Artificial intelligence is transforming how we communicate, work, and solve problems. From large language models to real-time analytics, AI’s potential is vast. But the infrastructure powering this revolution comes at a growing environmental and operational cost. If not addressed, the unchecked growth of AI workloads could undermine both corporate climate commitments and the long-term viability of the data center industry.

A 2023 report by the International Energy Agency (IEA) projected that global data center electricity use could exceed 1,000 terawatt-hours (TWh) by 2026 – more than the total electricity consumption of Japan. Much of this growth is linked to AI. Training a single AI model like GPT-3 can consume 1,287 megawatt-hours, enough to power 120 US homes for an entire year. But power isn’t the only concern.

GettyImages-1315096305
– Getty

Beyond energy: Emerging sustainability and ethical challenges

AI-ready data centers face a broader sustainability and reputational challenge that spans multiple dimensions:

  • Water stress: Cooling systems for high-density AI clusters often require millions of gallons of water per year. In regions already facing drought, this amplifies community tensions and exposes operators to regulatory scrutiny.
  • Grid strain and local energy inflation: AI-driven campuses require huge bursts of power, often reaching 100MW+ per facility. In some cases, their energy demand exceeds local industrial zones. This can place strain on local grids and even cause price hikes for nearby residents and businesses.
  • eWaste and hardware churn: The accelerated development cycle of AI chips means infrastructure is replaced more often. According to the UN, less than 20 percent of global eWaste is formally recycled. That means millions of tons of obsolete servers, GPUs, and networking gear could end up in landfills.
  • Carbon offset dependence: While many tech giants aim for net zero, the widespread use of carbon offsets (rather than actual emissions reductions) is increasingly under fire. Critics argue that offsets can mask unsustainable practices and delay real decarbonization.
  • Land use and biodiversity: Hyperscale facilities often require vast tracts of land, potentially impacting local ecosystems and biodiversity. This is especially concerning as more data centers expand into rural areas or regions with sensitive environmental zones.
  • Social license to operate: In some communities, large data center projects have met resistance due to perceived inequities in resource consumption versus local benefits. Without transparency and community engagement, projects risk reputational damage and legal delays.

Case in point: Microsoft and OpenAI’s sustainable buildout

Microsoft, in partnership with OpenAI, is investing in AI infrastructure designed to address these concerns head-on. Its $10 billion expansion includes two 315MW data centers in Wisconsin, engineered for high efficiency and powered by renewable energy. Microsoft’s broader goal is to run entirely on carbon-free energy by 2025 and become carbon negative by 2030.

These facilities also integrate AI-powered cooling systems, advanced heat reuse technologies, and rigorous water usage effectiveness (WUE) targets – setting a benchmark for what responsible AI infrastructure can look like.

Four strategies to lead with intelligence and responsibility

  1. Power with purpose: Embrace renewable energy and prioritize energy efficiency. Google’s infrastructure today delivers 6× more compute per joule than a decade ago, thanks to custom hardware and sustainability-led design.
  2. Optimize the compute: Deploy AI-specific chips like Nvidia’s H100 and use software compression techniques like quantization and distillation to reduce energy-intensive operations. A smaller model with 90 percent of the performance at 30 percent of the compute is a win–win.
  3. Reimagine cooling: Hyperscale operators like Meta are deploying liquid cooling with PUE below 1.1, a marked improvement over the global average of 1.58. More efficient cooling reduces both energy and water waste.
  4. Design for circularity: Plan for modular upgrades, refurbish and reuse server components, and embed recycling into the supply chain. Design out waste from the beginning and extend hardware lifespans wherever possible.

Sustainability is smart business

AI’s environmental cost is no longer theoretical. It affects electricity markets, water tables, eWaste landfills, and public trust. But it also represents an opportunity for innovation and leadership. Companies that lead on sustainable AI infrastructure will benefit from operational savings, regulatory favorability, and brand trust.

The AI revolution is unstoppable – but its impact doesn’t have to be unsustainable. We have the data, tools, and real-world models to build a future where intelligence and environmental responsibility go hand in hand.