A leading Chinese quantum physicist has sought to cool mounting commercial hype around quantum technology, asserting that quantum computing offers no acceleration for large language model training and remains years away from practical, widespread application.
Speaking at the 2026 Inclusion Bund Summit last week in Shanghai, Lu Zhaoyang, executive dean of the University of Science and Technology of China’s Shanghai Research Institute, cautioned that while the field is rapidly advancing, current systems remain limited.
"Quantum computing is very hyped, but right now it is actually quite 'raw'," Lu told attendees. Lu emphasized that quantum computing is not a universal "super machine" designed to accelerate everything.
Instead, he stated, it provides genuine computational speedups only for a narrow set of problems with specific mathematical structures. To date, he noted, there is no universally recognized scientific evidence demonstrating practical acceleration for financial applications, nor are there accepted quantum algorithms capable of boosting modern large language model (LLM) training.
Addressing what he calls widespread misconceptions, Lu explained that quantum computing is not simply equivalent to parallel computing. While quantum superposition allows multiple states to exist simultaneously, measurement collapses the system into a single definitive outcome. True quantum advantage requires pairing specific mathematical problems with custom-designed quantum algorithms. According to Lu, practical use cases remain largely confined to two domains: cryptographic challenges, such as prime factorization and discrete logarithms, and the direct simulation of quantum physical systems, including many-body physics, advanced materials, and chemistry.
Consequently, Lu warned against overestimating quantum capabilities in artificial intelligence and finance. He noted that advanced cryptographic systems like RSA, as well as the elliptic curve digital signatures underpinning cryptocurrencies like Bitcoin, face long-term threats from Shor's algorithm once fault-tolerant quantum computers mature.
Lu also reflected on the shifting landscape of quantum milestones. In 2019, Google’s 53-qubit Sycamore processor completed a random circuit sampling task in roughly 200 seconds, a feat claimed to take traditional supercomputers 10,000 years, marking a milestone known as "quantum supremacy." However, subsequent advancements by the USTC team in collaboration with the Shanghai Artificial Intelligence Laboratory demonstrated that optimized classical algorithms running on standard graphics processing units (GPUs) could handle similar tasks in about 17 seconds with lower energy consumption. Concurrently, Chinese researchers have advanced the "Jiuzhang" optical quantum prototype series, establishing new benchmarks in specific computational tasks and challenging early narratives of Western quantum dominance.
Despite these scientific strides, Lu stressed that practical, large-scale deployment remains a distant prospect. The primary bottleneck is not merely qubit quantity, but quality and stability. As qubit counts rise and quantum circuits deepen, error rates compound rapidly. Without advanced quantum error correction, larger systems become increasingly unreliable.
"If we want to build a skyscraper, we need much more solid materials, and that is quantum error correction," Lu said.
In light of these technical realities, Lu urged both markets and enterprises to exercise restraint. He noted that even elite global quantum research teams are still actively searching for commercially viable algorithms, warning that businesses promoting unverified applications, ranging from speculative drug discovery and quantum finance to outlandish concepts like "quantum pig farming", are pushing science-fiction narratives rather than engineering facts.
The true value of quantum computing, Lu concluded, lies not in an immediate, sweeping transformation of all computing tasks, but in its potential to decisively solve a select few critical, highly complex problems.
For cynical national security observers, it is important to note that public disclosures by leading Chinese researchers like Lu Zhaoyang, who are emphasizing the current limitations and "raw" state of quantum hardware for general commercial tasks like large language model training, may serve a strategic purpose in managing international technological competition. By publicly dampening expectations and steering the narrative away from commercial overhype, top-tier scientific figures in China can effectively downplay domestic breakthroughs in sensitive, dual-use domains.
As highlighted by milestones such as the "Jiuzhang" optical quantum prototype series, which successfully challenged early Western quantum supremacy narratives, China's research apparatus has achieved profound capabilities in specialized quantum processing. Furthermore, with the acknowledged threat that Shor's algorithm poses to foundational encryption standards like RSA and elliptic curve digital signatures, strategic downplaying of broad commercial readiness helps obscure the focused, high-priority strides being made in cryptanalytic-relevant quantum applications, allowing critical national security capabilities to advance under the radar across China.
