IBM has unveiled an ambitious plan to build what could become the world’s first large-scale, error-corrected quantum computer—a machine expected to significantly outperform current quantum systems and help solve scientific problems previously considered intractable. The project, named Starling, is slated for completion by 2028, with commercial availability via the cloud targeted for 2029.
The company has already begun constructing a dedicated facility in Poughkeepsie, New York, to house the computer. Each Starling system will be composed of a network of modular units, each containing a set of chips, connected together to scale processing power while preserving coherence. According to Jay Gambetta, vice president of IBM’s quantum initiative, the effort marks a shift from scientific experimentation to real engineering. “We’ve cracked the code for quantum error correction, and now we’ve moved from science to engineering,” he said.
Error correction remains the greatest hurdle in the development of practical quantum computers. Unlike classical computers, which use bits to represent data as 0s or 1s, quantum machines rely on qubits—units that can exist in multiple states simultaneously thanks to the principle of superposition. IBM’s qubits are built from superconducting circuits cooled to near absolute zero. But these systems are prone to noise and interference, and even minor errors can cascade during complex computations, rendering results useless.
To address this, error correction encodes a single “logical qubit” across multiple physical qubits. The cost is high: some schemes require dozens or even hundreds of physical qubits to represent a single logical unit. IBM’s approach, which uses a technique known as low-density parity check code, aims to balance performance and scalability, requiring about 12 physical qubits per logical qubit for memory storage. This is roughly on par with Amazon Web Services’ Ocelot system, which uses nine physical qubits per logical unit, and more efficient than Google’s surface code, which demands closer to 100.
What sets IBM apart, the company argues, is its ability to decode errors in real time using conventional FPGA chips. This on-the-fly decoding—interpreting whether a signal from the machine indicates an error or a valid result—is seen as crucial to making error correction feasible at scale. Neil Gillespie of UK-based quantum startup Riverlane praised the technical credibility of IBM’s decoder but noted that the broader industry is still experimenting with different architectures and approaches.
Starling will mark a significant leap from today’s quantum machines. The system is expected to support 200 logical qubits and perform up to 100 million consecutive logical operations with reliable accuracy. By contrast, current machines can only sustain a few thousand such operations. Gambetta noted that previous error-correction demonstrations by competitors like Google and Amazon were limited to single logical qubits. “They’re small-scale gadget experiments,” he said. “Starling will demonstrate error correction at a much larger scale.”
Still, the question remains whether Starling will be able to solve practical problems. Wolfgang Pfaff, a physicist at the University of Illinois Urbana-Champaign, said that some experts believe a billion logical operations may be needed to run useful algorithms in fields like chemistry or cryptography. While Starling may not hit that threshold, he called it “an interesting stepping-stone regime.” Pfaff, who has received IBM funding but is not involved with the Starling project, said the roadmap appears grounded in experimental and engineering reality. However, he warned that technical setbacks could delay progress. “This is the first time someone’s doing this,” he said.
IBM’s path to Starling is incremental. First, the company plans to showcase robust storage of error-corrected data on a chip named Loon later this year. In 2025, it aims to unveil Kookaburra, a module capable of both storing and processing quantum information. By 2027, IBM plans to interconnect two Kookaburra modules into a larger system named Cockatoo. Starling will follow, scaling up to 100 connected modules.
This modular approach reflects a shift in how the industry is attempting to scale quantum computers—favoring networks of smaller modules over monolithic chips. It also highlights IBM’s broader vision beyond Starling. The company’s next planned quantum system, Blue Jay, is designed to contain 2,000 logical qubits and perform a billion logical operations, reaching the scale that many believe is necessary for economic utility.
Asked about the naming convention, Gambetta smiled: “I like birds.” But beyond the names, the roadmap reflects IBM’s deep belief that the future of quantum computing lies in precise engineering and disciplined scaling.
Whether that vision becomes reality by 2028 remains to be seen—but IBM appears determined to lead the way.
Prepared by Navruzakhon Burieva
