As AI demand surges, performance-per-watt and rack-level density can prevent runaway infrastructure expansion and ease pressure on water and energy resources

The world is experiencing a challenging summer. This July, the average global temperature hit an all-time high. These record-setting heatwaves and sustained droughts are placing growing pressure on local water and energy supplies and are forcing governments to make tough decisions, including imposing new restrictions on data centers amid their expansion.

In recent assessments, the EU has called out data centers as part of the strain, particularly due to their growing demand for power and water used in cooling. This challenge also exists in the United States. The Federal Energy Regulatory Commission warns that the U.S. power grid could face significant stress this summer due to extreme heat and surging electricity demand, especially from the expansion of data centers. Regions like New England, the Midwest, the Mid-Atlantic, the Southwest and Texas are at heightened risk for power shortfalls during atypical weather events or supply disruptions.

With AI workloads exploding and extreme weather becoming increasingly common, the pressure on resources is only expected to rise. The question is no longer whether compute demand will grow — it’s how to support that growth without compounding environmental stress.

One of the clearest answers is to increase the amount of compute that can be delivered per data center.

If every watt of power and every rack of space can do more work, fewer data centers need to be built to meet demand. That translates directly into less land development, less power draw and less water used for cooling. The path to sustainable infrastructure isn’t just about sourcing cleaner energy or designing better cooling systems. It’s also about reducing the physical footprint of compute itself.

Right now, the industry is witnessing the opposite trend. The AI boom is pushing organizations toward larger clusters and deployments, often with highly specialized hardware that runs hot and draws massive power. If efficiency doesn’t keep pace, the only way to scale is to keep building, and the environmental costs of that are becoming clear.

This isn’t just a future problem. Around the world, permitting for new data centers is already being delayed or denied due to power constraints and community pushback. For example, Ireland’s grid operator has paused all new data center connections in the Dublin area until 2028 over concerns about electricity capacity, and local councils have rejected proposals citing insufficient energy infrastructure. In this context, infrastructure efficiency becomes not just a technical goal, but a gating factor for business and policy.

The industry needs to shift its focus toward maximizing real-world performance-per-watt, not just at the chip level, but at the rack-level and across entire systems. That kind of efficiency buys flexibility. It allows cloud and AI providers to scale within existing footprints, to deploy in constrained environments, and to avoid triggering new rounds of environmental and regulatory conflict.

In practical terms, this means rethinking compute architectures to maximize performance-per-watt through modern processor innovations. It also means rebalancing workloads and designing systems where throughput is matched by thermal and energy awareness. Efficiency can no longer be a side benefit. It has to be a primary goal.

The more work we can extract from every watt and square foot, the fewer data centers we need to build. The fewer data centers we build, the lower the water and energy burden we place on the regions we serve.

To support the next era of digital services, we need to shift from simply expanding infrastructure to also optimizing it, starting with the compute itself.