Topological qubits, superconducting roadmaps, and AI-powered error correction represent three competing visions for quantum's data center future
Thomas Niedermueller / Getty Images
The phrase "quantum computing is entering the data center" implies a single technology making a single move. It is not. Microsoft $MSFT, IBM $IBM, and Nvidia $NVDA are all pushing quantum into data center infrastructure. But they are doing so with different hardware, different business models, and different bets about where the bottleneck lies.
Understanding the gap between these approaches matters for anyone trying to assess which companies are building real capability and which are positioning for a future that may never arrive on their terms.
Microsoft's quantum strategy rests on a contrarian wager that has been nearly two decades in the making. Rather than use superconducting circuits, trapped ions, or any of the qubit types favored by its competitors, the company built its Majorana 1 chip around a new Topological Core architecture, leveraging a material it calls a topoconductor to observe and control Majorana particles. The argument is that topological qubits are immune to hardware errors. Unlike qubit implementations based on ions, electrons, or photons, topological qubits store quantum information in the topological properties of a physical system rather than in individual particles, making them theoretically more stable.
If that claim holds, it means less overhead spent on error correction and a more direct path to scale. Microsoft says the architecture can fit a million qubits on a single chip small enough to hold in one hand, according to a company announcement. The company is building a fault-tolerant prototype as part of the final phase of the DARPA Underexplored Systems for Utility-Scale Quantum Computing program. The physical form factor is designed for data center deployment: a quantum processing unit inside a standard server rack, drawing about 30 kilowatts.
"A quantum machine is very much part of the ecosystem. It sits right next to a hyperscaler," Zulfi Alam, corporate vice president of quantum at Microsoft, said, as reported by Quartz. The risk is real. Some experts have noted that Microsoft's claims about topological qubits have not been fully supported by peer-reviewed research, and the Nature paper accompanying the Majorana 1 announcement stopped short of providing definitive evidence of Majorana zero modes. Winfried Hensinger, a physicist at the University of Sussex, told Physics World that topological quantum computing is "probably 20–30 years behind the other platforms".
IBM's approach is less speculative. The company has shipped quantum processors on a public roadmap since 2020, using superconducting qubits, which remain the most common qubit type in the industry. Its current Heron processor contains 156 qubits and forms the foundation of its development roadmap. Where IBM distinguishes itself is in the concept it calls quantum-centric supercomputing: quantum processors tightly woven with classical CPUs and GPUs into a single compute fabric, where quantum accelerators tackle the parts of problems that classical machines cannot.
This is not a theoretical framework. At RIKEN in Japan, the company orchestrated its Heron processor in a closed-loop execution with the Fugaku supercomputer, according to IBM's research blog. The result was described as the largest and most accurate chemistry experiment ever performed on a quantum computer.
IBM's roadmap for fault tolerance is detailed. IBM Quantum Starling, targeted for 2029, will be built in a new quantum data center in Poughkeepsie, New York, and is expected to perform 20,000 times more operations than current quantum computers, according to an IBM announcement. Starling will run 100 million quantum gates on 200 logical qubits. A follow-on system called Blue Jay, projected for 2033, would scale to 2,000 qubits running one billion gates.
"We're moving into a data center model where we're actively looking at co-located CPU and GPUs with quantum processors," Jerry Chow, CTO of quantum-centric supercomputing at IBM, said, as reported by Quartz. IBM's vulnerability is that superconducting qubits are noisy, and error correction requires many physical qubits to represent each logical qubit. In a paper published in Nature, IBM researchers outlined a new error-correction scheme, quantum low-density parity-check codes, that would require about one-tenth the number of qubits that surface codes require, according to IEEE Spectrum. Whether that reduction proves sufficient at scale remains an open question.
Nvidia is not building a quantum processor. The company has been explicit about this. "Nvidia doesn't make quantum computers, but we dedicate ourselves to creating accelerated computing stacks to enable quantum computers," CEO Jensen Huang said at the company's Quantum Day event, according to The Next Platform. Instead, Nvidia is building the tools that sit between quantum hardware and useful computation.
In April, the company announced Ising, the first family of open-source quantum AI models, according to an Nvidia announcement. The models target two bottlenecks: calibration (keeping qubits tuned) and error-correction decoding (interpreting measurement data quickly enough to fix errors before they accumulate). Nvidia claims the Ising decoding models are up to 2.5 times faster and three times more accurate than pyMatching, the current open-source industry standard.
The hardware connector is NVQLink, an open interconnect announced in October 2025 that couples quantum processors to GPU systems. NVQLink supports 17 quantum hardware builders, five controller builders, and nine U.S. national labs. It delivers 40 petaflops of AI performance at FP4 precision with a GPU-to-QPU throughput of 400 gigabits per second and latency under four microseconds.
The business logic mirrors Nvidia's broader playbook. As Tom's Hardware noted, Ising's models are open source, but the stack they run on is not: the decoder requires NVQLink's low-latency interconnect, the calibration workflows run through CUDA-Q, and the deployment tooling targets Nvidia hardware. The company remains deeply integrated with the quantum computing industry despite not building quantum hardware. Where IBM and Google $GOOGL have explored machine learning for quantum error correction internally, those efforts are tied to proprietary hardware, according to InfoQ. Nvidia is positioning Ising as a hardware-agnostic, open model layer that can be integrated across platforms.
The differences are real. Microsoft is pursuing a physics-first approach, aiming to solve the error problem at the hardware level. IBM is iterating on proven qubit technology while building a hybrid classical-quantum infrastructure. Nvidia is not touching qubits at all, betting that whoever builds the quantum processor will still need GPUs to make it work.
But the three strategies share one assumption: that quantum processors will operate alongside classical hardware, not replace it. The quantum processing unit sits next to the CPU and GPU, and the workload moves between them. The question is who controls the architecture that ties them together.
For enterprise buyers, the practical implication is that quantum is not one product category to evaluate. It is at least three competing visions of how computation gets reorganized, with different timelines, different risks, and different dependencies. None has won yet.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
