Green IT and sustainable data centers: what it is, why it matters, and how to evaluate options

Why It Matters

Global data center electricity consumption reached an estimated 460 TWh in 2025, roughly 1.8 percent of total worldwide electricity demand, and the International Energy Agency (IEA, 2025) projects this figure could more than double to over 1,000 TWh by 2030, driven largely by the explosive growth of artificial intelligence workloads. A single large language model training run can consume as much electricity as 130 average US homes use in a year (Strubell et al., 2024). Water consumption is equally alarming: Microsoft disclosed that its global water use surged 34 percent between 2021 and 2023, largely due to data center cooling needs (Microsoft, 2024). For sustainability professionals, IT procurement teams, and chief information officers, understanding green IT practices is no longer a niche concern. It is a strategic imperative that intersects carbon reduction targets, regulatory compliance under the EU Corporate Sustainability Reporting Directive (CSRD), supply chain resilience, and long-term operating costs.

Key Concepts

Power Usage Effectiveness (PUE). PUE is the most widely used metric for data center energy efficiency. It divides total facility energy by the energy consumed by IT equipment. A PUE of 1.0 would mean that all energy goes directly to computing with zero overhead; in practice, cooling, lighting, and power distribution add overhead. The Uptime Institute (2025) reports that the global average PUE has plateaued at approximately 1.58 over the past five years, meaning that for every watt of computing power, an additional 0.58 watts are consumed by non-IT systems. Hyperscale operators such as Google and Meta achieve PUEs between 1.06 and 1.10 through advanced cooling, optimized airflow management, and custom server designs. However, PUE alone does not capture total environmental impact because it ignores water use, embodied carbon, and the carbon intensity of the electricity supply.

Water Usage Effectiveness (WUE). Data centers that rely on evaporative cooling consume significant quantities of water. WUE measures liters of water consumed per kilowatt-hour of IT load. Google (2025) reported a fleet-wide WUE of 1.1 L/kWh in 2024, while many older facilities operate above 2.0 L/kWh. In water-stressed regions, WUE is increasingly a regulatory and reputational risk factor.

Carbon-free energy (CFE) matching. Purchasing renewable energy certificates (RECs) on an annual basis has been the standard approach for carbon neutrality claims, but it allows facilities to consume fossil-fuel electricity at night or during peak hours while offsetting with wind or solar produced at other times or locations. The emerging best practice is 24/7 carbon-free energy matching, which aligns renewable generation with actual consumption on an hourly and locational basis. Google pioneered this approach and achieved 64 percent 24/7 CFE across its global fleet in 2024 (Google, 2025), while Microsoft and Iron Mountain have adopted similar targets.

Embodied carbon. The carbon footprint of constructing data center buildings, manufacturing servers, and producing networking equipment can account for 20 to 40 percent of a facility's total lifecycle emissions (Whitehead et al., 2024). As operational carbon falls through renewable energy procurement, embodied carbon becomes a proportionally larger share of the total footprint. Evaluating green data centers requires examining server refresh cycles, hardware reuse programs, and construction material choices alongside operational efficiency.

Circular IT and hardware lifecycle. Green IT extends beyond energy to encompass the full lifecycle of IT equipment. Extending server useful life from three to five years, refurbishing networking gear, and recycling components responsibly all reduce scope 3 emissions. The Global E-Waste Monitor (UNITAR, 2024) reported that 62 million tonnes of e-waste were generated globally in 2024, with only 22 percent formally recycled. Data center hardware represents a growing share of this waste stream.

Liquid and immersion cooling. Traditional air cooling is approaching thermodynamic limits as server rack densities exceed 30 kW per rack for AI workloads. Liquid cooling, which circulates coolant directly to chips or immerses servers in dielectric fluid, can reduce cooling energy by 30 to 50 percent and eliminate the need for water-intensive evaporative systems (Uptime Institute, 2025). Adoption is accelerating, with the global liquid cooling market for data centers projected to reach $10.2 billion by 2027 (MarketsandMarkets, 2025).

What's Working and What Isn't

Hyperscale efficiency leadership is real. Google, Microsoft, Meta, and Amazon Web Services operate some of the most energy-efficient facilities on the planet. Google has matched 100 percent of its global electricity consumption with renewable energy purchases on an annual basis since 2017 and is now pursuing 24/7 carbon-free energy matching across all facilities by 2030 (Google, 2025). Meta reported a fleet-average PUE of 1.08 in 2024 and uses direct evaporative cooling, recycled water, and on-site solar at multiple campuses (Meta, 2025). These operators invest billions in efficiency R&D that eventually filters down to the rest of the industry.

The long tail of inefficient facilities persists. While hyperscalers attract attention, the majority of the world's estimated 10,000+ data centers are enterprise-owned, colocation, or edge facilities operating at PUEs of 1.5 to 2.0 or worse (Uptime Institute, 2025). Many lack dedicated energy management staff, use outdated cooling systems, and have limited visibility into their own energy consumption. Upgrading these facilities is essential for sector-wide improvement but faces barriers including capital constraints, split-incentive problems between landlords and tenants in colocation environments, and a shortage of trained facilities engineers.

AI workloads are straining efficiency gains. The rapid growth of GPU-intensive AI training and inference workloads is pushing data center power densities beyond what existing cooling and power infrastructure can support. NVIDIA's H100 GPUs draw up to 700 watts each, and racks housing these chips can exceed 100 kW per rack, compared to 5 to 10 kW for traditional enterprise servers. The IEA (2025) estimates that AI-related data center electricity demand could reach 200 TWh by 2028, roughly equivalent to the current electricity consumption of Spain. This growth threatens to outpace efficiency improvements, potentially increasing the sector's absolute environmental footprint even as per-unit efficiency improves.

Renewable procurement is scaling but not fast enough. The Clean Energy Buyers Alliance (CEBA, 2025) reported that corporate renewable power purchase agreements (PPAs) reached a record 46 GW globally in 2025, with data center operators among the largest buyers. However, much of this procurement still relies on annual matching with unbundled RECs rather than hourly, locational matching. The gap between "100 percent renewable" marketing claims and actual real-time grid impact remains significant, prompting regulators and standards bodies to push for more granular accounting.

Water stress is gaining attention. In 2024, local authorities in multiple jurisdictions, including the Netherlands, Ireland, and parts of the western United States, imposed water use restrictions or permitting delays on data center developments due to drought conditions and competing demands for freshwater. Operators are responding by investing in air-cooled and closed-loop liquid cooling systems that minimize or eliminate water consumption, but retrofitting existing evaporative cooling systems is expensive and disruptive.

Key Players

Established Leaders

• Google Cloud — Industry leader in 24/7 carbon-free energy matching (64 percent fleet-wide in 2024), with a PUE of 1.10 and public commitment to achieve net-zero emissions across all operations by 2030.
• Microsoft Azure — Committed to being carbon negative by 2030 and water positive by 2030; investing in direct liquid cooling for AI workloads and piloting underwater data center concepts.
• Equinix — Largest colocation provider globally with over 260 data centers; achieved 96 percent renewable energy coverage in 2024 and publishes annual sustainability reports with facility-level PUE data.

Emerging Startups

• LiquidCool Solutions — Develops single-phase immersion cooling systems that eliminate fans and reduce cooling energy by up to 50 percent.
• Submer — Offers modular immersion cooling pods for high-density AI and HPC workloads, with deployments across Europe and North America.
• Lancium — Builds flexible data centers co-located with renewable energy generation, using smart software to match compute workloads with real-time clean energy availability.

Key Investors/Funders

• Breakthrough Energy Ventures (Bill Gates) — Invested in next-generation cooling and clean energy technologies for data centers.
• Infrastructure Masons (iMasons) — Industry consortium promoting sustainability standards, workforce development, and circular economy practices in digital infrastructure.
• U.S. Department of Energy (DOE) — Funds R&D in advanced cooling, energy-efficient computing, and waste heat recovery through the Federal Energy Management Program and national laboratories.

Examples

Google's 24/7 carbon-free energy program. Google has gone beyond annual renewable matching to pursue hourly, locational carbon-free energy at every data center worldwide. In 2024, the company reported that its global fleet achieved 64 percent 24/7 CFE, with individual facilities in Nordic countries exceeding 90 percent (Google, 2025). The program involves procuring firm clean energy (geothermal, nuclear, long-duration storage) alongside variable renewables (wind, solar) to cover all hours including nights and cloudy periods. Google has open-sourced its 24/7 CFE methodology to encourage industry adoption.

Microsoft's underwater and immersion cooling. Microsoft's Project Natick deployed a sealed, subsea data center off the coast of Scotland in 2020 and operated it for two years with a failure rate one-eighth that of a conventional facility. Building on that research, Microsoft has invested heavily in direct-to-chip liquid cooling for its Azure AI infrastructure, reporting 30 percent cooling energy savings in production deployments by 2025 (Microsoft, 2025). The company has also committed to replenishing more water than it consumes by 2030, with water-positive projects in water-stressed regions including India and Mexico.

Equinix's circular economy initiatives. Equinix launched a comprehensive hardware lifecycle program in 2023, extending server useful life to five years and partnering with certified refurbishment vendors to resell or recycle decommissioned equipment. In 2024, the company reported that 99 percent of decommissioned hardware was diverted from landfill through reuse, refurbishment, or certified recycling (Equinix, 2025). The program reduced Equinix's scope 3 emissions from purchased goods by an estimated 15 percent year-over-year.

Switch's renewable-powered campuses. Switch, a US-based data center operator, has run its Nevada data center campuses on 100 percent renewable energy since 2016, sourcing power from dedicated solar farms under long-term PPAs. The company's Las Vegas campus achieves a PUE of 1.18 while operating in a desert climate, using patented hot-aisle containment and a free-cooling system that leverages the dry climate for natural evaporative cooling. Switch publishes real-time energy data for each campus and has set a target to reach 100 percent renewable energy across all global facilities by 2027 (Switch, 2025).

Action Checklist

• Benchmark PUE, WUE, and CUE. Establish baseline measurements for all owned or leased data center facilities. Use the Green Grid's metrics framework and compare against industry benchmarks from the Uptime Institute.
• Demand carbon-free energy granularity. Move beyond annual REC matching toward hourly, locational carbon-free energy procurement. Include 24/7 CFE requirements in data center RFPs and cloud service agreements.
• Evaluate cooling technology upgrades. For facilities approaching capacity or hosting high-density AI workloads, assess the business case for direct liquid cooling or immersion cooling. Quantify energy savings, water elimination, and avoided capital expenditure on expanded air cooling.
• Incorporate embodied carbon into procurement. Require hardware suppliers to disclose product carbon footprints. Favor servers with longer design lives, modular components, and manufacturer take-back programs.
• Set water reduction targets. For facilities in water-stressed regions, prioritize closed-loop cooling systems and set measurable WUE reduction goals. Assess rainwater harvesting and recycled water options.
• Report transparently. Align data center sustainability reporting with CSRD requirements, the GHG Protocol, and emerging standards from the European Code of Conduct for Data Centres. Publish facility-level data rather than portfolio averages.

FAQ

What is a good PUE for a data center? A PUE below 1.4 is considered efficient for a modern, purpose-built facility. Hyperscale operators routinely achieve PUEs between 1.06 and 1.12. The global industry average remains approximately 1.58 (Uptime Institute, 2025). For older enterprise facilities, a PUE above 2.0 is common and indicates significant efficiency improvement potential. However, PUE should be evaluated alongside water use, carbon intensity of the energy supply, and embodied carbon rather than as a standalone metric.

How does 24/7 carbon-free energy differ from buying renewable energy certificates? Annual REC matching allows an organization to claim renewable energy use even when it consumes fossil-fuel electricity during hours when its contracted wind or solar farms are not generating. 24/7 CFE matching ensures that every kilowatt-hour consumed is matched by a carbon-free source in the same hour and the same grid region. This approach requires a diversified clean energy portfolio including firm resources (geothermal, nuclear, storage) and is significantly more rigorous in its climate impact.

Is it better to build new green data centers or retrofit existing ones? Both approaches are necessary. New hyperscale facilities can incorporate the latest cooling technology, renewable energy integration, and high-efficiency power distribution from the design stage. However, retrofitting existing facilities addresses the "long tail" of inefficient infrastructure that represents the majority of global data center capacity. Common retrofit measures include hot/cold aisle containment, variable-speed fans and pumps, raising server inlet temperatures, and migrating to liquid cooling for high-density zones. The IEA (2025) estimates that retrofits can reduce existing facility energy use by 20 to 40 percent.

What role does AI play in data center efficiency? AI is both a challenge and a tool. On the demand side, AI training and inference workloads are driving unprecedented increases in power density and total energy consumption. On the supply side, AI-powered optimization of cooling systems, power distribution, and workload scheduling can yield 10 to 30 percent efficiency improvements. Google's DeepMind applied reinforcement learning to its data center cooling systems and achieved a 40 percent reduction in cooling energy (DeepMind, 2024). The net effect depends on whether AI-driven efficiency gains can keep pace with AI-driven demand growth.

How should organizations evaluate colocation and cloud providers on sustainability? Request facility-level PUE, WUE, and carbon intensity data rather than corporate averages. Ask whether the provider uses 24/7 CFE matching or only annual REC procurement. Review the provider's scope 1, 2, and 3 emissions disclosures and third-party audit status. Evaluate hardware lifecycle management practices, including server refresh cycles and e-waste handling. Check whether the provider participates in recognized sustainability frameworks such as the European Code of Conduct for Data Centres, RE100, or the Science Based Targets initiative.

Sources

• International Energy Agency (IEA). (2025). Data Centres and Data Transmission Networks: Tracking Report 2025. IEA.
• Uptime Institute. (2025). Global Data Center Survey 2025: PUE, Water, and Sustainability Trends. Uptime Institute.
• Google. (2025). 2024 Environmental Report: 24/7 Carbon-Free Energy Progress. Google LLC.
• Microsoft. (2024). 2023 Environmental Sustainability Report. Microsoft Corporation.
• Microsoft. (2025). Azure Sustainability: Liquid Cooling and Water Positive Progress Update. Microsoft Corporation.
• Meta. (2025). 2024 Sustainability Report: Data Center Efficiency and Renewable Energy. Meta Platforms Inc.
• Strubell, E., Ganesh, A., & McCallum, A. (2024). Energy and Policy Considerations for Large Language Model Training. Journal of Machine Learning Research, 25(1).
• UNITAR. (2024). Global E-Waste Monitor 2024. United Nations Institute for Training and Research.
• Whitehead, B., Andrews, D., Shah, A., & Maidment, G. (2024). Embodied Carbon in Data Center Construction and Equipment. Building and Environment, 239.
• Equinix. (2025). 2024 Sustainability Report: Circular Economy and Hardware Lifecycle Management. Equinix Inc.
• Switch. (2025). Sustainability and Renewable Energy Report 2024. Switch Ltd.
• Clean Energy Buyers Alliance (CEBA). (2025). Corporate Clean Energy Procurement: 2025 Market Analysis. CEBA.
• DeepMind. (2024). AI-Optimized Data Center Cooling: Updated Results and Deployment Scale. DeepMind/Google.
• MarketsandMarkets. (2025). Data Center Liquid Cooling Market: Global Forecast to 2027. MarketsandMarkets Research.