Selling before perfection: a Chinese startup tests a shortcut to quantum com...

by ZHOU Mo

When QBoson, a privately owned Chinese quantum computing startup, set its targets in early 2024, the ambition sounded almost trivial: sell at least one quantum computer.

At the time, quantum machines were still widely seen as fragile laboratory devices, wrapped in cables and confined to research settings. Among peers, meeting experimental benchmarks was difficult enough; talking about sales felt premature.

Reality moved faster. Orders arrived sooner than expected, and hand-built prototypes could no longer keep up. By late 2025, the company had begun building a factory in Shenzhen, positioning itself as China's first attempt at scaled manufacturing of special-purpose photonic quantum computers.

The move highlights a tension facing the global quantum industry. While the race to build a fault-tolerant, universal quantum computer continues, practical applications remain distant. QBoson is betting that narrower, task-specific machines can reach the market first, even if they fall short of the industry's long-term ideal.

A factory before a breakthrough

Many quantum computing companies in China have emerged from elite universities or operate with support from state-linked institutions. QBoson, by contrast, does not fit that mold. Founded in 2020 by WEN Kai, a Stanford graduate, and MA Yin, a former aerospace engineer, the company has remained fully private and deliberately non-consensus in its strategy.

While competitors debated superconducting qubits, trapped ions or other technical routes, QBoson made a more basic decision. It chose to focus on special-purpose quantum machines rather than pursue a fully general-purpose system.

"The logic was simple," Ma said in an interview with Jiemian News, "We opted for something that might be usable sooner rather than later."

That choice runs against the dominant narrative in quantum computing. Universal machines, in theory capable of running arbitrary algorithms, remain the field’s goal. Progress is measured in qubits, but qubits are fragile. Correcting their errors can require thousands of physical qubits to produce a single stable logical one.

Even global leaders are advancing cautiously. IBM has pushed its roadmap for 2,000 qubits to 2033. Google, despite unveiling a 105-qubit processor, has yet to set a timeline for reaching 1,000 qubits.

"Quantum computers are still experimental devices," a technician at CIQTEK, a Chinese quantum hardware company, told Jiemian News. "In China, qubit counts are usually in the low double digits, and fidelity is hard to sustain."

Ma said feedback from potential customers suggests that meaningful quantum advantage would likely require at least 1,000 error-corrected computational qubits, putting universal systems years, if not decades, away from practical deployment.

A "quantum GPU" bet against the mainstream

Rather than wait for that breakthrough, QBoson chose a narrower path.

Its photonic quantum computers do not attempt to build complex logic gates. Instead, they exploit the physical behavior of photons, allowing systems to evolve naturally toward optimal solutions for specific classes of problems, particularly combinatorial optimization. By avoiding gate-based universality, the machines also sidestep much of the error-correction burden.

Ma compares the approach to classical computing. "If a universal quantum computer is a CPU, what we're building is closer to a GPU," he said, less flexible but efficient at certain parallel tasks.

The analogy reflects a broader bet that quantum computing may follow earlier technology cycles, where specialized accelerators reached the market long before fully general systems emerged.

Not everyone is convinced. Researchers working on universal quantum systems argue that special-purpose machines face limits of their own. Fixed architectures restrict application scope, and advances in classical algorithms and AI hardware continue to narrow the performance gap.

One industry researcher said such systems could solve certain problems faster but added that it remained unclear whether customers would be willing to pay for dedicated quantum hardware as classical alternatives continued to improve.

QBoson's response has been to focus less on hardware novelty and more on usability. Inspired by Nvidia, the company has built a software stack designed to attract existing AI developers rather than a small pool of quantum specialists.

Its Kaiwu SDK, positioned as a CUDA-like compiler, allows developers to write quantum algorithms in Python using PyTorch, without learning quantum physics. On top sits a quantum AI framework based on Boltzmann machines, followed by industry-specific applications.

"The interface doesn't change," Ma said. "That's the point — we're trying to let existing AI developers migrate over, rather than retrain them from scratch."

Early traction, unanswered questions

Commercial traction remains tentative. Like many quantum companies worldwide, QBoson's first customers have been supercomputing centers, which offer public funding, technical validation and relatively low commercial risk.

In 2025, QBoson deployed a system at the Chengdu Supercomputing Center, marking China's first integration of a special-purpose quantum computer into a national supercomputing cluster. Similar projects are under discussion elsewhere.

Within the industry, such deployments are often viewed as exploratory rather than revenue driven. Engineers at a rival quantum firm said these projects tend to align more with local government expectations than sustained commercial demand.

Cloud access offers another testing ground. Through China Mobile Cloud, Alibaba Cloud and Huawei Cloud, users can access QBoson's machines on a pay-per-use basis. The company said about 39% of users come from biopharmaceuticals, including collaborations with the Guangzhou National Laboratory, XtalPi and BGI, while universities account for much of the remaining usage.

Revenue figures have not been disclosed. Industry insiders say most Chinese quantum firms remain loss-making, relying on venture funding rather than repeat customers.

QBoson is also hedging its future. In October 2025, it completed a Series A++ funding round, earmarking capital for general-purpose photonic quantum computing. Ma said the company plans to start with chip-level research before moving toward full systems, with a roadmap expected from 2026.

International competition is intensifying. NTT in Japan has claimed laboratory-scale systems with 100,000 qubits, while D-Wave has signaled a shift toward gate-based machines, underscoring how unsettled the field remains.

The deeper divide lies in ecosystems. Ma estimates that the United States hosts about 50 quantum hardware companies and 500 to 600 downstream algorithm firms, backed by hundreds of venture capital investors. In China, fewer than 10 hardware players remain, with limited downstream innovation and heavy reliance on state-backed customers.

"Globally, IBM and Google are still the leaders," an executive at China Telecom Quantum told Jiemian News. Chinese firms, he said, move quickly but remain in catch-up mode on core technologies.

For now, QBoson's wager is pragmatic rather than visionary: that in a field chasing perfection, there may be a market, however narrow, for quantum machines that work today. Whether that bet leads to a sustainable business, or merely buys time before universal quantum computing arrives, remains an open question.