Best of Linkedin: Green ICT & Sustainable AI CW 34/ 35
Sustainable AI is shifting from pledges to plumbing. Grid operators in Denmark, California and Ontario moved to price or ration data center power, builders turned to liquid cooling, tape storage and engineered timber to cut physical footprints, and frugal AI tooling showed model choice and workload routing can cut energy use by orders of magnitude without new hardware. Research kept revising the problem's scale upward, from water and land use to energy spent beyond training, while governance work pushed to make green AI claims verifiable. Efficiency only counts once it is measured, disclosed and designed in.
Date
September 4, 2026
Green ICT & Sustainable AI
Thomas Allgeyer

Methodology: Every two weeks we collect most relevant posts on LinkedIn for selected topics and create an overall summary only based on these posts. If you´re interested in the single posts behind, you can find them here: https://linktr.ee/thomasallgeyer. Have a great read!

Listen to our podcast

If you prefer listening, check out our podcast summarizing the most relevant insights from Green ICT & Sustainable AI CW 34/ 35:

Data Center Policy and Grid Access

  • Denmark ranks data centers below households and green-transition projects for grid access under new emergency legislation
  • Spain proposes hourly clean-energy matching for new data centers, with 80 percent or higher as the practical floor
  • California's SB 886 and SB 887, which would recover grid costs from operators and confirm no CEQA exemption, stall ahead of a final House vote despite bipartisan support
  • Canada sets a national framework for data center development centered on community input and sovereignty
  • UK's Crown Commercial Service weights sustainability at up to 20 percent of tender scores under PPN 06/21 and SECR
  • EU's target of tripling data center capacity by 2027 traces to a real estate report on bandwidth growth, not computing demand

Community Impact and Local Resistance

  • 79 percent of Americans back US AI leadership, but only 14 percent want a data center nearby
  • Five North American projects withdrawn or rejected in a single week, while Texas puts 200GW of proposed capacity under audit
  • Ontario creates a new rate class for loads over 1MW, Ofgem proposes charging up to £712,500 per MW for a grid offer
  • Texas researchers trace community resistance to lost agency, not just land or noise
  • The AI Siting Ledger proposes eight disclosure questions on water, power, land and closure liability before construction
  • Geraldton workshop launches the Green Compute at the Edge Challenge across energy, water, infrastructure and community design
  • Mongolia's sovereign AI strategy opens with a draft Data Law before infrastructure capacity
  • Amazon commits to no potable water for cooling and net-positive water return at its India data centers

Data Center Hardware and Physical Infrastructure

  • Dell's IR7000 liquid-cooled rack cuts cooling energy by up to 74 percent and needs 75 percent fewer racks
  • IBM's Climate-Controlled Diamondback tape system cuts costs 84 percent and energy use 97 percent for unstructured storage
  • Engineered timber emerges as a lower-carbon structural alternative to concrete and steel for data centers
  • Closed-loop cooling and reclaimed water are shown to meaningfully cut freshwater demand versus standard systems
  • ISO/IEC 30134 pushes operators beyond PUE toward a fuller KPI set spanning renewable energy, carbon, water and heat reuse

AI Model and Software Efficiency

  • A Nature commentary urges smaller, task-appropriate models over ever-larger ones as the default
  • The Frugal AI Switchboard finds DeepSeek-V3 54 times smaller, 4 times less impactful and 900 times cheaper than a frontier model, at 2 percent less accuracy
  • Honey trims AI coding agent output by 39 percent fewer tokens and 53 percent less code at full test-pass rate
  • Eco Router, a hackathon build, routes AI compute jobs to the grid with the lowest real-time carbon intensity
  • Tret launches an open-source workbench routing tasks to the most carbon-efficient model available
  • Grids run at roughly 30 percent utilization and GPU clusters at 30 to 60 percent, making orchestration the faster lever
  • Runtime governance on agentic AI cuts operational spend 60 to 83 percent by killing runaway loops and idle capacity
  • Code-level profiling shows 20 percent of execution time driving 40 percent of energy use, undercutting "faster is greener"
  • GPT-5.6's automated red-teaming alone is estimated at 308 MWh, a quarter of GPT-3's full pretraining energy

Corporate Investment and Reported Deals

  • WSJ finds roughly $3 trillion in future AI infrastructure commitments across nine major tech companies
  • Meta's $27 billion Hyperion data center shows how such commitments are financed through joint ventures, not direct debt
  • CPP Investments and Equinix complete a $4 billion acquisition of Nordic operator atNorth
  • Microsoft and Magnora ASA advance new Finland sites as the Nordics emerge as a data center hub
  • An insolvent 220MW Portuguese solar plant is pitched for conversion into a green AI compute campus
  • Saudi Arabia announces xAI's expansion and 50MW of new data center capacity at LEAP2026
  • QTS Data Centers publishes its eighth annual sustainability report

Measuring AI's True Environmental Footprint

  • Allianz Research puts data center carbon footprint 57 percent above IEA estimates, projecting $154 billion in annual climate damages by 2030
  • A UNU report projects 945 TWh of electricity and 9.3 trillion liters of water use by 2030, with inference at 80 to 90 percent of AI's energy
  • The same UNU report finds greener energy sourcing can raise water use 30-fold and land use 100-fold
  • CBS finds AI water-footprint estimates vary up to 2,000-fold by methodology, with US data centers below golf-course irrigation levels
  • 2024 data center use reaches roughly 415 TWh and 62 million tonnes of e-waste, only 22 percent of it recycled
  • An analysis of 3,800 servers finds age a poor predictor of performance, pointing to 11.7 million kWh in potential savings

AI Governance, Trust and Compliance

  • Decades of board reviews show all-green dashboards routinely preceding failures, since effort metrics cannot register rising risk
  • NIST IR 8286r1 aims to translate AI risk into enterprise risk register terms boards already use
  • Emissions measurement stays reliable for self-run AI compute but breaks down once AI is bought in as a service
  • The Green Software Foundation's SCI for AI standard is still maturing as the answer for third-party disclosure
  • A RAG-based compliance pipeline for ESRS and BRSR performs well but exposes a persistent human trust gap
  • Australia's Climate Governance Forum flags data centers reaching 13 percent of the national grid by 2035

Events, Training and Community Building

  • Bundesverband Green Software opens September registration for a three-part training series in Germany
  • Sustainable IT Action Netherlands hosts a Delft meetup on the true cost of AI
  • Sustainable and trustworthy AI take main-stage slots at ALL IN, GreenIO London and LEAP2026
  • A Swiss Green Economy Symposium convenes European infrastructure and sustainability institutions
  • New podcast releases span French, Danish and English audiences on measuring digital impact and observability
  • A Tamil Nadu faculty session explores whether AI can drive sustainability without becoming unsustainable itself
  • A new Springer Nature book chapter details resource-efficient software design across the DevOps lifecycle

Subscribe to newsletter

Subscribe to receive the latest blog posts to your inbox every week.

Please confirm your GDPR consent to join our mailing list.
By subscribing you agree to with our Privacy Policy.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
No items found.

Want to see the posts voices behind this summary?

This week’s roundup (CW 34/ 35) brings you the Best of LinkedIn on Green ICT & Sustainable AI

→ 72 handpicked posts that cut through the noise

→ 35 fresh voices worth following

→ 1 deep dive you don’t want to miss