What the plan actually says
On 4 September 2026 the Cyberspace Administration of China (CAC) published the Implementation Plan for Promoting the Coordinated Digital and Green Transformation (2026-2030). Seven bodies issued it: the CAC, the NDRC, MIIT, and the ministries of Ecology and Environment, Housing and Urban-Rural Development, Agriculture and Rural Affairs, and Commerce. According to the CAC's published text, the headline computing target is that by 2030, renewable electricity consumption at computing facilities and other key sectors should reach the level of the host province's renewable electricity consumption responsibility weight.
The plan is also notable for what it leaves out. It promises to "significantly raise" energy efficiency in computing facilities and 5G base stations, but sets no numeric efficiency target. It gives no percentage for the renewable share. Anyone who wants a number must look to a different instrument.
The operational detail sits in the computing-facilities section. The CAC text supports liquid cooling, waste-heat recovery and low-power chips. It encourages exploring 800V high-voltage DC power architecture and racks above 100 kW. It directs computing resources toward regions rich in wind, solar and hydropower, and backs direct green-power supply, large-capacity storage and green electricity and certificate trading. It also calls for retrofitting older, small and scattered facilities. For 5G base stations, it favours low-power chips, shared sites, off-peak power use and priority use of green electricity. Yicai's coverage notes that the plan has four focus areas and 14 key tasks, of which green computing infrastructure is one.
The case for tough rules, stated fairly
The strongest argument for binding rules is arithmetic. The NDRC says computing-related electricity use is expected to rise by more than 100 billion kWh a year, on average, during the 15th Five-Year Plan period, according to its explainer on the new weight system. Demand growing that fast can overwhelm grid planning and lock in fossil generation. A regulator that waits for voluntary adoption may find the build-out already done. Seen that way, a vague aspiration would be a serious failure.
Why the provincial-weight design is better than it looks
The plan does not set a rule for each facility. It points computing facilities at a number that already exists in law, the provincial weight, and the machinery behind that number has just been rebuilt. The NDRC and NEA measures on minimum renewable consumption and consumption responsibility weights took effect on 1 August 2026, per Xinhua. They allow shortfalls to be covered by buying green certificates within three months. The consequences for missing a target are warnings, public notices and, for enterprises, a credit record.
This matters for three reasons.
- Compliance is flexible. An operator can meet its obligation through on-site generation, direct green supply or certificate purchases. A fixed rule such as "60% of hours matched" would ignore the fact that a Guizhou hydropower site and a Guangdong site face very different supply conditions.
- The target moves with the grid. The NDRC explainer says provinces with lower current weights get larger annual increases. The data-centre target therefore rises with the local grid instead of being frozen in a standard that ages badly.
- It avoids a PUE-style trap. Efficiency ratios reward whoever reports the best number. Renewable consumption weights measure the thing policymakers care about, which is where the electrons come from.
The plan's siting language also points in the right direction. Steering compute toward renewable-rich regions is a market-compatible nudge, provided the network can carry the traffic. Latency-tolerant training workloads can move. Inference serving users in Shanghai cannot move far. A policy that treats the two alike would hurt innovation, and the plan's "guide" and "support" verbs suggest the drafters understand that.
Where the design is weak
Three gaps deserve attention.
First, the plan stays silent on the number. Operators cannot plan capital spending against "the provincial weight" until provinces publish theirs. Xinhua reports that the NEA will issue 2030 weights alongside the 2026 weights this year, which would fix that gap if it happens on schedule. Until then, investors are guessing.
Second, the NDRC says newly built data centres in national hub nodes already face green-power ratio requirements, and that computing facilities are among the emerging sectors to be brought into assessment. The CAC plan does not say how the new target relates to those ratios. Overlapping instruments invite confusion about which one binds a given facility. They also invite duplicated reporting.
Third, certificate-based compliance is only as good as the certificates. If a certificate bought in one province covers generation that was never physically available to the data centre, the target can be met on paper while the local grid stays fossil-heavy. The measures' ban on provinces restricting cross-provincial certificate trading helps liquidity, but it does not settle whether the claims are additional.
What would make it work
The test is whether Beijing keeps this flexible. Three practices would help: publish provincial weights early, let direct-supply and storage arrangements count in full, and report facility-level energy and renewable data openly so outsiders can check results. The retrofit provision for older, small and scattered facilities needs the same discipline. A retrofit programme backed by diagnostics and support is a useful tool. One that quietly closes small operators would simply consolidate the market around large incumbents.
On current evidence, the plan chooses a measure tied to supply rather than a ceiling on compute, and that is a defensible choice for a sector still scaling. Other jurisdictions weighing data-centre rules, including those tempted by moratoria or hard efficiency caps, could learn from it. Judge it on the 2030 weights and on whether operators can actually buy the power they are told to use.