What is China’s national computing power network, and how is it supposed to work?

From infrastructure to coordination

The national policy language covers heterogeneous compute, interconnection, scheduling, metering, charging, settlement and coordination with electricity. National Data Administration figures for end-July 2026 put total intelligent compute at 2.45 million PFLOPS (FP16), with 1.45 million PFLOPS — about 60% — inside the national monitoring scope. By early September the trial platform covered 1,462 compute facilities.

Monitoring is not the same as fungibility

Seeing resources on a platform does not mean workloads can seamlessly move among them. CCTV reporting on 24 September said monitoring and cross-provincial invocation mechanisms remain incomplete and many resources are still “islands”; it also showed backbone fibre routes in Kaifeng being shortened because older paths were not designed for today’s compute-latency requirements. Hardware differences, software stacks, data locality, latency, network cost and commercial agreements can all prevent capacity from behaving like one national pool. We therefore distinguish visibility, interconnection, scheduling and economic fungibility.

What changes if coordination improves

If those barriers can be reduced, fragmented capacity could become more usable and workloads could be matched more effectively with available infrastructure and energy. Whether this improves total cost per useful completed task remains the key economic test.

Evidence, claims and limits

• National Data Administration, 28 Aug 2026 — official source
• National integrated computing-power network portal — platform portal
• National Data Administration, 3 Sep 2026 — monitoring-platform coverage
• CCTV via Digital China, 24 Sep 2026 — resource islands, cross-provincial invocation and network adaptation