Energy Nov 6, 2026 · The Motley Fool
Oklo (OKLO) Q3 2026 Earnings Call Transcript
Earnings call where DeWitte discusses Oklo's progress and expanding customer interest in powering data centers wherever power can be found.
Jacob DeWitteOklo
// updated daily · 125 signals
Field signals
What the people leading data-center sustainability just published — water, cooling, heat reuse, energy and measurement. Every link is checked before it appears. Summaries are ours; the work is theirs.
Energy
Energy Oct 27, 2026 · MIT Technology Review
How solid-state transformers could help power the AI boom
Article on the transformer shortage straining grids as data centers add demand. It explains how semiconductor-based solid-state transformers might first be used to supply power to data center server racks.
Casey CrownhartMIT Technology Review
Energy Oct 27, 2026 · Bloomberg (via Yahoo Finance)
Crusoe CEO on AI Energy Demand and Building in Texas
Video interview in which Lochmiller discusses AI compute demand, Texas data center scrutiny, Crusoe's closed-loop cooling water use, pairing sites with wind and gas power, and stranded hydro abroad.
Chase LochmillerCrusoe
Energy Oct 23, 2026 · Koomey Analytics blog
PBS show on energy and environmental effects of data centers
Blog post on a PBS show about data centers' energy and environmental effects in which Koomey was interviewed. He criticizes its Jevons paradox framing and links to his own articles on forecasting.
Jonathan KoomeyKoomey Analytics
Energy Oct 19, 2026 · CNBC
Google's head of energy and power on how AI data centers can help improve the grid
In a CNBC video interview, Corio discusses how Google's AI data centers, including a planned Minnesota site with an Xcel Energy agreement, could support the power grid.
Amanda Peterson CorioGoogle
Energy Oct 13, 2026 · University of Chicago (UChicago News / EPIC)
What are data centers, and why do they need so much power? An expert explains
In this interview, Chien explains data center power and water demands, projects their growing share of US electricity, and proposes grid flexibility, colocated generation, and community disclosure measures.
Andrew A. ChienUniversity of Chicago
Energy Oct 13, 2026 · Felicis
Power Player (Crusoe CEO Chase Lochmiller on Powering the AI Buildout)
Interview-based profile in which Lochmiller calls energy AI's main bottleneck and describes Crusoe's energy-first, vertically integrated model, its modular Spark units, and its battery and nuclear power efforts.
Chase LochmillerCrusoe
Energy Oct 6, 2026 · Google (The Keyword blog)
Why we're backing America's existing nuclear plants
Announces a Google deal with Constellation Energy for 890 MW of nuclear uprates at six PJM plants, taking Google's total enabled uprate and restart capacity above 1.5 GW.
Amanda Peterson CorioGoogle
Energy Oct 3, 2026 · The Twenty Minute VC (20VC) podcast
20VC: The Future of Datacentres: What You Need to Know | Why Everyone Gets GPU Depreciation and AI's Energy Costs Wrong | Who Really Makes Money From AI & Why Most Moats Don't Exist with Chase Lochmiller
Crusoe's CEO talks about what limits AI data center construction, whether data centers raise household energy bills, how many planned projects will get built, and GPU economics.
Chase LochmillerCrusoe
Energy Oct 2026 · Uptime Institute Intelligence
Data center energy growth: substantial but constrained
Projects global data center energy use rising from about 500 TWh in 2025 to 1,100 TWh by 2030, limited by power, supply chains and overstated announced capacity.
Jay DietrichUptime Institute
Energy Sep 30, 2026 · X (@MLiebreich)
Post on US data centre capacity needs
Liebreich argues the US does not need around 1 TW of new data centres and instead needs roughly 50 GW by 2035.
Michael LiebreichLiebreich Associates
Energy Sep 24, 2026 · MIT Technology Review
AI is dominating the conversation at Climate Week
Reports that AI dominated New York Climate Week. Covers climate-tech funding flowing to data-center products, rising Big Tech emissions as data centers expand, and new gas plants built to meet demand.
Casey CrownhartMIT Technology Review
Energy Sep 15, 2026 · AI Infra Summit 2026
From Challenge to Opportunity: Rethinking Power in AI Data Centers (AI Infra Summit 2026 speaker session)
Conference talk in the AI Data Center track on rethinking power in AI data centers, presented by Gadepally as co-founder of Bay Compute, which optimizes power and cooling.
Vijay GadepallyMIT Lincoln Laboratory
Energy Sep 15, 2026 · AI Infra Summit 2026
Reality Check: Evolving the Grid to Meet Growing Power Demand
A session in the AI Data Center Track on adapting the power grid to rising demand, with Norris speaking as Google's Head of Market Innovation.
Tyler NorrisDuke University
Energy Sep 5, 2026 · CNBC
Fervo Energy's enhanced geothermal project aims to power data center boom
CNBC report featuring Latimer on how Fervo's enhanced geothermal development aims to supply power amid surging demand from hyperscalers and data centers.
Tim LatimerFervo Energy
Energy Aug 28, 2026 · Koomey Analytics blog
My keynote address for the Ohio Manufacturers' Association
Keynote on electricity demand growth and data centers, arguing that demand growth is local rather than uniform and that media accounts have inflated it.
Jonathan KoomeyKoomey Analytics
Energy Aug 26, 2026 · Volts (David Roberts podcast)
Making data centers flexible so they can serve the grid rather than stress it out
Podcast interview in which Sivaram explains how Emerald AI's Conductor software shifts, slows or relocates AI workloads and dispatches onsite resources so data centers can connect to grids faster.
Varun SivaramEmerald AI
Energy Aug 25, 2026 · CNBC
Emerald AI CEO Varun Sivaram: We transform AI data centers into flexible grid allies
A video interview in which Sivaram explains how Emerald AI helps data centers adjust their power use when the grid is strained, and discusses current AI energy demand.
Varun SivaramEmerald AI
Energy Aug 25, 2026 · Emerald AI blog
Emerald AI Raises $150M Series A at $1.05B Valuation
Sivaram announces a $150M funding round for software that lets AI data centers cut power use during grid stress. He describes demonstrations, utility partnerships and plans to scale.
Varun SivaramEmerald AI
Energy Aug 22, 2026 · arXiv
PowerSlider: Exploiting Phase Asymmetry for LLM Serving under Demand Response
Proposes an LLM serving system that meets time-varying power caps from grid demand-response events. It directs power cuts to the inference stages where they degrade performance least.
Udit GuptaCornell Tech
Energy Aug 15, 2026 · Business for Good Podcast
Ep. 196 - The Climate Startup That Raised Billions and Made It Big: Fervo Energy's Geothermal Breakthrough
Latimer discusses Fervo's IPO, its Cape Station project in Utah, and enhanced geothermal as firm carbon-free power for a grid strained by AI and data center demand.
Tim LatimerFervo Energy
Energy Aug 12, 2026 · Cleaning Up (Substack)
AI Has An Electricity Problem (Cleaning Up Episode 270)
Liebreich hosts EPRI chief Arshad Mansoor to debate how fast AI data center power demand will grow, grid strain, flexible connection, public opposition and nuclear's role.
Michael LiebreichLiebreich Associates
Energy Aug 12, 2026 · Earnings Whispers (Fervo Energy Q2 2026 earnings call)
Fervo Energy Q2 2026 earnings call transcript
Latimer names AI data centers as a major driver of US load growth. He presents behind-the-meter geothermal as a bridge to grid timelines and acknowledges local pushback.
Tim LatimerFervo Energy
Energy Aug 12, 2026 · Cleaning Up (podcast)
AI Has An Electricity Problem | Ep270: Arshad Mansoor
Liebreich and EPRI's CEO debate how much new US data-centre power demand to expect. They also cover grid queues, data-centre flexibility, public opposition and the role of small modular reactors.
Michael LiebreichLiebreich Associates
Energy Aug 7, 2026 · Oklo earnings call (transcript via StockAnalysis)
Oklo Q2 2026 Earnings Call Transcript
DeWitte covers plans to power Meta from a 1.2 GW Ohio campus, local opposition to data centers, the low water use of advanced reactors, and heat-driven cooling work with Vertiv.
Jacob DeWitteOklo
Energy Jul 28, 2026 · Uptime Institute
Uptime Institute Global Data Center Survey Results 2026
Uptime's 16th annual survey finds continued industry growth driven by high-density and AI workloads. Operators face constraints from power availability, grid reliability, costs, supply chains and staffing.
Andy LawrenceUptime Institute
Energy Jul 24, 2026 · CNBC
Oklo CEO Jacob DeWitte: AI is transforming the future of nuclear energy
In a CNBC video interview, DeWitte argues that AI is speeding up reactor design and engineering productivity, helping move advanced nuclear deployment forward.
Jacob DeWitteOklo
Energy Jul 1, 2026 · ESA Φ-lab Collaborative Innovation Network (Φ-talk)
Energy for AI and AI for energy
In an online talk with Thomas Spencer, Singh presents IEA findings on the electricity needs of AI data centres and how Earth observation could track data-centre growth.
Siddharth SinghInternational Energy Agency
Energy Jul 2026 · USENIX OSDI '26
SPADE: Signal-Aware DAG Scheduling and Dynamic Provisioning for Data Processing Clusters
Co-authored paper presenting a scheduler that uses job DAG structure and time-varying signals like carbon intensity or energy price, cutting such costs 32.9% on a Spark prototype without losing throughput.
Noman BashirMIT
Energy Jun 30, 2026 · Google (The Keyword blog)
Read Google's 2026 Environmental Report
Google introduces its 11th annual environmental report, which covers managing its operational footprint, scaling clean energy for growing data center demand, and using AI to help cut emissions.
Kate BrandtGoogle
Energy Jun 18, 2026 · Lawrence Berkeley National Laboratory (OSTI)
United States Data Center Energy Usage Report: 2025 Update
LBNL bottom-up model estimating that US data centers could use about 11.8% of national electricity by 2030, with scenarios ranging from 9.5% to 15.3%.
Arman ShehabiLawrence Berkeley National Laboratory
Energy Jun 9, 2026 · CNBC (Closing Bell: Overtime)
Crusoe CEO weighs in on concerns around waining data center demand
In a TV interview, Crusoe's CEO addresses worries that demand for data centers may be weakening and gives his view of the AI infrastructure sector's current state.
Chase LochmillerCrusoe
Energy Jun 3, 2026 · MIT Technology Review
How virtual power plants could provide energy for data centers
Covers a Google deal with Voltus to use a virtual power plant in the PJM grid to help supply its data centers, and questions whether enough customers will opt in.
Casey CrownhartMIT Technology Review
Energy Jun 2026 · Lawrence Berkeley National Laboratory
United States Data Center Energy Usage Report: 2025 Update
LBNL report, co-authored by Shehabi, updating national estimates and projections of US data center electricity use.
Arman ShehabiLawrence Berkeley National Laboratory
Energy Jun 2026 · The Innovator
Interview Of The Week: Siddharth Singh On AI's Energy Use
IEA analyst discusses how AI-driven data centre growth strains grids through clustered demand, connection backlogs and gas reliance, and argues for cleaner power alongside AI tools for energy optimisation.
Siddharth SinghInternational Energy Agency
Energy Jun 2026 · Lawrence Berkeley National Laboratory
United States Data Center Energy Usage Report: 2025 Update
Bottom-up modeling update projecting US data centers could use about 11.8% of national electricity by 2030, within a scenario range of 9.5% to 15.3%.
Arman ShehabiLawrence Berkeley National Laboratory
Energy Jun 2026 · Lawrence Berkeley National Laboratory
United States Data Center Energy Usage Report: 2025 Update
LBNL update estimating US data centers could use about 11.8% of national electricity by 2030 (649 TWh reference case), with scenarios spanning roughly 9.5% to 15.3%.
Arman ShehabiLawrence Berkeley National Laboratory
Energy May 13, 2026 · The Motley Fool
Oklo (OKLO) Q1 2026 Earnings Call Transcript
On the earnings call, DeWitte describes data center customer relationships including Meta and Switch, the planned Ohio nuclear campus, and nuclear as firm power for AI infrastructure.
Jacob DeWitteOklo
Energy May 6, 2026 · arXiv
OpenG2G: A Simulation Platform for AI Datacenter-Grid Runtime Coordination
Presents an open-source simulator for coordinating AI datacenter power flexibility with the electric grid, combining production-derived workload data, grid simulators and pluggable controllers.
Mosharaf ChowdhuryUniversity of Michigan
Energy May 5, 2026 · Sustainability by Numbers (Substack)
How much electricity does AI consume? [2025 summary]
Uses IEA data to estimate that data centres used about 1.5% of global electricity in 2025, with AI-focused facilities using about 0.5%, and looks at regional concentration and per-query energy.
Hannah RitchieOur World in Data
Energy May 4, 2026 · arXiv
From Barrier to Bridge: The Case for AI Data Center/Power Grid Co-Design
The paper argues that synchronized AI training loads break the grid's load-diversity assumption, so data center and power sectors should co-develop through joint planning, control and market innovation.
Noman BashirMIT
Energy Apr 16, 2026 · International Energy Agency
Key Questions on Energy and AI
IEA report examining data centre electricity growth, grid and supply chain bottlenecks, affordability and AI's energy-sector uses, projecting data centre demand roughly doubling by 2030.
Thomas SpencerInternational Energy Agency
Energy Apr 2026 · International Energy Agency
Key Questions on Energy and AI
An IEA special report updating its 2025 work, covering AI-driven power demand, gas, nuclear and batteries at data centers, grid impacts of AI clusters, and physical AI.
Thomas SpencerInternational Energy Agency
Energy Mar 25, 2026 · Uptime Institute Journal
As emissions soar, operators look to carbon capture
Gas-fired power for data center growth threatens net-zero goals. Dietrich suggests operators partner with developers on gas plants that use carbon capture and storage, despite its cost and site limits.
Jay DietrichUptime Institute
Energy Mar 23, 2026 · Koomey.com / GridLab
New report out today: Separating fact from fiction in data center electricity forecasts: A guide for regulators
A guide for regulators on judging data center electricity forecasts, pointing to uncertain demand and efficiency growth and to inflated or duplicate interconnection requests.
Jonathan KoomeyKoomey Analytics
Energy Mar 6, 2026 · University of Chicago News (Big Brains podcast)
Could data centers break our power grid? (Big Brains podcast)
In this podcast, Chien says AI data centers pose real grid risks if nothing is done, and argues that data centers should work with grids to lower emissions.
Andrew A. ChienUniversity of Chicago
Energy Dec 5, 2025 · Substack
Substack note on the study Flexible Data Centers: A Faster, More Affordable Path to Power
Post highlighting a PJM modeling study finding flexible data center interconnection with on-site capacity cuts system costs, keeps grid power available over 99% of hours, and speeds connections.
Tyler NorrisDuke University
Energy Nov 20, 2025 · Alphabet Investor Relations
Alphabet's Data Center Energy Strategy Call 2025
On this investor call, Peterson Corio describes Google's energy-first approach: it assesses power supply, reliability and impact before committing to new data center investments.
Amanda Peterson CorioGoogle
Energy Nov 12, 2025 · arXiv
Distribution and Management of Datacenter Load Decoupling
The paper studies how to distribute and manage energy resources that decouple datacenter power capacity from grid load, assessing grid carbon reduction, datacenter-grid cooperation and economic viability.
Andrew A. ChienUniversity of Chicago
AI footprint
AI footprint Oct 20, 2026 · Penn Engineering (Innovation & Impact podcast)
Powering the AI Revolution: The Future of Data Centers with Ben Lee
In a podcast episode, Lee discusses data center electricity and water needs and how chip design, architecture and operations could make AI more efficient and lower its emissions.
Benjamin C. LeeUniversity of Pennsylvania
AI footprint Oct 17, 2026 · Climate Hope Affiliates webinar series (UCC Environmental Justice Ministries)
The Environmental Footprint of AI
Virtual talk on the scale and growth of AI's electricity demand, water use and carbon emissions.
Alex de Vries-GaoVrije Universiteit Amsterdam / Digiconomist
AI footprint Aug 26, 2026 · asim.dev
Per token is coming
Post predicting flat-rate AI subscriptions will give way to per-token pricing, offering tips to use fewer tokens and arguing cost is the best proxy for environmental impact.
Asim HussainGreen Software Foundation
AI footprint Aug 14, 2026 · Mack Institute for Innovation Management, Wharton
Watch: AI Data Centers: Separating Fact from Fiction
A panel webinar featuring Lee looks at debates over AI data centers' electricity demand, water use and economic impact, separating speculation from what is still unknown.
Benjamin C. LeeUniversity of Pennsylvania
AI footprint Jul 20, 2026 · Our World in Data
How much energy do data centers and artificial intelligence use?
Estimates that data centers used about 1.5% of global electricity in 2025 and AI-focused ones about 0.5%, and stresses that demand is concentrated in a few local grids.
Hannah RitchieOur World in Data
AI footprint Jul 9, 2026 · Microsoft On the Issues
Responsibly building the AI future
Foreword to Microsoft's 2026 sustainability report, co-written with Brad Smith. It covers a 25% emissions rise driven by datacenter expansion, global water replenishment and server circularity.
Melanie NakagawaMicrosoft
AI footprint Jul 5, 2026 · arXiv
Lights, Camera, Carbon: Architectural Scaling Laws for Video Generation Energy Consumption
Introduces a framework that estimates energy use of text-to-video models from architecture and generation settings, reporting under 3% error across six open-source models.
Sasha LuccioniHugging Face
AI footprint Jul 2026 · ACM HotCarbon 2026 Workshop on Sustainable Computer Systems (via Amazon Science)
Amortizing AI training carbon footprint: Challenges, limitations, and a path forward
The paper looks at how to spread AI training emissions across inference requests. It finds that boundary, functional-unit and lifetime choices can shift per-request estimates roughly tenfold.
Noman BashirMIT
AI footprint Jul 2026 · Nature Reviews Clean Technology
Strategies and design for increasing AI sustainability
This co-authored review covers the energy, carbon and water impacts of AI data centers and surveys strategies to reduce them.
Shaolei RenUC Riverside
AI footprint Jul 2026 · ACM International Conference on Supercomputing (ICS 2026), Belfast
Scaling AI Computing Sustainably: A Journey Towards Sustainable AI (ICS 2026 invited keynote)
Invited keynote on how to scale AI computing sustainably, addressing the environmental footprint of AI systems.
Carole-Jean WuMeta (FAIR)
AI footprint Jul 2026 · ACM International Conference on Supercomputing (ICS)
Scaling AI Computing Sustainably: A Journey Towards Sustainable AI (Invited Keynote, ICS 2026)
Wu's talks page lists an invited keynote at ICS 2026 on scaling AI computing sustainably and the path toward sustainable AI.
Carole-Jean WuMeta (FAIR)
AI footprint Jun 20, 2026 · arXiv
The Language-Energy Divide: Measuring Energy Costs of Multilingual LLM Inference
Measures LLM inference energy across languages and finds large gaps from costlier scripts and longer outputs for low-resource languages. The highest-energy languages often had the lowest accuracy.
Mosharaf ChowdhuryUniversity of Michigan
AI footprint Jun 15, 2026 · The Microsoft Cloud Blog
Scaling AI with 8 to 20x energy efficiency
Co-authored post drawing on a Joule study, arguing that per-query AI energy at scale is lower than earlier estimates and that combined efficiency gains could reduce it 8 to 20 times.
Melanie NakagawaMicrosoft
AI footprint May 18, 2026 · davidmytton.blog
Energy use of AI inference – estimates and efficiency opportunities
Blog post estimating per-query energy for frontier-model inference, higher figures for long reasoning tasks, aggregate daily use at a billion queries, and how agentic workloads complicate estimates.
David MyttonArcjet / Console
AI footprint May 18, 2026 · davidmytton.blog
Energy use of AI inference – estimates and efficiency opportunities
Compares recent per-query estimates of AI inference energy, notes that disclosures from different companies are hard to compare, and discusses how reasoning and agentic workloads affect consumption.
David MyttonArcjet / Console
AI footprint May 13, 2026 · arXiv
Towards Resource-Efficient LLMs: End-to-End Energy Accounting of Distillation Pipelines
Presents a framework that tracks GPU energy at every stage of LLM distillation, including often-ignored teacher costs, and compares the energy and quality of distillation approaches.
Sasha LuccioniHugging Face
AI footprint May 6, 2026 · arXiv
From Cradle to Cloud: A Life Cycle Review of AI's Environmental Footprint
Reviews studies of AI's environmental impacts against an eight-stage life cycle framework, finding inconsistent scope and reliance on coarse CO2e proxies, and proposes more comparable measurement and reporting approaches.
Sasha LuccioniHugging Face
AI footprint May 1, 2026 · arXiv
The Hidden Cost of Thinking: Energy Use and Environmental Impact of LMs Beyond Pretraining
Measures energy, emissions and water across the full Olmo 3 development pipeline, finding reasoning post-training and development overhead dominate costs that typically go unreported.
Emma StrubellCarnegie Mellon University
AI footprint Apr 13, 2026 · Massachusetts Climate Action Network
The Carbon Cost of the Internet
A virtual talk on the carbon footprint of internet use and data centers, including AI's expected growth in energy demand and how climate advocates can help reduce it.
Eric MasanetUC Santa Barbara (Bren School)
AI footprint Mar 31, 2026 · Advanced Computing Community (online talk)
How to Give AI a Bad Carbon Footprint with David Patterson
Online talk offering tongue-in-cheek advice on worsening AI's carbon footprint. It addresses common fallacies about AI emissions, the main factors behind them, and ways to build more sustainable AI systems.
David PattersonGoogle
AI footprint Mar 5, 2026 · Energy vs Climate podcast
The Hidden Power Bill of Artificial Intelligence with MIT's Vijay Gadepally
Podcast interview in which Gadepally discusses the energy footprint of AI, including agentic AI, and efficiency measures that could substantially reduce AI energy use.
Vijay GadepallyMIT Lincoln Laboratory
AI footprint Mar 5, 2026 · Energy vs Climate podcast (Bespoke Podcasts)
The Hidden Power Bill of Artificial Intelligence with MIT's Vijay Gadepally
Gadepally explains why agentic AI and inference dominate AI energy use, says efficiency gains get little priority, and suggests flexible data center loads could help the grid.
Vijay GadepallyMIT Lincoln Laboratory
AI footprint Mar 5, 2026 · Energy vs Climate podcast
The Hidden Power Bill of Artificial Intelligence with MIT's Vijay Gadepally
Podcast guest Gadepally discusses AI's energy footprint, including higher energy use by agentic AI, whether efficiency gains keep pace with demand, and one change that could sharply cut consumption.
Vijay GadepallyMIT Lincoln Laboratory
AI footprint Mar 5, 2026 · Energy vs Climate (podcast)
The Hidden Power Bill of Artificial Intelligence with MIT's Vijay Gadepally
Gadepally talks about AI's energy footprint, including the higher energy use of agentic systems, inference's growing share, how hard energy is to measure, avoidable waste, and grid flexibility.
Vijay GadepallyMIT Lincoln Laboratory
AI footprint Jan 28, 2026 · KQED
The Real Cost of AI Slop (Close All Tabs)
In this podcast, O'Donnell and Casey Crownhart explain the energy and water costs of generative AI, including the cost of video generation, inference versus training, and companies not disclosing data.
James O'DonnellMIT Technology Review
AI footprint Oct 29, 2025 · MIT Technology Review
Exclusive eBook: The Math on AI's Energy Footprint
Subscriber eBook co-authored with Casey Crownhart. It covers AI energy use and emissions across model building, single queries, fuel sources and future demand, and argues that small per-query impacts add up.
James O'DonnellMIT Technology Review
Measurement
Measurement Oct 3, 2026 · arXiv
Estimating data center water use: Best practices and critical questions
Shows that study design alone can produce a roughly 24-fold range in data center water estimates. Proposes 14 best practices and uses them to score 31 existing studies.
Eric MasanetUC Santa Barbara (Bren School)
Measurement Sep 9, 2026 · Uptime Institute Intelligence
Coverage and quality concerns persist in EU's EED data
Uptime Intelligence analysis of continuing gaps in coverage and data quality in data center energy reporting under the EU Energy Efficiency Directive.
Jay DietrichUptime Institute
Measurement Aug 14, 2026 · Uptime Institute Intelligence
Data center operators gather sustainability data, but quietly
Analysis finding that operators still collect sustainability data during the AI build-out but seldom publish it, linked to unflattering results and politicization of sustainability.
Andy LawrenceUptime Institute
Measurement Jul 13, 2026 · arXiv
Energy Calculus: A Compositional Algebra of Energy in Computational Systems
Proposes a formal algebra that treats energy as a first-class primitive, composing measured energy signatures across computations with explicit propagation of measurement uncertainty.
Mosharaf ChowdhuryUniversity of Michigan
Measurement Jun 28, 2026 · ISCA 2026 Workshops/Tutorials
Full-Stack Carbon Accounting and Life Cycle Assessment for AI Systems (ISCA 2026 tutorial)
A hands-on tutorial, introduced by Gupta, on open-source tools for modeling embodied and operational carbon of AI systems, from chip supply chains to cloud-level attribution.
Udit GuptaCornell Tech
Measurement Jun 2026 · ACM e-Energy 2026
Go with the Flow: Analyzing the Carbon Footprint of Green Streaming
The paper analyzes video streaming emissions, arguing that green streaming must account for both server and network delivery paths across regional grid carbon differences, not just low-carbon data center regions.
Prashant ShenoyUMass Amherst
Measurement Jun 2026 · ACM e-Energy '26 (Banff)
CarbonX: An Open-Source Tool for Computational Decarbonization Using Time Series Foundation Models
The co-authored paper presents an open-source tool that uses time-series foundation models to forecast and fill in grid carbon intensity data for carbon-aware computing, using one model across many grids.
Prashant ShenoyUMass Amherst
Measurement Jun 2026 · ACM e-Energy '26
CarbonX: An Open-Source Tool for Computational Decarbonization Using Time Series Foundation Models
Presents an open-source tool that uses time series foundation models for computational decarbonization tasks such as forecasting grid carbon intensity.
Prashant ShenoyUMass Amherst
Measurement May 31, 2026 · arXiv
Position: Evaluation of ML Resource Utilization Requires Model Life Cycle Assessment
Position paper arguing that AI energy and environmental costs should be measured by life cycle assessment covering hardware manufacturing, training, experimentation and inference, not single runs.
Emma StrubellCarnegie Mellon University
Measurement May 31, 2026 · arXiv (ICML 2026 Position Paper Track)
Evaluation of ML Resource Utilization Requires Model Life Cycle Assessment
A position paper arguing that AI energy and environmental evaluation should use life cycle assessment covering embodied hardware and operational costs, not single training runs or inferences.
Emma StrubellCarnegie Mellon University
Measurement May 2026 · ACM/SPEC ICPE 2026 (UMass LASS)
CarbonShare: Carbon-Fair Allocation for Shared Clusters
Co-authored paper proposing carbon fairness policies that cap a shared cluster's total emissions and split that budget among applications, implemented in LXD and evaluated on industry workloads.
Prashant ShenoyUMass Amherst
Measurement Apr 6, 2026 · arXiv
The Energy Cost of Execution-Idle in GPU Clusters
Defines execution-idle, when GPUs draw high power during low activity, measures its share of time and energy in an academic AI cluster, and tests two mitigation prototypes.
Emma StrubellCarnegie Mellon University
Measurement Mar 17, 2026 · Green Software Foundation
After Five Years, I'm Stepping Down as Executive Director of the Green Software Foundation
Hussain announces his departure from the Green Software Foundation and reviews its work, including the SCI specification, SCI for AI, the Impact Framework and the Carbon Aware SDK.
Asim HussainGreen Software Foundation
Measurement Mar 15, 2026 · arXiv
DCGen 1.1 Technical Report: Generating Datacenter Configurations (including IT, Power, Cooling)
Describes DCGen, a tool that generates datacenter configurations covering IT, cooling and power at target scales. It supports research on datacenter design, operations and grid interaction scenarios.
Andrew A. ChienUniversity of Chicago
Measurement Feb 24, 2026 · U.S. House Committee on Science, Space, and Technology
Testimony of Dr. Eric Masanet before the House Committee on Science, Space, and Technology
Congressional testimony arguing that insufficient data prevents confident answers about the energy and environmental footprint of the U.S. AI data center buildout.
Eric MasanetUC Santa Barbara (Bren School)
Measurement Jan 29, 2026 · arXiv
Where Do the Joules Go? Diagnosing Inference Energy Consumption
A measurement study of 46 generative AI models on H100 and B200 GPUs that shows large energy differences by task and utilization, plus a framework explaining what drives them.
Mosharaf ChowdhuryUniversity of Michigan
Cooling
Cooling Oct 2026 · Open Compute Project, 2026 OCP Global Summit
Cooling the AI Revolution: Scaling Performance Through Advanced Liquid Cooling Technologies
Executive-session talk, co-presented with Auras Technology, on scaling AI system performance through advanced liquid cooling technologies.
Ali HeydariNVIDIA
Cooling Oct 2026 · Open Compute Project (2026 OCP Global Summit)
Cooling the AI Revolution: Scaling Performance Through Advanced Liquid Cooling Technologies
Executive session at the 2026 OCP Global Summit, co-presented with Auras Technology, on using advanced liquid cooling technologies to scale AI performance.
Ali HeydariNVIDIA
Cooling Sep 15, 2026 · Uptime Intelligence (Uptime Institute)
IT-agnostic DLC systems: key considerations
An overview of how to design, build and operate the secondary fluid networks used in water-based direct liquid cooling systems for data centers.
Daniel BizoUptime Institute (Uptime Intelligence)
Cooling Sep 14, 2026 · Uptime Institute (Uptime Intelligence)
IT-agnostic DLC systems: key considerations
Bizo reviews considerations for designing, building and running shared water-based direct liquid cooling loops that can support hardware from several vendors across overlapping IT refresh cycles.
Daniel BizoUptime Institute (Uptime Intelligence)
Cooling Jun 29, 2026 · Data Center Frontier
Data Center Insights: Joe Capes, Trane Technologies/LiquidStack
Capes argues that the hard part of high-density AI infrastructure is deployment: integrating power, cooling, controls and IT, and standardizing installation and operations across sites.
Joe CapesLiquidStack
Cooling May 30, 2026 · arXiv
Maximizing Compute Capacity in AI Data Centers through Cooling, Energy Storage, and Computing Adaptation
Proposes ComputeAmp, a framework that jointly manages cooling, battery storage and computing adaptation so power-limited AI data centers can extract more compute within power and water limits.
Shaolei RenUC Riverside
Cooling Mar 4, 2026 · Electronic Design
The Liquid-Cooled Plumbing Behind AI's Power Surge
Malouin's bylined article argues that direct-to-chip liquid cooling, modular power delivery and integrated liquid-cooled racks can raise AI compute density within existing data center footprints.
Bernie MalouinJetCool Technologies / Flex
Cooling Nov 2025 · Trane, Cool Air, Hot Takes podcast (Season 3, Episode 1)
Nvidia's Ali Heydari on the AI Revolution
Heydari explains why AI racks moved from air to liquid cooling, why NVIDIA chose a PG25 coolant, and the tradeoffs of higher loop temperatures, digital twins and heat recovery.
Ali HeydariNVIDIA
Siting
Siting Nov 4, 2026 · MIT Technology Review
Texas was the new epicenter of the AI data center buildout. That might be kaput.
Argues that Democratic wins in Texas on platforms critical of data centers could slow the state's large AI buildout. It notes that local governments and unions may resist sweeping restrictions.
James O'DonnellMIT Technology Review
Siting Oct 21, 2026 · CNBC
'MAKE DEALS': Trump says data centers should pay communities a 'dividend' to win support
In a CNBC interview in Berlin, Lochmiller described power availability as the main constraint on AI data centers and cited permitting and local community engagement as growing hurdles.
Chase LochmillerCrusoe
Siting Sep 3, 2026 · Latitude Media (Catalyst podcast)
What comes after the data center backlash?
Janous talks about local opposition to data centers over water, power prices and AI. He says developers should fund grid capacity, and he is doubtful about behind-the-meter generation.
Brian JanousCloverleaf Infrastructure
Siting Sep 1, 2026 · Andy Masley (Substack)
AI safety and the data center backlash
Argues that the public backlash against data centers does little for AI safety or harms it. He urges safety advocates not to spread misleading data center statistics that could erode trust.
Andy MasleyIndependent
Siting Aug 6, 2026 · Meta Newsroom
Why Meta Builds Its Own AI Data Centers
A video interview in which Meta's VP of Data Centers covers how AI data centers are cooled, how much water they use, and how Meta picks sites.
Rachel PetersonMeta
Siting Jun 22, 2026 · The Official Microsoft Blog
Powering the next wave of AI: Expanding capacity with our new datacenter in Pecos
Announces a roughly 2 GW Texas datacenter campus with a co-located natural gas plant at launch, closed-loop cooling that needs only an initial water fill, and use of nonpotable water.
Noelle WalshMicrosoft
Siting May 22, 2026 · Democracy Now!
"AI Resist List": Karen Hao on Data Center Resistance, Tech Billionaires, "Empire of AI" & More
Interview in which Hao discusses bipartisan community opposition to data centers, citing their power demand, freshwater use and effect on household utility bills.
Karen HaoIndependent
Water
Water Sep 22, 2026 · Semafor
'Hold us accountable' on data centers, Amazon sustainability chief says
Speaking at a Semafor event, Hurst said Amazon is reviewing its data center practices. She said it plans to publish more site-level metrics on energy and water effectiveness.
Kara HurstAmazon
Water Aug 27, 2026 · ThursdAI
Andy Masley on ThursdAI Aug 27, 2026 — Data-Center Backlash by the Numbers
Podcast appearance in which Masley goes through figures on data-center water and power use and explains how he found a large unit error in a water claim in Empire of AI.
Andy MasleyIndependent
Water Jul 2026 · Microsoft
2026 Environmental Sustainability Report: Responsibly building the AI future
Microsoft's annual report on fiscal-year 2025 covers renewable electricity matching, emissions trends, and circularity. It adds site-level datacenter water and electricity metrics. Nakagawa co-signed the foreword with Brad Smith.
Melanie NakagawaMicrosoft
Water Jul 2026 · arXiv (accepted at ACM e-Energy '26)
Balancing Bits and Drops: Stress-Adjusted Water Management for Data Centers
Proposes a water-stress-adjusted accounting framework for data center water use, including off-site electricity water, and evaluates water-carbon aware scheduling, rainwater harvesting and dry cooling.
Shaolei RenUC Riverside
Water Jun 25, 2026 · Trellis Impact 26
Can Water and Energy Both Win? Data Center Strategies Changing the Answer
Panel featuring Wellise on cutting data center water use without raising carbon emissions, covering cooling choices, site selection, and reuse of waste heat and water.
Christopher WelliseEquinix
Water Jun 22, 2026 · Fortune
Nvidia says its new data center design will fix AI’s water problem
Heydari is quoted on NVIDIA's closed-loop warm-water liquid cooling design, which the company says nearly eliminates on-site cooling water use by its newest AI server racks.
Ali HeydariNVIDIA
Water Jun 19, 2026 · arXiv
AI Data Centers and the Water Use Feedback Loop
Preprint looking at how AI data centres use water for cooling, how water scarcity limits where they can be sited, and how AI tools could make water systems more efficient.
Shaolei RenUC Riverside
Hardware
Hardware Jun 15, 2026 · arXiv
Architecture Carbon Tool v3: Enabling Sustainability-aware Silicon System Design Exploration
Whitepaper on ACT3, an updated platform for modeling carbon in silicon system design. It adds richer modeling, analysis telemetry and design-space exploration, shown through case studies.
Carole-Jean WuMeta (FAIR)
Hardware Jun 14, 2026 · arXiv / IEEE Micro
Google's Training Supercomputers from TPU v2 to Ironwood: Architectural Stability, Scale, Resilience, Power Efficiency, and Sustainability Across Five Generations
Co-authored paper reviewing five generations of Google TPU training supercomputers, reporting large gains in performance per watt and lower carbon per operation across generations.
David PattersonGoogle
Hardware Jun 14, 2026 · arXiv (slated for IEEE Micro, July/August 2026)
Google's Training Supercomputers from TPU v2 to Ironwood: Architectural Stability, Scale, Resilience, Power Efficiency, and Sustainability Across Five Generations
Co-authored review of five generations of Google TPU training supercomputers. It covers scaling, resilience, and gains in performance per watt and in carbon emissions per floating-point operation.
David PattersonGoogle
Hardware Feb 24, 2026 · Digiconomist
AI's Sustainability Blind Spot: Electronic Waste
Presents a supply-side estimate that AI servers could generate about 131 to 225 kilotons of e-waste annually by 2030, and calls for greater data center transparency.
Alex de Vries-GaoVrije Universiteit Amsterdam / Digiconomist
Hardware Jan 15, 2026 · arXiv (v2 revision)
CarbonClarity: Understanding and Addressing Uncertainty in Embodied Carbon for Sustainable Computing
Proposes a probabilistic framework that models embodied hardware carbon as distributions. It shows large uncertainty at advanced nodes and finds that chiplets and mature nodes reduce both carbon and uncertainty.
Udit GuptaCornell Tech
Hardware Nov 11, 2025 · Meta AI Research (NeurIPS 2025)
CATransformers: Carbon Aware Transformers Through Joint Model-Hardware Optimization
Presents a framework that co-optimizes Transformer models and hardware accelerators for both operational and embodied carbon. It reports total emissions reductions of up to 30% without losing accuracy or latency.
Carole-Jean WuMeta (FAIR)
Policy
Policy Nov 4, 2026 · MIT Technology Review
What's next for US energy policy after the midterm elections?
Says a divided federal government makes state and local energy action more important. It reports that voters backed restrictions in nearly all of 41 local ballot measures on data centers.
Casey CrownhartMIT Technology Review
Policy Oct 14, 2026 · International Energy Agency (IEA)
Powering Artificial Intelligence: Strategies for integrating data centres into electricity systems
Report requested by France's G7 Presidency. It reviews data centre policies in 34 jurisdictions and proposes principles for fitting data centres into electricity systems affordably, sustainably and securely.
Siddharth SinghInternational Energy Agency
Policy Mar 4, 2026 · Google (The Keyword blog)
Supporting the White House Ratepayer Protection Pledge: Google's approach for responsible energy growth
Announces that Google is signing the White House Ratepayer Protection Pledge and sets out how it will protect ratepayers and grid reliability as its data center energy use grows.
Amanda Peterson CorioGoogle
Policy Feb 10, 2026 · arXiv
The Global Landscape of Environmental AI Regulation: From the Cost of Reasoning to a Right to Green AI
Surveys environmental AI rules in eleven jurisdictions, finds they rarely require model-level inference energy disclosure, and proposes transparency mandates, user rights and international coordination.
Sasha LuccioniHugging Face
Heat reuse
Heat reuse Jun 28, 2026 · The Impact Equation (podcast)
Mark Bjornsgaard: £Billion plan to Heat Homes Using AI Data Centres
Podcast interview where the Deep Green founder describes capturing AI data centre waste heat to warm local infrastructure and homes, and how solving the economics enabled commercialization.
Mark BjornsgaardDeep Green
Heat reuse Apr 13, 2026 · The Techne Connect (podcast)
Mark Bjornsgaard, Founder & Chief Innovation Officer of Deep Green
Podcast interview on Deep Green's approach to reusing waste heat from data centres, the rising power demands driven by AI, and how distributed computing could change data centre design.
Mark BjornsgaardDeep Green
Inclusion is not endorsement — of us or by us. Spotted an error? Tell us via the bid form.