work like one big computer.

IBM tried this. The company joined two chips with a wire 80 centimetres long. The wire is made of aluminium.

The test worked well. About 99 times out of 100, the information arrived correctly. IBM wants a working quantum computer by 2029.

computer

A machine that works with information.

chip

A small flat piece inside a computer that does the work.

wire

A long thin piece of metal that carries signals.

join

To put two things together.

centimetre

A small unit of length; one hundred make a metre.

test

To try something to see if it works.

information

Facts or data.

correctly

In the right way, without a mistake.

Level 2 - Elementary

Building a quantum computer gets harder as you add more qubits to a single chip. For several years, researchers have argued that the practical answer is to link several chips instead, the way ordinary computers are built from many parts.

IBM has now shown that this can work. Its researchers connected two chips, each with 120 qubits, using four links made of superconducting aluminium. The team calls them L-couplers. Quantum information travelled 80 centimetres between the chips, which is a long way in this field.

The accuracy was high. On average, 99 percent of the information arrived intact, and the best measurement reached 99.29 percent. The connection also carried entanglement, a shared quantum state between particles, and it kept working after the equipment was warmed up and cooled down three times.

There are problems left to solve. Magnetic interference reduced the quality of the signal, and the couplers need to become more efficient before they can be used in a real machine. The work is a preprint on the arXiv website and has not yet been checked by other scientists. IBM wants to connect hundreds of chips and deliver a fault-tolerant quantum computer by 2029.

qubit

The basic unit of information in a quantum computer.

link

A connection between two things.

superconducting

Carrying electricity with no resistance, usually when very cold.

accuracy

How close something is to being correct.

intact

Complete and not damaged.

entanglement

A quantum link where two particles share one state.

interference

Unwanted signals that spoil a measurement.

preprint

A scientific paper shared before other experts have reviewed it.

Level 3 - Intermediate

Every proposal for a useful quantum computer eventually collides with the same wall. Error correction demands enormous numbers of physical qubits to protect a much smaller number of logical ones, but a single superconducting chip cannot be enlarged indefinitely. Fabrication yields fall, wiring becomes impossible to route, and the refrigerator that holds the chip near absolute zero runs out of room. The alternative, pursued seriously for the last few years, is modularity: build chips of a manageable size and connect them.

IBM has now published the most convincing demonstration of that approach to date. Its researchers joined two 120-qubit superconducting processors with four couplers made from superconducting aluminium, which the team calls L-couplers, and moved quantum states across a distance of 80 centimetres. Average transfer fidelity was 99 percent, with a best figure of 99.29 percent. Crucially, the link transmitted entanglement rather than merely classical bits, which is what distinguishes a quantum interconnect from an ordinary cable, and it continued to function across three thermal cycles of warming and recooling.

The hardware is designed for the cryogenic systems IBM unveiled in August 2026, and the company's stated ambition is to link hundreds of chips inside those units on the way to a fault-tolerant machine by 2029. Jay Gambetta, who directs IBM Research and was not involved in this particular study, described the connection as a step forward. The team itself frames the result as the first multi-channel quantum interconnect between large superconducting processors, and therefore as a building block rather than a finished component.

The caveats deserve equal billing. Magnetic interference degraded signal quality, and the couplers lose more energy than a production system could tolerate, so efficiency work remains. The paper is a preprint on arXiv and has not passed peer review. A 99 percent transfer fidelity also sounds better than it is in context: error-corrected architectures generally want several more nines before a link stops being the weakest element in the chain. What the result establishes is that the modular route is not blocked, which is a different and more modest claim than saying it is open.

error correction

Techniques that detect and fix mistakes in stored or sent information.

logical qubit

A reliable qubit built from many error-prone physical ones.

fabrication yield

The share of manufactured chips that work properly.

modularity

Building something from separate units that connect together.

fidelity

How faithfully information is preserved during an operation.

interconnect

The hardware that carries information between separate chips.

cryogenic

Relating to extremely low temperatures.

Level 4 - Advanced

The quantum computing industry has spent a decade quoting qubit counts, and the figure has been close to meaningless for most of that time. What determines whether a machine can run a useful error-corrected algorithm is not how many physical qubits sit on a die but the product of their coherence, their gate fidelities, and the connectivity available between them. Monolithic scaling attacks the first three and surrenders the fourth: beyond a few hundred qubits, fabrication yield collapses, the input and output wiring becomes a thermal and geometric nightmare, and the dilution refrigerator itself becomes the binding constraint. Modular architectures invert the trade, accepting an interconnect penalty in exchange for the ability to keep adding modules.

IBM's new preprint is the strongest evidence so far that the penalty may be survivable. The team coupled two 120-qubit superconducting processors through four aluminium L-couplers and transferred quantum states across 80 centimetres, achieving an average fidelity of 99 percent and a peak of 99.29 percent. Two details matter more than the headline figure. The first is that entanglement itself was transmitted, which is the operative test for a quantum channel, since a link that carries only classical correlations is of no use for distributing logical qubits across modules. The second is thermal cycling: the couplers survived three warm and recool sequences, which speaks to the mechanical and material robustness that any serviceable system will need, as opposed to a one-off laboratory result that degrades the moment the fridge is opened.

The hardware targets the cryogenic platform IBM introduced in August 2026, and the roadmap calls for linking hundreds of chips within those enclosures en route to fault tolerance by 2029. Jay Gambetta, the director of IBM Research, who did not work on this study, called the demonstration a step forward, which is both corporate understatement and a defensible reading. The team's own framing is more precise: a first multi-channel quantum interconnect between large superconducting processors, offered as a building block.

Scepticism should be aimed at the right target. The problem is not that the fidelity is low in absolute terms but that it is low relative to everything else in the stack. Two-qubit gates on good superconducting hardware now sit comfortably above 99.5 percent and are pushing towards 99.9, and surface-code thresholds are unforgiving about the worst element in a circuit. An interconnect at 99 percent, with acknowledged magnetic interference and energy losses that the authors concede must improve, would be the dominant error source in any modular logical qubit built today. That is a tractable engineering gap rather than a physical barrier, which is precisely the useful conclusion. The modular path has not been shown to work; it has been shown not to be foreclosed, and until peer review has had its turn, even that should be held loosely.

coherence

How long a qubit holds its quantum state before it decays.