China’s national supercomputing network tops 500 AI agents online

The significance here is less the raw agent count than the shift in who can run serious scientific computing: conversational assembly and reusable components lower the barrier for researchers without coding expertise. If that holds, demand for compute shifts from a few large labs toward a much wider, more fragmented user base, which is exactly the kind of load a national network is built to pool. The open question is whether usage concentrates in a handful of the cited scenarios or spreads across the full matrix.