Green IT and sustainable data centers: where the efficiency and investment momentum is heading

Why It Matters

Global data center electricity consumption reached an estimated 510 TWh in 2025, roughly 2.2% of total worldwide electricity demand, and the International Energy Agency (IEA, 2025) projects this could double to over 1,000 TWh by 2030 as artificial intelligence workloads scale exponentially. Water withdrawals for cooling already rival those of some mid-sized cities; a single hyperscale facility can consume 5 million gallons per day during peak summer months (Uptime Institute, 2025). At the same time, corporate net-zero commitments and tightening regulations, from the EU Energy Efficiency Directive to the US SEC climate disclosure rules, are turning data center sustainability from a voluntary initiative into a compliance imperative. For sustainability professionals, understanding where efficiency gains, investment capital, and regulatory pressure are converging is essential to shaping credible digital infrastructure strategies.

Key Concepts

Power Usage Effectiveness (PUE) is the industry's primary efficiency metric, defined as total facility energy divided by IT equipment energy. The global average PUE has improved from 1.58 in 2020 to 1.45 in 2025 (Uptime Institute, 2025), but hyperscale operators like Google and Meta routinely achieve PUE values between 1.08 and 1.12. A PUE of 1.0 would mean zero energy overhead for cooling, lighting, and power distribution.

Water Usage Effectiveness (WUE) measures litres of water consumed per kilowatt-hour of IT energy. As evaporative cooling expands, WUE has become a contested metric. Microsoft (2024) reported a 34% increase in water consumption between 2021 and 2023, prompting a shift toward air-cooled and liquid-cooled designs in water-stressed regions.

Carbon-free energy (CFE) matching goes beyond purchasing annual renewable energy certificates (RECs) by targeting hour-by-hour matching of electricity consumption with carbon-free sources. Google pioneered this concept and achieved 64% average CFE matching across its global fleet by 2024 (Google, 2025). The 24/7 Carbon-Free Energy Compact, convened by the United Nations and Sustainable Energy for All, now has over 160 signatories.

Liquid cooling encompasses direct-to-chip and immersion cooling technologies that transfer heat away from processors using specialised fluids rather than air. AI accelerator chips like NVIDIA's B200 generate thermal design power loads exceeding 1,000 watts per chip, making traditional air cooling thermally and economically unviable at density (NVIDIA, 2025).

Embodied carbon refers to the greenhouse gas emissions from manufacturing, transporting, and installing data center equipment and building materials. While operational carbon has received most attention, embodied carbon can account for 20 to 40% of a data center's lifecycle emissions (IEA, 2025).

What's Working

Liquid cooling deployment is accelerating. The global liquid cooling market for data centers grew 45% year-over-year in 2025 and is projected to reach $15.6 billion by 2028 (Dell'Oro Group, 2025). Direct-to-chip cooling, where coolant circulates through cold plates attached to processors, has become the default for new AI training clusters. Equinix deployed direct liquid cooling across 18 facilities in 2025 and reported 30% reductions in cooling energy per rack compared to precision air cooling (Equinix, 2025). Immersion cooling, where entire servers are submerged in dielectric fluid, is gaining traction for edge deployments and high-density AI inference. GRC (Green Revolution Cooling) now operates immersion-cooled deployments in 14 countries.

Renewable and nuclear power procurement is expanding. Hyperscale operators signed over 15 GW of new renewable power purchase agreements (PPAs) in 2025 alone (BloombergNEF, 2026). Amazon Web Services became the world's largest corporate buyer of renewable energy, with over 30 GW of cumulative capacity. Nuclear energy is attracting fresh attention: Microsoft signed a 20-year agreement to restart Unit 1 of the Three Mile Island plant, now rebranded as the Crane Clean Energy Center, securing 835 MW of carbon-free baseload power (Microsoft, 2024). Google and Amazon have announced agreements with Kairos Power and X-energy, respectively, for small modular reactor (SMR) deployments targeting the early 2030s. Oracle disclosed plans to power a 1 GW data center campus with three SMRs (Oracle, 2025).

AI-driven facility optimisation is delivering measurable savings. Google's DeepMind-developed cooling optimisation system has been in production since 2016, but newer implementations using reinforcement learning achieved 12 to 15% reductions in total facility cooling energy across multiple operators in 2025 (Google DeepMind, 2025). Schneider Electric's EcoStruxure IT platform, deployed across over 10,000 sites, uses predictive analytics to reduce unplanned downtime and optimise energy distribution in real time.

Waste heat reuse is scaling in Northern Europe. Denmark, Sweden, and Finland have implemented regulatory frameworks that incentivise data centers to feed rejected heat into district heating networks. Meta's Odense data center in Denmark supplies waste heat to approximately 12,000 homes, displacing natural gas (Meta, 2025). Stockholm Data Parks, a collaboration between the City of Stockholm and operators, aims to supply 10% of the city's residential heating from data center waste heat by 2030.

What's Not Working

Water consumption is rising faster than efficiency gains. Despite WUE improvements at individual facilities, total industry water consumption is increasing as capacity scales. Microsoft disclosed that its global water use rose to 7.8 billion litres in 2024, up from 6.4 billion in 2022 (Microsoft, 2025). In water-stressed regions like Arizona, Northern Virginia, and parts of India, new data center permits face growing community opposition over aquifer depletion.

Grid constraints are delaying new capacity. In Ireland, EirGrid has paused new data center grid connections in the Dublin area, which already hosts over a quarter of the country's electricity demand from data centers (EirGrid, 2025). Northern Virginia, the world's largest data center market, faces interconnection queues of three to five years due to transmission bottlenecks. These delays push operators toward on-site generation, including diesel backup, which undermines decarbonisation goals.

Embodied carbon measurement remains immature. While operational carbon tracking is now standard, fewer than 15% of data center operators conduct full lifecycle carbon assessments that include embodied emissions from concrete, steel, servers, and networking equipment (Uptime Institute, 2025). Without standardised methodologies, comparisons across facilities are unreliable and Scope 3 emissions from the IT supply chain remain largely unquantified.

Greenwashing risks in REC procurement persist. Annual REC matching allows operators to claim 100% renewable energy while consuming grid power that may be predominantly fossil-fuelled on an hourly basis. A 2025 analysis by Ember found that for some European data center operators, actual hourly CFE match rates were as low as 38% despite annual REC claims of 100% (Ember, 2025). The gap between certificate-based claims and real emission reductions undermines credibility and has attracted regulatory scrutiny.

E-waste from server refresh cycles is mounting. The average server lifecycle has shortened from five years to three years as AI workloads demand faster hardware replacement. The Global E-waste Monitor (UNITAR, 2025) estimates that IT equipment contributes 10.5 million tonnes of e-waste annually, with recovery rates below 20% for critical minerals like cobalt and rare earth elements.

Key Players

Established Leaders

• Google — Pioneer of 24/7 CFE matching and AI-driven cooling optimisation; operating at 1.10 average PUE across its global fleet.
• Microsoft — Committed to water-positive operations by 2030; signed the largest nuclear PPA for a data center (835 MW from Crane Clean Energy Center).
• Equinix — World's largest colocation provider; deploying liquid cooling at scale and targeting climate-neutral operations by 2030.
• Schneider Electric — Leading provider of power management and data center infrastructure management (DCIM) software, deployed across over 10,000 sites.

Emerging Startups

• LiquidCool Solutions — Develops single-phase liquid immersion cooling systems for high-density IT workloads with zero water consumption.
• ZutaCore — Israeli startup offering two-phase direct-on-chip dielectric cooling that enables PUE values below 1.03 in pilot deployments.
• Lancium — Designs flexible, behind-the-meter data centers co-located with renewable generation, enabling curtailment-based load flexibility.

Key Investors/Funders

• Brookfield Asset Management — Invested over $30 billion in renewable energy assets, increasingly targeting data center power procurement partnerships.
• BlackRock — Launched dedicated infrastructure funds for digital infrastructure with sustainability criteria, deploying capital into green data center REITs.
• Breakthrough Energy Ventures — Backed multiple startups in advanced cooling, next-generation geothermal, and SMR technologies relevant to data center decarbonisation.

Real-World Examples

Google's 24/7 Carbon-Free Energy Programme. Google has systematically pursued hourly CFE matching since 2020. By 2024, its global fleet averaged 64% CFE on an hourly basis, with individual campuses in Denmark, Finland, and Oregon exceeding 90% (Google, 2025). The programme combines long-term PPAs, advanced energy storage, and grid-aware workload scheduling that shifts deferrable compute tasks (such as batch processing and model training) to hours when local grids are cleanest. Google has open-sourced its methodology and electricity maps data to encourage industry-wide adoption.

Microsoft's Crane Clean Energy Center Nuclear Agreement. In September 2024, Microsoft signed a 20-year PPA with Constellation Energy to restart Three Mile Island Unit 1, a 835 MW pressurised water reactor that had been shuttered in 2019 for economic reasons. The agreement provides Microsoft with carbon-free baseload power for its Azure cloud and AI workloads in the PJM Interconnection region, which still derives over 30% of its electricity from fossil fuels (Microsoft, 2024). The deal signalled to the market that hyperscalers are willing to underwrite nuclear restarts, catalysing similar announcements from Amazon and Oracle.

Meta's Odense Waste Heat Recovery. Meta's hyperscale data center in Odense, Denmark, captures waste heat from its server halls and feeds it into the municipal district heating system operated by Fjernvarme Fyn. The system delivers approximately 100,000 MWh of heat annually, enough to warm around 12,000 Danish homes and displacing roughly 15,000 tonnes of CO2 per year from natural gas boilers (Meta, 2025). The project was enabled by Denmark's regulatory requirement that large electricity consumers offer waste heat to local utilities at cost, creating a replicable policy model for other jurisdictions.

Equinix Liquid Cooling Rollout. Equinix deployed direct-to-chip liquid cooling across 18 International Business Exchange (IBX) data centers in 2025, targeting facilities hosting the densest AI training and inference workloads. Early results showed cooling energy reductions of 30% per rack compared to hot-aisle containment with precision air conditioning, and the company reported that liquid-cooled racks achieved power densities of 60 to 100 kW per rack versus 15 to 20 kW for air-cooled equivalents (Equinix, 2025). The rollout is part of Equinix's broader strategy to reach climate-neutral operations across its 260+ data centers by 2030.

Action Checklist

• Benchmark your PUE and WUE. Measure and publicly report facility-level PUE and WUE using the ISO/IEC 30134 standard; set annual improvement targets.
• Transition to hourly CFE matching. Move beyond annual REC procurement toward 24/7 carbon-free energy matching; use tools like Electricity Maps or Google's CFE methodology to track hourly performance.
• Evaluate liquid cooling for new builds. Assess direct-to-chip or immersion cooling for any facility expected to host AI accelerators or rack densities above 30 kW per rack.
• Quantify embodied carbon. Conduct lifecycle carbon assessments for new builds and major retrofits, covering concrete, steel, server hardware, and networking equipment.
• Engage on grid constraints proactively. Work with utilities and regulators to secure grid connections early, explore on-site renewables and battery storage, and consider flexible load participation in demand-response programmes.
• Explore waste heat reuse. Identify district heating networks, industrial processes, or agricultural operations near your facilities that could absorb rejected heat.
• Develop server circularity programmes. Extend hardware lifecycles through firmware updates and workload optimisation; establish certified refurbishment and recycling partnerships for decommissioned equipment.
• Prepare for regulatory compliance. Map exposure to the EU Energy Efficiency Directive (reporting obligations from 2024), SEC climate disclosures, and national data center energy regulations; assign compliance ownership.

FAQ

Will AI workloads make data centers unsustainable? AI is driving a step-change in electricity and cooling demand, but it is also enabling efficiency gains. AI-optimised cooling, workload scheduling, and predictive maintenance can offset a portion of the incremental energy use. The net outcome depends on whether the industry deploys new capacity with best-available efficiency technologies (liquid cooling, CFE procurement, waste heat reuse) or simply expands legacy air-cooled, fossil-powered infrastructure. The IEA (2025) projects that without aggressive efficiency measures, data center electricity demand could reach 1,000 TWh by 2030, but with best-practice deployment, growth could be contained to 650 to 700 TWh.

Is nuclear power a realistic option for data centers? Nuclear power provides carbon-free, 24/7 baseload electricity, making it uniquely suited to data center loads. The Microsoft-Constellation deal, Amazon's X-energy SMR agreement, and Oracle's 1 GW SMR campus plan demonstrate serious commercial intent. However, SMR technology is pre-commercial with first deployments expected no earlier than 2029 to 2031, and existing nuclear restarts face regulatory and community approval timelines of two to four years. Nuclear is likely to be part of a diversified energy portfolio rather than a standalone solution.

How should operators balance water and energy efficiency? Evaporative cooling is one of the most energy-efficient methods, but it consumes large volumes of water. In water-abundant regions (Nordics, Pacific Northwest), evaporative cooling can be appropriate. In water-stressed areas (US Southwest, Middle East, parts of India), operators should prioritise air-cooled or liquid-cooled designs that eliminate or drastically reduce water use. Some operators are adopting closed-loop liquid cooling systems that recirculate coolant without evaporative loss, achieving both low PUE and near-zero WUE.

What regulations should data center operators prepare for? The EU Energy Efficiency Directive (Article 12) requires data centers above 500 kW to report PUE, WUE, renewable energy share, and waste heat reuse by January 2025, with annual updates thereafter. The EU Corporate Sustainability Reporting Directive (CSRD) broadens Scope 3 disclosure requirements. In the US, the SEC climate disclosure rules require public companies to report material climate risks and Scope 1 and 2 emissions, with Scope 3 under safe-harbour provisions. Singapore has introduced a moratorium on new data centers unless they meet a PUE below 1.3 and commit to green energy procurement. These requirements are tightening globally and will likely expand to cover embodied carbon and e-waste within the next two to three years.

How can smaller operators compete on sustainability? Smaller operators and colocation customers can leverage shared infrastructure efficiencies, select providers with strong CFE records, and adopt modular liquid cooling solutions that are now available at rack scale. Joining industry initiatives like the Climate Neutral Data Centre Pact (Europe) or the iMasons Climate Accord provides access to benchmarking tools, procurement frameworks, and peer learning. Sustainability does not require hyperscale budgets; it requires disciplined measurement, transparent reporting, and strategic technology choices.

Sources

• International Energy Agency (IEA). (2025). Data Centres and Data Transmission Networks: Tracking Report 2025. Paris: IEA.
• Uptime Institute. (2025). Global Data Center Survey 2025: PUE, Water Use, and Sustainability Trends. New York: Uptime Institute.
• Dell'Oro Group. (2025). Data Center Physical Infrastructure Quarterly Report: Liquid Cooling Market Forecast. Redwood City, CA.
• BloombergNEF. (2026). Corporate Energy Market Outlook: Global PPA Volumes and Pricing, 2025 Year in Review. London: BNEF.
• Google. (2025). 2024 Environmental Report: 24/7 Carbon-Free Energy Progress and Methodology. Mountain View, CA.
• Microsoft. (2024). Crane Clean Energy Center Power Purchase Agreement Announcement. Redmond, WA.
• Microsoft. (2025). 2024 Environmental Sustainability Report: Water, Energy, and Carbon Data. Redmond, WA.
• Meta. (2025). Odense Data Center Waste Heat Partnership: Annual Impact Report. Menlo Park, CA.
• Equinix. (2025). Liquid Cooling Deployment Report: Energy and Density Performance Across 18 IBX Facilities. Redwood City, CA.
• NVIDIA. (2025). Blackwell Platform Data Sheet: Thermal Design Power and Cooling Requirements. Santa Clara, CA.
• Ember. (2025). Hourly Carbon-Free Energy Matching in European Data Centers: Certificate Claims vs. Reality. London: Ember.
• EirGrid. (2025). Shaping Our Electricity Future: Data Centre Grid Connection Policy Update. Dublin.
• UNITAR. (2025). Global E-Waste Monitor 2025: IT Equipment Waste Streams and Recovery Rates. Bonn: United Nations Institute for Training and Research.
• Oracle. (2025). Data Center Campus Design: Small Modular Reactor Power Strategy. Austin, TX.
• Google DeepMind. (2025). Reinforcement Learning for Data Center Cooling: 2025 Deployment Results. London.
• GRC (Green Revolution Cooling). (2025). Immersion Cooling Deployments: Global Footprint Update. Austin, TX.