AI and Software Efficiency

• Agent to agent boundaries compound AI energy costs and need governing
• Ponytail agent skill curbs AI coding over engineering with a seven step ladder
• CodeWeave cut PyTorch topk latency by 45.5%
• Carbon-Tracking-MCP estimates AI coding session CO2eq from local logs
• Prompt wording significantly changes LLM inference energy
• AI efficiency metrics are shifting from tokens per watt to outcomes per resource
• AI energy savings remain unproven outside narrow machine learning

‍

Data Centre Infrastructure, Energy and Water

• Liquid cooling becomes the baseline as racks pass 40 kW toward 100 kW
• Data center value is shifting from megawatts and PUE to what power produces
• Cooling design now asks whether chilled water is needed at all
• AWS returns about 15 litres per litre used in Aragon
• Only 1.8% of heat is reused across EU reporting sites
• US data centers need $110 billion of mostly gas power by 2030
• Norway activates a floating data center powered by ocean waves

‍

Policy, Standards and Reporting

• The European Commission data center label grades PUE and WUE from A to G
• Scotland requires impact assessments for data center projects above 50MW
• A WEF playbook finds carbon outcomes are locked in at siting and design
• Microsoft and Meta commitments show ESG returning under new names
• DIST gives organisations an open source schema for technology impact reporting
• The Software Water Handbook becomes an official Green Software Foundation project
• SFIA 9 adds a Sustainable Technology and GreenOps skills view

‍

Tools, Products and Partnerships

• Flexera unifies cloud cost and carbon tracking with Greenpixie
• GreenPT joins the Eden AI gateway as a renewable energy option
• Greenference partners with Opper AI on sovereign green AI
• Vigyanlabs launches FEMTO, an air cooled on premise AI appliance
• The eFootprint tool models environmental and cost trade offs before implementation
• Couchbase and SustainableIT.org argue responsible AI belongs in the data layer
• A new ESG and IT operations framework adds AI skills for green coding

‍

Research, Reports and Events

• A GitHub survey finds 71% concerned about AI’s environmental impact
• Climate Week NYC surfaced about 125 GW of new AI driven power demand
• Nebius reports a 2025 average PUE of 1.25 against 1.54 for the industry
• A ZUG panel flagged missing AI provider emissions data
• Green IO Paris runs 2 to 3 December with speaker calls closing 20 October
• Apache Community Over Code hosts its first Green Software and Sustainability track
• Calls are open for The Green Cloud Blueprint book and a sustainable AI special issue

‍

Thanks to Navveen Balani, Nina Benoit, Denis Kraft, Elisa Filippi, Marjolein Pilon, Oliver Cronk, Ketan Dhende, Melissa Shaffer, Kamran Ahmed, Christopher Misra, Harsh Tyagi, Josh Harbert, Nolwenn Godard, Amy Luers, Jonathan Medrano, Salih ALPTEKİN, Bernhard Lorentz, Muhammed Mirshad DCCA®, Pascal Joly, Lee Hodgkinson, Matthias Haymoz, Shaolei Ren, Cal Innes, Iegor Trieshchov, Carolyn Pistone, Ibrar Ahmed, Dr. Nitish Pathak, Eric Sjöstedt, Jay Monga, Dr. Daniel Jeffrey Koch, Philipp Wiesner, Lubomila J., Christopher Wellise, Marija Verner, Bjorn-Soren Gigler, Denise Mejia, Christoffer Linder, Meghan Horvath, Neil Yeoh, Aydin Mir Mohammadi, Julian Lecker, Gaël DUEZ, Dr. Katia Chaban, Esmeralda Colombo, Kevin Leslie, Ruben Vilhena, Anita Schüttler, Julien Nioche and everyone else who contributed insights to this edition. Find the full list of posts and voices on LinkedIn: