AI’s Power Problem: Data Centers Face Energy Squeeze

Artificial intelligence is driving a new buildout of data centers. The growth is fast. The power demand is rising with it. Utilities, chipmakers, and policymakers are now racing to keep pace. The stakes are high for climate goals and for the availability of electricity in busy grids.

Behind the boom: why AI needs so much power

AI models are growing larger. They run more often. Training new systems takes vast computing power. Running them for everyday tasks, known as inference, is the bigger and ongoing load. Each prompt, image, or video processed in the cloud uses energy. At scale, that adds up.

The International Energy Agency (IEA) warns the surge is significant. In its 2024 outlook, the agency said electricity consumption from data centres, AI and cryptocurrencies “could double by 2026”. That forecast includes both new AI training clusters and the rapid spread of AI features in search, productivity tools, and media apps.

Hardware is part of the story. Advanced accelerators pack more transistors into each chip. They deliver major performance gains, but they also draw more power per rack. Dense liquid cooling is becoming standard in new halls. That allows higher compute density, yet it increases site-level energy complexity and water management needs.

Grid constraints move to the foreground

Electric utilities and grid operators are adjusting plans. Interconnection queues are long in several regions. New large-load customers must wait for transmission upgrades. In fast-growing metros, power for new campuses is tight. This is not only an AI issue, but AI is a prominent driver in 2024 and beyond.

Project timelines are stretching. Developers are looking farther from city centers. Some are clustering near existing substations or generation. Others explore colocating near renewable plants or adopting on-site generation and storage. Reliability remains a key factor. AI clusters need high uptime and stable voltage. That can require substation upgrades and new lines, which take years to permit and build.

Companies pledge cleaner compute

Technology firms say the AI boom must be paired with cleaner energy. Many have long-term targets on climate and water. Google has stated a goal to “operate on 24/7 carbon-free energy by 2030”. Microsoft has pledged to be “carbon negative by 2030” and “water positive by 2030”. Those objectives push companies to sign new clean power contracts, invest in storage, and shift workloads to match renewable output.

Corporate buyers are also evolving contracts. They want power in specific locations and hours, not just annual credits. That supports the buildout of firm, clean capacity such as advanced storage and geothermal. It also pushes grid transparency, so companies can measure when energy used is actually carbon-free.

Efficiency race: do more with fewer watts

Efficiency is the other lever. Chipmakers are improving performance per watt. Data centers adopt advanced cooling. Model developers are cutting computational cost. Techniques like quantization, pruning, and mixture-of-experts architectures reduce energy per inference. Software stacks are being tuned to run models at lower precision, such as 8-bit or 4-bit, without major quality loss.

Scheduling matters too. Some providers shift non-urgent training to off-peak hours or regions with spare capacity. Others cache results or use smaller models when appropriate. These steps are invisible to users but can lower energy use and bills while maintaining performance.

Policy and transparency catch up

Regulators are seeking more data from the sector. In Europe, the revised Energy Efficiency Directive introduces reporting duties for large data centers, including energy and water metrics. The goal is clearer baselines and better benchmarking across facilities. More jurisdictions are discussing disclosure rules, building codes for high-density cooling, and incentives for heat reuse.

National energy agencies are also updating demand forecasts. They are factoring in AI clusters, electrification of transport and buildings, and industrial loads like chip fabs. The mix varies by country. Where grids are already constrained, permitting and planning reform are on the agenda. The aim is to connect new clean generation and wires faster, without compromising community input or environmental safeguards.

Environmental concerns and equity questions

Environmental groups support transparency and efficiency, but they warn that net impacts depend on location and timing. If new load lands in regions with carbon-intensive generation, emissions can rise before new clean plants arrive. Water use is a related concern. Evaporative cooling cuts power, but it consumes water. In drought-prone areas, that raises local risks. Communities near proposed sites want clear assessments and shared benefits, such as infrastructure upgrades or heat recovery for district systems.

Supporters of AI argue the technology can deliver climate benefits. It can help operators forecast wind and solar, optimize industrial processes, and cut waste in logistics. They say better algorithms and hardware will narrow the energy gap over time. Both views can be true. Outcomes will hinge on choices made in siting, procurement, and design.

What to watch next

• Grid buildout speed: Progress on transmission lines, substations, and interconnection reforms will shape where and how fast AI capacity can grow.
• Clean power procurement: More granular, hourly matching deals and long-term contracts for firm, clean resources could become the norm for large buyers.
• Hardware roadmaps: New chips and cooling systems that boost performance per watt will influence the total footprint of AI services.
• Model efficiency: Wider use of smaller, specialized models and low-precision inference could cut routine energy use at scale.
• Water strategies: Shifts toward closed-loop or non-potable cooling, and siting in cooler climates, can ease local water stress.
• Heat reuse: Capturing server waste heat for buildings or industry could improve overall energy efficiency where infrastructure allows.

The bottom line

AI is a powerful new strain on digital infrastructure. The industry’s response—more clean energy, greater efficiency, and smarter operations—will determine the environmental cost of the boom. Policymakers are pushing for transparency and faster grid upgrades. Companies are making bolder procurement and engineering bets. The IEA’s warning that demand “could double by 2026” captures the urgency. The next two years will show whether the sector can scale without overrunning climate goals and local grids.

The choices are concrete: where to build, what to buy, and how to run it. The outcome will be measured in megawatts, emissions, and the reliability users expect. AI’s promise is large. So is the power bill.