The wires, not the qubits, were holding quantum computers back — and that's why IBM bought Boeing and GM's lab

What has been holding quantum computers back is not the number of qubits — it is the wiring.

Qubits sit inside a cooling chamber kept near absolute zero, while the electronics that control them operate at room temperature outside. Each qubit requires its own wire, and heat flows in along those wires, warming the chamber. Adding more qubits means more wires and more heat, making it increasingly difficult to scale up.

When a US research team moved the qubit control electronics inside the cooling chamber, the error rate for two-qubit operations fell to one-tenth of what conventional approaches produce.

According to Nature, Vol. 655, the team unveiled a quantum processing unit that bundles the qubit chip and its control electronics together, connected by a superconducting cable. Researchers at HRL Laboratories published the findings.

HRL is a private US research and development firm jointly owned by Boeing and General Motors. IBM announced July 23 that it had signed an agreement to acquire the laboratory.

Different companies build quantum computers in different ways. IBM and Google use superconducting circuits. HRL traps electrons inside silicon.

The silicon chip itself was fabricated using a 200-millimeter wafer process common in conventional semiconductor factories. The chip the team built can hold up to 18 qubits.

One challenge remained: where inside the cooling chamber to place the control electronics. Positioning them right next to the qubits shortens the wires but generates heat — a critical problem in a zone that must be kept near minus 273 degrees Celsius.

Keeping the electronics at room temperature outside has been the standard approach. The team chose a middle ground: inside the cooling chamber but away from the qubits, in a zone maintained at minus 269 degrees Celsius.

The link between the control chip and the qubits is a superconducting ribbon cable made by coating a thin film with niobium, a metal known for its superconducting properties.

Within a strip just over 1 centimeter wide, 296 signal lines run side by side in a single layer. Crosstalk between adjacent lines is suppressed to one part in ten thousand.

The ribbon cable carries less than 10 microwatts of heat toward the qubit zone from the minus-269-degree section — roughly one-thousandth of the power needed to light a single miniature LED.

The results were clear. Single-qubit operations produced errors at a rate of 1.7 per 10,000 attempts. Two-qubit operations erred at 3.5 per 1,000 — a tenfold improvement over the previous best for this type of system.

For quantum computers to be practically useful, they need the ability to correct their own errors.

The approach works by distributing a single piece of information across multiple qubits and continuously cross-checking them during computation to detect discrepancies — much like passing the same message through several people so that if one or two get it wrong, the original can still be recovered.

Ordinarily, adding more components increases the probability of failure. Error correction is only meaningful if the opposite holds: adding more qubits must reduce the final error rate.

The team compared a seven-qubit configuration with a three-qubit one. The seven-qubit setup produced errors at a rate of 5 per 1,000 operations; the three-qubit setup erred at 2.4 per 100. Adding more qubits cut the error rate by a factor of 4.7.

IBM announced the HRL acquisition agreement on July 23. The company did not disclose the price and said it expects to close the deal within the third quarter.

IBM has until now focused exclusively on superconducting technology. The company plans to release "Starling," a self-error-correcting quantum computer, in 2029, followed by the larger-scale "Blue Jay" in 2033.

However, a significant gap remains between this research and a practical quantum computer. The chip can accommodate 18 qubits, but the actual experiments used only six or seven.

The team identified several remaining challenges: wiring technology that can be mass-produced, control chips that consume less power per qubit, and methods to reduce performance variation across individual components.

DOI: 10.1038/s41586-026-10754-7

Members of the HRL Quantum Team and Collaborators. A digitally controlled silicon quantum processing unit. Nature 655, 1154-1159 (2026).