Peng Cheng Laboratory's staged rollout makes capacity in operation the near-term measure for Shenzhen strategists assessing domestic AI infrastructure delivery.
Peng Cheng Laboratory has brought 10,000P of domestic computing capacity into trial operation. The laboratory’s announcement and the 21st Century Business Herald placed the release at the World Computing Network Conference (WCNC), held in Shenzhen on October 15–16. The capacity belongs to Peng Cheng Cloud Brain III, a large-scale compute facility that the laboratory is still building.
China News Service independently reported the release and the laboratory’s description of the new phase as “build while operating.” In the reporting, P denotes petaflops, a measure of computing performance; Pandaily also identifies the unit this way. The update therefore establishes an operating-capacity marker during construction.
The February city target provides the comparison point. A Shenzhen government transcript said 4,500P had been powered on December 29 and that the project was expected to build a 16,000P cluster by the end of October. The municipal statement sets 16,000P as a dated target. The October conference announcements establish 10,000P in trial operation.
Trial operation moves the project beyond its initial power-on stage
The laboratory said 10,000P of domestic capacity had formally gone online for trial operation. It described the system as a domestic intelligent-computing supernode architecture intended to address data-transfer bottlenecks and support domestic large-model training and research on networks for AI-driven science. The announcement places the 10,000P update in an operating phase during continuing construction.
The preceding public milestone was smaller and described a different stage. China Science Daily reported on January 2 that 4,500P of first-phase compute boards and auxiliary equipment passed a power-on test on December 29. That report said the project had entered a build-and-debug phase, providing the baseline for the later 10,000P trial-operation disclosure.
The two updates give the facility a visible progression: initial equipment power-on followed by a larger tranche of capacity in trial operation. They also preserve the laboratory’s own framing that building work and operation are proceeding together. The staged sequence is the key operating evidence disclosed so far.
The 16,000P goal remains the benchmark for October
The Shenzhen transcript set out the project’s earlier schedule in unusually specific terms. It paired the 4,500P power-on milestone with an expectation that a 16,000P cluster would be built before October ended. The statement appeared in a February city press conference on Shenzhen’s technology and economic plans, placing the date and scale on the public record months before the latest conference announcement.
The October disclosure narrows the remaining gap to a clear numerical question. The difference between 10,000P in trial operation and the city’s 16,000P cluster target is 6,000P. The 21st Century Business Herald also described 16,000P as the scale of the completed system, reinforcing that the higher figure refers to the facility’s planned configuration.
The two figures serve distinct purposes in assessing domestic compute supply. The trial-operation milestone shows capacity in an operating phase. The municipal target specifies the intended cluster scale and date. A subsequent laboratory or city statement can establish whether the October target was met and what operating status applies to the full figure.
The next update will determine the scale that is actually operating
For corporate strategy teams evaluating local capacity, the sequence separates hardware milestones from capacity that the laboratory says is online. The 4,500P update documented powered equipment, and the 10,000P release documents trial operation. Each update reports a distinct stage of availability.
The laboratory’s announcement also connects Cloud Brain III to domestic large-model training and research into AI-for-science networks. Its staged approach means the facility can supply capacity before the project reaches its planned size. The public record leaves external users, utilization of the 10,000P, and the full cluster’s operating schedule undisclosed.
The immediate event to watch is a dated release from Peng Cheng Laboratory or Shenzhen confirming whether the 16,000P cluster has been built by the end of October and stating its operating stage. The confirmed position is 10,000P of domestic capacity in trial operation.