About Peng Cheng Laboratory

Peng Cheng Laboratory

Peng Cheng Laboratory (PCL) is a major research institution in Shenzhen, Guangdong, and part of China's national strategic science and technology forces. Its headquarters was started in March 2018 as a provincial laboratory established by the Guangdong Provincial Government and funded and managed by the Shenzhen Municipal Government. Headquartered in Shenzhen's Nanshan District, it conducts strategic, forward-looking and basic research in network communication, artificial intelligence and cyberspace security. Its director is computer scientist Gao Wen, an academician of the Chinese Academy of Engineering. The laboratory is best known for the Peng Cheng Cloud Brain series of AI computing platforms — Cloud Brain II exceeded 1,000 petaflops and ranked among the world's strongest AI training systems — and for its role in building China Computing NET, the national network linking the country's computing centres.

Establishment and governance

Peng Cheng Laboratory was started in March 2018, in the first wave of Guangdong's provincial laboratories — large, newly founded research institutions through which Guangdong and its cities set out to build strategic research capacity in the Greater Bay Area. Established by the Guangdong Provincial Government, it is funded and managed by the Shenzhen Municipal Government, an arrangement that gives a provincial-level laboratory the resources and speed of China's richest technology city. From Shenzhen's Nanshan District — home to much of the city's technology industry — the laboratory was charged with strategic, forward-looking and basic research in the information field: network communication, artificial intelligence and cyberspace security. It is counted among China's national strategic science and technology forces, the small set of laboratories expected to undertake missions of national significance.

Director: Gao Wen

The laboratory's founding and continuing director is Gao Wen (Wen Gao), one of China's most decorated computer scientists. An academician (member) of the Chinese Academy of Engineering, an ACM Fellow and an IEEE Fellow, he is a Boya Chair Professor at Peking University, where he directs the Faculty of Information and Engineering Sciences, and serves as a deputy to the 14th National People's Congress. He has engaged for decades in research on artificial intelligence, multimedia, computer vision, pattern recognition, image processing and virtual reality, has published six books and more than 300 papers in international journals, and has received seven State Awards in Science and Technology as first accomplisher, together with honours including the National May 1 Labor Medal (2023), the Wu Wenjun AI Highest Achievement Award (2023) and the Ho Leung Ho Lee Foundation's Prize for Scientific and Technological Progress (2022). Gao's research identity — video coding, computer vision and AI systems — is written into the laboratory's agenda, from its Cloud Brain platforms to its 'digital retina' work on visual data for smart cities, a model generation, utilisation and communication paradigm he co-authored with laboratory and Peking University colleagues.

Peng Cheng Cloud Brain

The laboratory's signature infrastructure is the Peng Cheng Cloud Brain series of AI computing platforms. Cloud Brain II, built in Shenzhen in 2020, was described by Gao Wen as, at completion, the world's strongest machine for artificial intelligence training — stronger than the contemporary systems of Microsoft and Google, with computing power equivalent to about 4,000 NVIDIA A100 GPUs. Notably, the machine is built on domestic technology: its processors are Huawei Kunpeng CPUs and Ascend NPUs, connected by an all-optical network with very low latency between nodes. With Cloud Brain II the laboratory could train large models and support scientific research, industrial applications and key technology development. Cloud Brain III, under construction from 2024 with completion expected around the end of 2024 to the beginning of 2025, was designed at 16 exaflops-scale ambition — 16,000 petaflops of intelligent computing power, sixteen times Cloud Brain II — intended to support 6G technology and large-model training as a major scientific infrastructure for intelligent network communication research and data-intensive intelligent science. In June 2026 Cloud Brain III topped both the IO500 Global List and the Research List for storage performance at the International Supercomputing Conference in Hamburg, with a score of 603,334 — a new world record, and the laboratory's twelfth consecutive championship in the IO500 track, continuing Cloud Brain II's long-standing top position. Its storage architecture — control-data separation with I/O direct connection, shared network cards expanding bandwidth and DPUs connecting directly to SSD endpoints at a single-chassis throughput of 500 GB/s — was designed for the storage demands of the AI agent era.

Pengcheng Mind and large models

On its Cloud Brain platforms the laboratory trains its own large models. In 2023 Gao Wen announced the Pengcheng Mind (Pengcheng series) large language model: an open model intended to provide services for enterprises' digitalisation. The Pengcheng Mind series is the subject of the laboratory's international academic outreach — Gao Wen's distinguished seminars abroad are billed on 'Peng Cheng Cloud Brain and Mind Series of Large Model' — presenting the platform and the models as a single system: domestic computing infrastructure purpose-built for open large-model development, at a moment when access to foreign AI accelerators is restricted.

China Computing NET

Beyond Shenzhen, the laboratory is an architect of national computing integration. China Computing NET is the major national computing cluster project linking supercomputing networks and artificial intelligence data centres across China — the computing-power counterpart to a national grid. At the second Greater Bay Area Computing Power Industry Conference and first China Computing NET Conference in Shaoguan in May 2023, Gao Wen announced that Peng Cheng Laboratory would establish China's first computing power dispatching (distributing) centre in Shaoguan, connecting the northern Guangdong city's computing, big data and communication network facilities to China Computing NET and promoting data communication across the Greater Bay Area; at the same conference the Computing Power Cities Alliance was established, formally connecting computing power in provinces including Guizhou, Sichuan and Xinjiang to the network. Gao's public argument for the build-out is economic: he has said that when China's computing power surpasses that of the United States, its GDP is expected to follow — computing power, like electricity in the industrial age, is in his framing the basis on which large models and the industries above them are built, and not something to economise on.

Research fields

The laboratory's research portfolio spans its three founding fields. In network communication, it works on the intelligent networks its Cloud Brain III is designed to serve, including research supporting 6G. In artificial intelligence, it covers large models, computer vision — including the digital retina paradigm for smart cities, in which front-end visual sensors extract compact features for a city brain rather than streaming raw video — and the computing systems AI requires. In cyberspace security, it addresses the security of networks and intelligent systems. As national strategic infrastructure, its platforms are offered beyond its own teams: Cloud Brain exists to let researchers and enterprises train models and run data-intensive science on domestic hardware at national scale.

Place in Shenzhen and the Greater Bay Area

Peng Cheng Laboratory is the flagship of Shenzhen's effort to add fundamental research capacity to its industrial technology base, and a pillar of the Greater Bay Area's science infrastructure alongside the laboratory's sister institutions in Guangming and elsewhere in the region. Its computing networks tie the Bay Area to national resources — the Shaoguan dispatching centre links the region's demand to inland computing supply — while its conferences and alliances make Shenzhen a convening point for China's computing power industry. For Guangdong's provincial laboratory system, PCL is the information-technology anchor, as the Songshan Lake and Shenzhen Bay laboratories are for materials and biomedicine.

Significance

Peng Cheng Laboratory represents a distinctive Chinese answer to a strategic problem: how to guarantee world-class AI computing when the best foreign accelerators may not be available. Its answer — build the machines domestically, at record scale, on Huawei silicon and all-optical networks; top the world's storage rankings with them; train open large models on them; and wire them into a national computing network — has been executed under one director, Gao Wen, since 2018. The laboratory's Cloud Brain series and China Computing NET work make it one of the institutions through which China's computing sovereignty is being built, from a headquarters in Nanshan District, Shenzhen.