China’s Sunway Supercomputer Achieves Breakthrough in AI-Driven Quantum Chemistry Simulation
China’s high-performance computing (HPC) research community has reached a new milestone. Scientists using the Sunway Oceanlite supercomputer have successfully simulated complex quantum chemistry at the molecular scale by combining artificial intelligence (AI) with traditional supercomputing.
The project — which ran across 37 million processor cores — demonstrates how AI-driven models can extend the reach of classical computing into scientific problems once thought solvable only by quantum computers. The achievement represents a significant step forward for China’s AI and supercomputing ecosystem, as well as for the broader global effort to bridge classical and quantum computation.
Understanding the Challenge: Quantum Chemistry at Molecular Scale
Quantum chemistry is one of the most demanding fields in computational science. It seeks to understand how atoms and electrons interact at the quantum level — knowledge that underpins advances in drug discovery, materials science, and clean energy research.
In theory, every molecule’s behavior can be determined by solving the Schrödinger equation, which describes how the quantum state of a system evolves. However, as the number of electrons in a molecule increases, the number of possible configurations — known as the quantum state space — grows exponentially.
For example, modeling a molecule with just a few dozen electrons can require more calculations than the most advanced classical supercomputers can handle. As a result, scientists have traditionally relied on approximation methods to simplify these equations, but such methods quickly lose accuracy when applied to complex or strongly correlated systems.
This is why quantum computers — which process information in quantum bits (qubits) rather than binary ones — are considered the long-term solution for these problems. Yet, since practical quantum computers are still limited in scale and reliability, researchers have been exploring alternative approaches to push classical computing closer to quantum precision.
The Sunway Approach: Fusing AI With Classical Supercomputing
The breakthrough achieved by the Sunway team offers a powerful example of this new direction.
Instead of trying to directly simulate quantum systems with brute-force computation, the researchers used a machine learning model called a Neural-Network Quantum State (NNQS). This AI-based framework can “learn” to approximate the behavior of electrons within a molecule by observing sample configurations and adjusting its parameters to minimize energy prediction errors.
In simple terms, the neural network is trained to represent a molecule’s wavefunction — the mathematical function that encodes the probability of where electrons are likely to be found. By combining physical laws with data-driven learning, the NNQS method merges AI scalability with quantum-level accuracy, offering a promising new route for quantum chemistry simulation.
This AI-augmented approach allows traditional high-performance computers to model molecular systems that were previously out of reach, without the need for fully functional quantum hardware.
Inside the Sunway Oceanlite Supercomputer
The experiment was carried out on China’s Sunway Oceanlite supercomputer, a next-generation system designed and built by the National Supercomputing Center in Wuxi. Oceanlite is based on the Sunway SW26010-Pro processor, which contains 384 computing cores per chip and supports high-precision floating-point formats such as FP16, FP32, and FP64.
What sets this processor apart is its heterogeneous architecture, which includes a small number of management cores that coordinate work across millions of lightweight Compute Processing Elements (CPEs). These CPEs are designed for massively parallel workloads, making them ideal for large-scale simulations and AI-driven scientific research.
To take advantage of this architecture, the researchers built a hierarchical communication model that efficiently synchronized data across 37 million cores. They also implemented a dynamic load-balancing algorithm to distribute workloads evenly, ensuring that no computational resources remained idle during the run.
The combination of algorithmic design and hardware optimization was key to achieving the unprecedented level of efficiency reported in this project.
Record-Breaking Scale and Computational Efficiency
The team’s target system for simulation contained 120 spin orbitals, a measure of the possible electron states in a molecule. Simulating such a system requires an enormous number of variables and interactions to be processed simultaneously — far beyond the capacity of most existing supercomputers or approximation methods.
Using the Sunway Oceanlite supercomputer, the researchers achieved 92% strong scaling and 98% weak scaling efficiency across 37 million CPE cores.
In high-performance computing, “strong scaling” measures how effectively a system speeds up a fixed workload when adding more processors, while “weak scaling” measures how performance holds up when the workload increases proportionally with the number of processors. Achieving near-perfect values in both metrics demonstrates exceptional synchronization between software and hardware — a rare accomplishment even among the world’s most advanced computing systems.
This simulation of a 120-orbital molecular system represents the largest AI-driven quantum chemistry calculation ever performed on a classical supercomputer. It is also one of the first large-scale demonstrations of how machine learning can be tightly integrated with HPC architecture to model quantum systems at a molecular scale.
Implications for AI, Quantum Computing, and Supercomputing
Although this achievement does not replace the need for true quantum computers, it highlights a growing convergence between artificial intelligence, quantum science, and high-performance computing.
Here are some of the broader implications:
-
Advancing Quantum Research with AI Surrogates
Neural-network quantum states (NNQS) allow researchers to explore molecular systems and reaction dynamics with quantum-like precision, even without access to qubit-based machines. -
Bridging the Gap Before Quantum Maturity
Quantum hardware remains limited in qubit count, error correction, and scalability. AI-driven simulations on exascale supercomputers like Sunway Oceanlite provide a practical interim solution for quantum chemistry, condensed matter physics, and materials modeling. -
Demonstrating China’s HPC Capabilities
The project underscores China’s continued leadership in high-performance computing innovation. The Sunway Oceanlite system, along with Tianhe and other national facilities, forms part of China’s broader strategy to develop indigenous, energy-efficient supercomputing architectures. -
Enabling Cross-Disciplinary Research
The integration of AI algorithms into physics-based computation offers new pathways for interdisciplinary science — from computational chemistry to biophysics and nanotechnology. -
Optimizing for Future Hybrid Systems
The software and scaling techniques developed for this project could inform future AI–HPC–quantum hybrid architectures, where machine learning models help optimize quantum workloads and vice versa.
Comparing Supercomputing and Quantum Computing Approaches
Traditional supercomputers like Sunway Oceanlite process information sequentially or in parallel using classical bits, which represent either 0 or 1. In contrast, quantum computers operate using qubits, which can exist in multiple states simultaneously thanks to superposition.
While quantum machines promise exponential speed-ups for certain tasks, their hardware remains experimental and extremely difficult to scale. Noise, decoherence, and error correction remain major barriers.
By using AI to approximate quantum behavior on classical hardware, the Sunway team effectively brings some quantum advantages to existing infrastructure — without relying on fragile qubits. This AI-assisted simulation approach could serve as a crucial stepping stone toward the broader adoption of quantum-enhanced computation in scientific research.
A Global Context: Race Toward AI-Accelerated HPC
China’s breakthrough comes amid a global race to integrate AI into high-performance computing. The United States, Japan, and the European Union are all developing AI-accelerated exascale systems, designed to handle mixed workloads in science, climate modeling, and defense.
In the U.S., systems such as Frontier and Aurora already use GPU acceleration to train large AI models and perform large-scale simulations. Japan’s Fugaku supercomputer has also explored combining machine learning with physics-based models.
What distinguishes the Sunway achievement is the scale of parallelism — 37 million cores working in synchronization — and the use of an entirely homegrown architecture optimized for AI and scientific workloads. This reinforces China’s push for technological self-reliance in semiconductor design and supercomputing hardware.
Future Directions: Toward AI–Quantum Integration
The researchers behind the Sunway project have stated that they plan to continue refining their NNQS framework and exploring its potential for simulating larger and more complex quantum systems.
Longer term, the fusion of AI algorithms, supercomputing infrastructure, and emerging quantum hardware could pave the way for hybrid computational ecosystems. In such systems, classical HPC clusters handle large-scale data management and machine learning, while quantum processors tackle specific, high-precision subproblems.
This integrated approach could transform not only quantum chemistry but also fields such as molecular biology, semiconductor design, and energy storage research, where quantum-scale behavior drives macroscopic performance.
Conclusion
China’s Sunway Oceanlite supercomputer has achieved a notable scientific and technological milestone by using 37 million processor cores to perform an AI-driven quantum chemistry simulation at an unprecedented scale.
While this does not yet replace quantum computing, it demonstrates how the boundaries between AI, HPC, and quantum research are rapidly converging. By leveraging neural networks to approximate quantum systems, researchers are finding practical ways to explore molecular-level phenomena with precision that was previously unattainable on classical systems.
The work reflects both the technical maturity of China’s supercomputing infrastructure and the global trend toward AI-enhanced scientific discovery — a pathway that could shape the next generation of computational science worldwide.