China showcased a fully domestically built AI cluster containing 100,000 compute cards, completed this year, at a Shenzhen computing conference.

The cluster was presented at the 2026 National High-Performance Computing Conference in Shenzhen. The CCTV report describes it as China’s first fully domestically built 100,000-card AI supercluster and states that construction was completed in 2026. It also reports that actual hardware was showcased.

The scale is expressed in compute cards, not servers or racks. The source does not identify the card manufacturer, accelerator architecture, number of servers, or interconnection design. It also provides no measured computing throughput or power-consumption figure.

Ten major application achievements accompanied the presentation. The report describes support for applications such as a digital-twin brain and large-scale universe simulation, but does not enumerate all ten achievements or provide their benchmark results. These remain capabilities described by the report rather than independently documented performance measurements in the supplied material.