---
title: "China pairs AI with its power grid: a 2027 target for 'compute-power and electricity' synergy"
date: 2026-11-09
category: Policy & Governance
site: NeuroAI
canonical: https://neuroai.site/a/na-policy-ai-energy-synergy
language: en
---

# China pairs AI with its power grid: a 2027 target for 'compute-power and electricity' synergy

> A September 2025 NDRC–NEA opinion pushes 'AI+' into eight energy sub-sectors and makes compute-power/electricity coordination a stated national goal — but the 2027 milestones are targets, not results.

A data centre in the desert and a wind farm on the steppe are, on paper, two different industries. In Beijing's latest energy plan, they are now the same problem. The logic is simple: the models that run the grid need the grid's power, and the grid's surplus clean energy needs flexible, nearby demand. China has just written that loop into policy.

In September 2025, the National Development and Reform Commission (NDRC) and the National Energy Administration (NEA) issued the *Implementation Opinion on Promoting High-Quality Development of "AI+" Energy* (《关于推进"人工智能+"能源高质量发展的实施意见》, document No. 国能发科技〔2025〕73号, dated 2025-09-04). It is the sector-specific follow-on to the State Council's broader *Opinion on Deeply Implementing the "AI+" Action* (《关于深入实施"人工智能+"行动的意见》, 国发〔2025〕11号, signed 2025-08-21, published 2025-08-26).

**The document is in force — the targets are not yet met**

This is a point worth stating plainly. Both opinions are **published and in force** as central government policy. What they contain, however, are forward-looking milestones, not accomplished facts. The energy opinion sets two checkpoints:

- By **2027**: an initial "AI+energy" innovation system is built; the foundation for **compute-power and electricity coordinated development** (算力与电力协同发展) is solidified; AI empowers core energy technologies with "significant breakthroughs"; **five or more** professional large models (大模型) are deeply applied across power grids, generation, coal and oil/gas; **more than ten** replicable demonstration projects appear; **about one hundred** typical application scenarios are explored; and **one hundred** technical standards are formulated.

- By **2030**: China's energy-sector AI reaches a world-leading level, with a matured compute-power/electricity coordination mechanism and a "green, economical, safe and efficient" compute energy-use model.

So when you read "AI+energy," read it as a 2027 agenda with a 2030 horizon — not a description of today's grid.

**Eight lanes, not one**

The opinion lays out eight implementation directions: "AI+ power grid," "AI+ new energy forms," "AI+ new energy (renewables)," "AI+ hydropower," "AI+ thermal power," "AI+ nuclear power," "AI+ coal" and "AI+ oil and gas." The framing is deliberately broad because energy is where China already has strong digital foundations, high-quality data and dense application scenarios — exactly the conditions the NEA cited when explaining the document.

The NEA was candid about the gaps, too: AI reliability is still insufficient for the highest-stakes decisions (nuclear safety, real-time grid dispatch), data foundations are weak, and there is a "reverse distribution" problem — compute clusters and renewable resources sit in different places.

**Compute-power meets the power system**

The "compute-power and electricity" (算电) thread is the part that links this file to a wider policy push. The NDRC and NEA have been folding it into the new power-system plan, explicitly listing "compute-electricity coordination" (算电协同) as a key task in the 15th Five-Year new-power-system planning. The mechanism is concrete:

- Treat compute facilities as a **priority power-supply object**, matching their load profiles with local green-power resources.

- Expand **green-power direct connection** (绿电直连): a May 2025 NDRC–NEA notice lets new-energy plants supply users "point-to-point" through dedicated lines. As of an April 2026 NEA briefing, **24 provincial-level regions** had issued or drafted supporting policies and **99 green-power-direct projects** had been approved, with total new-energy capacity of **34.05 GW (3405万千瓦)**.

- Run **compute-electricity coordination pilots** exploring joint renewable-output/compute-load forecasting and flexible load control.

Older, still-active rules reinforce the direction. The *Special Action Plan for Green and Low-Carbon Data Centre Development* (《数据中心绿色低碳发展专项行动计划》) — issued in 2024 — requires that by end-2025, new large and ultra-large data centres at national hub nodes keep power usage effectiveness (PUE) at or below **1.2**, and that green-power share at those hubs exceed **80%**.

**Why it matters**

Two trends converge here. First, AI's hunger for both compute and clean power is now a planning problem for the energy ministry, not just a procurement problem for tech firms. Second, "AI+energy" is one of the few AI-policy areas where China pairs a deployment push with an explicit clean-energy and grid-stability condition — the 2030 phrase "green, economical, safe and efficient compute energy-use" is the tell.

For global readers, the takeaway is that China is trying to make its AI build-out a *customer* of its renewable build-out, closing the loop between the two instead of treating them as separate industrial campaigns.

## What readers can do now

- **If you run compute infrastructure in China**: plan for green-power direct connection and PUE ≤ 1.2 at hubs; these are existing requirements, not proposals.

- **If you work in energy or grid software**: the eight sub-sectors and the 100-standards target define a procurement and partnership map through 2027.

- **If you track climate/AI**: watch the 34 GW of approved green-power-direct projects as the first real test of "compute meets clean power" at scale.

## Honest limitations

I relied on official texts (gov.cn, ndrc.gov.cn, nea.gov.cn) and NEA press briefings; I did not independently audit provincial green-power-direct project lists. The "five models / ten projects / one hundred scenarios / one hundred standards" figures are the policy's own 2027 targets, not current achievement levels. The 34.05 GW and 24-province figures are from an April 2026 NEA briefing and may be updated. I have not compared this policy against provincial energy-AI plans individually.

## Key takeaways

- The NDRC–NEA "AI+ energy" opinion (Sept 2025) is in force, but its 2027/2030 milestones — including compute-power/electricity coordination — are targets, not results.

- It spans eight energy sub-sectors and explicitly makes "compute-electricity coordination" part of 15th Five-Year power-system planning, backed by a May 2025 green-power-direct-connection notice.

- Tangible, already-active rules (PUE ≤ 1.2 at hubs, 80% green-power share) mean the clean-power side of the loop is further along than the AI side.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-policy-ai-energy-synergy
Free to quote with attribution and a link to the original.
