Live Nov 18 · Cohort 3Nov 18, 19:00 MSK — free hands-on AI agents live workshop · Live Cohort 3 starts Nov 23Join Nov 18 Live →Cohort 3 (Nov 23)
ReleaseInsight · On RadarGuardian→ original

AI Data Centers Plan 181 GW of Their Own Gas Plants as Grid Queues Grow

One of the biggest bottlenecks for AI data centers is grid connection. According to BloombergNEF, developers in the US, the UK and Europe have announced about 181 GW of on-site gas plants because the queue for a grid connection is too long. That is almost four times the peak electricity demand of the entire United Kingdom. If everything is built, emissions in the US alone would reach about 345 million tonnes of CO₂-equivalent per year.

AI-processed from Guardian; edited by Hamidun News
AI Data Centers Plan 181 GW of Their Own Gas Plants as Grid Queues Grow
Source: Guardian. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceGuardian ↗
Editor-in-Chief
181 GW of on-site gasproposals in the US, the UK and Europe
Key metric #2About 90% of the planned gas plants are in the US
Key metric #3By 2035, AI data center demand will grow almost fourfold, to 2,354 TWh

When people talk about a shortage of capacity for AI, they usually mean GPUs. A new BloombergNEF study points to a different queue: the one for grid connections. Data center developers don't want to wait for years, and they plan to build their own gas-fired power plants right next to the servers.

What the analysts counted

Proposals in the US, the UK and continental Europe add up to about 181 GW of gas generation at the data center sites themselves. For scale, that is almost four times the peak electricity consumption of the entire United Kingdom.

Most of these plans, about 90%, are in the US. If everything is built, those plants alone would emit about 345 million tonnes of CO₂-equivalent per year. That is comparable to the annual emissions of roughly 72.8 million fossil-fuel cars.

Demand, meanwhile, will keep growing. According to the report's forecast, by 2035 electricity consumption by AI data centers will increase almost fourfold to 2,354 TWh, or about 7% of global electricity demand. Power is needed not only for servers but also for cooling systems and auxiliary equipment.

Mark Daly of BloombergNEF calls speed to power one of the main constraints for developers. In his words, the climate impact depends not only on the volume of consumption but also on how quickly grids can connect data centers and whether clean generation appears nearby.

"Temporary" could drag on

Developers claim the gas plants are needed only for a while, until the grid connection arrives. Daly is skeptical. Once a plant is built and running, the economics may push the owner to keep using it. This creates a long-term dependence on fossil fuel, known as lock-in.

In the UK this is already happening in practice. According to Future Energy Networks, in 2024–2025 planned data centers submitted more than 100 requests for a gas connection because of delays in getting access to the grid. The Wapseys Wood project in Buckinghamshire and Quest Park in Bedfordshire want to build their own gas plants. They are among the first data centers being fast-tracked through the nationally significant infrastructure projects regime, which allows bypassing local planning authorities.

Government policy reinforces the situation. According to the Guardian, UK authorities have allowed data centers to build and operate temporary fossil-fuel plants until they are connected to the grid, under the so-called "data center covenant." Environmental campaigners call this a red line. According to Oliver Hayes of Global Action Plan, no one seriously believes that such plants will be temporary.

What this means for the industry

For engineers and product managers the takeaway is simple: energy is becoming as much of a bottleneck as chips. If you are planning large training or inference capacity, your timeline depends on grid connection and local generation as much as on GPU deliveries.

For founders and investors it is also a reputational and regulatory risk. Projects with their own gas get a quick start but may come under pressure from environmental requirements, especially as politicians discuss electrification at a climate summit and the EU holds its first AI summit.

The limitations are worth remembering. These are proposals and plans, not built plants, and not all of them will be realized. The emissions estimates also refer to a scenario in which everything planned is carried out. Even so, the report's key point stands: speeding up grid connections and adding clean generation near new demand is the most reliable way to avoid tying AI infrastructure to gas for decades.

Story

Key Questions Answered

What the analysts counted?

Proposals in the US, the UK and continental Europe add up to about 181 GW of gas generation at the data center sites themselves. For scale, that is almost four times the peak electricity consumption of the entire United Kingdom. Most of these plans, about 90%, are in the US. If everything is built, those plants alone would emit about 345 million tonnes of CO₂-equivalent per year. That is comparable to the annual emissions of roughly 72.8 million fossil-fuel cars. Demand, meanwhile, will keep growing.

What are the key details on "Temporary" could drag on?

Developers claim the gas plants are needed only for a while, until the grid connection arrives. Daly is skeptical. Once a plant is built and running, the economics may push the owner to keep using it. This creates a long-term dependence on fossil fuel, known as lock-in. In the UK this is already happening in practice. According to Future Energy Networks, in 2024–2025 planned data centers submitted more than 100 requests for a gas connection because of delays in getting access to the grid.

What this means for the industry?

For engineers and product managers the takeaway is simple: energy is becoming as much of a bottleneck as chips. If you are planning large training or inference capacity, your timeline depends on grid connection and local generation as much as on GPU deliveries. For founders and investors it is also a reputational and regulatory risk.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…