Damao Technology has signed a strategic cooperation agreement with Shanghai Pudong Environmental Energy Development, and the first deliverable, per Jiefang Daily's Shanghai Observer and China National Radio, is a compute-power coordination platform serving the Dawn intelligent computing center in Pudong. That center is the first 10,000-GPU-class cluster in northern Pudong, a national pilot for compute-grid coordination, with a designed power capacity of 18 megawatts.
Compute-grid coordination tackles the most physical constraint an AI data center runs into, which is electricity. AI server power density keeps climbing, and at 18 megawatts of design capacity, grid stability ends up deciding whether the compute itself stays stable. Because the environmental-energy company operates the facility, the utility side and the compute side sit inside the same park, coordinating grid load, green-power supply and training-job scheduling on one platform.
The national-pilot status pushes this beyond a single campus. The hard problems here are well known, from timing mismatch between supply and demand to green-power volatility, plus price swings that land directly in training budgets. The platform's job is to turn adjustable load into a grid-schedulable resource, running training jobs off-peak and trading time for cheaper electricity and flexibility for capacity.
Put a 10,000-GPU cluster in northern Pudong and you're really watching Shanghai's compute buildout take another step. From Lingang to Pudong, the city has spent the past two years laying out intelligent-computing facilities at a brisk pace, treating AI compute as foundational infrastructure for the digital economy. For tenants, the shift is already visible. Power certainty is climbing to the top of site-selection criteria, ahead of rent and network quality.
The observable test sits in the meters. Past PUE, a new set of operating metrics takes shape, covering green-power share, job off-peak rate and electricity cost per token. If the Dawn center runs this platform end to end, compute-grid coordination gets its first replicable reference model.
