IBM has revealed a detailed roadmap to build Starling, the world´s first large-scale, error-corrected quantum computer by 2028, with the goal of making it accessible to users via the cloud in 2029. Starling will be housed at a purpose-built data center in Poughkeepsie, New York, and will consist of a network of interconnected modules, each equipped with cutting-edge quantum chips. According to Jay Gambetta, IBM´s vice president for quantum initiatives, construction of the facility has already begun, signaling the company´s rapid progress toward overcoming quantum computing´s most significant hurdle: error correction.
The field of quantum computing has long grappled with error rates that undermine complex calculations. Unlike classical computers, which use bits, quantum computers utilize qubits capable of representing multiple values simultaneously through superposition. However, their fragile nature leads to frequent computational errors, especially when operations alter adjacent qubits unintentionally. To address this, IBM will deploy a low-density parity check code that requires only 12 physical qubits to construct a single logical qubit in memory—a number that compares favorably to algorithms from competitors like Amazon Web Services, while being much more hardware-efficient than Google´s surface code.
A unique feature of Starling is its real-time decoding capability. IBM´s design leverages an FPGA-driven decoding algorithm that swiftly identifies and corrects errors during computation, a step that industry experts such as Neil Gillespie from Riverlane regard as a major boost to the system´s credibility. While skeptics note that it´s not yet clear which quantum architecture will ultimately prevail, IBM´s modular approach—in which multiple quantum chip modules are networked together—reflects a recent industry trend aiming for scalable, fault-tolerant designs.
Starling is expected to comprise 200 logical qubits and be capable of performing up to 100 million logical operations with high accuracy, dwarfing the capabilities of current machines that operate at only a few thousand. IBM intends to reach this milestone by first introducing incremental machines: Loon, for robust error correction storage on a single chip this year; Kookaburra, capable of both storage and computation next year; and Cockatoo, combining two computation modules by 2027. Ultimately, these advances will culminate in Starling’s network of around 100 modules.
Despite the technical ambition, some experts temper expectations, noting that truly practical quantum advantage may require even more extensive capabilities—potentially billions of error-corrected logical operations. Still, the approach is viewed as both compelling and realistic in its engineering. Looking further ahead, IBM plans to develop Blue Jay, a successor to Starling with ten times the number of logical qubits and a billion logical operations, potentially heralding a new era in solving massive scientific and commercial problems using quantum hardware.
