For years, data center design was driven by a straightforward goal: build enough capacity to meet demand. That approach made sense when workloads were more predictable, but AI is changing that equation.
Today’s infrastructure needs to support higher-density, dynamic workloads while making better use of power and resources. This shift is about more than efficiency for efficiency’s sake. Using infrastructure more effectively can help lower operating costs, improve resilience, and make it easier to adapt as workloads evolve.
With more than 1,500 data centers in development in the United States, operators have an opportunity to build these advantages into infrastructure from day one rather than trying to retrofit them later. The next generation of AI infrastructure is therefore less about building with speed or relying on any single technology, and more about taking a systems-level approach that considers power, cooling, compute, and operations together.
Why Efficiency Is Becoming a Design Priority
Growing workloads and shifting infrastructure requirements make data centers more complex to design and operate. As AI drives higher-density workloads, every infrastructure decision can have implications elsewhere in the system.
A change to the power architecture, for example, can affect cooling requirements. Cooling decisions can influence energy and water consumption, while compute utilization can change power and cooling needs. These interdependencies mean that efficiency cannot be addressed by a single team or discipline in isolation.
Teams must coordinate decisions and consider the upstream and downstream consequences of their choices. This helps operators balance the need for new capacity against how effectively they use existing infrastructure.
This has direct implications for both cost and resilience. Using resources more effectively can lower operating costs, while designing systems to work together reduces inefficiencies and potential failure points. Addressing these considerations during design can also reduce reliance on costly retrofits later.
A Systems-Level Approach to Efficiency
Power usage effectiveness (PUE) has long been a standard measure of data center efficiency, but it doesn’t capture every dimension of modern infrastructure. The industry’s expectations for what constitutes an efficient facility have changed. Average PUE has improved from 2.50 in 2007 to 1.52 in 2026, while newer facilities routinely achieve PUEs of 1.3 or lower. A level of efficiency that would’ve been considered strong two decades ago is now well above what’s expected of modern facilities.
This progress illustrates why operators must look beyond a single measure. A lower PUE is important, but AI infrastructure also needs to account for water, carbon, compute utilization, energy reuse, and interaction with the grid.
A broader framework can consider:
-
Power usage effectiveness: Energy efficiency across operations.
-
Water usage effectiveness (WUE): Water consumption tied to operations and cooling.
-
Carbon usage effectiveness (CUE): Carbon impact of IT energy consumption.
-
Energy reuse effectiveness (ERE): How effectively a facility reuses the energy it generates/consumes.
-
Compute power efficiency (CPE): IT equipment utilization relative to energy overhead.
-
Grid-aware operations (GAE): How workloads shift or reduce power consumption in response to grid conditions.
Together, these metrics give operators a fuller view of the trade-offs between capacity, demand, and utilization.
Designing Power, Cooling, and Compute Around Real Workloads
No single system determines infrastructure efficiency. Power, cooling, and compute all must work together, particularly as high-density AI workloads place new demands on data center environments.
That starts with designing infrastructure around how workloads behave day to day. From there, integrated power and cooling approaches can improve deployment reliability and make scaling more efficient, while modular approaches give operators the flexibility to adapt as requirements change.
As more AI infrastructure is deployed in the United States, this approach also supports a more responsive infrastructure supply chain and closer collaboration between technology providers. Growing investment in US-based manufacturing is bringing engineering, integration, and production capabilities closer together, helping the industry respond more quickly as infrastructure needs evolve.
Measuring Efficiency Through the Infrastructure Lifecycle
Efficiency is not a one-time design decision. It needs to be evaluated throughout development, deployment, and ongoing operations.
A balanced set of metrics shows operators how infrastructure is performing and where there’s room to improve. In turn, tracking energy, water, carbon, energy reuse, compute utilization, and grid interaction provides a fuller picture than any single measure.
As AI workloads and infrastructure requirements evolve, ongoing measurement provides operators with the visibility needed to adjust and maintain performance over time.
Building for What Comes Next
AI infrastructure will continue to scale, but capacity alone will not be enough. Operators need to design with efficiency, reliability, and adaptability in mind from the beginning.
By taking a systems-level approach across power, cooling, compute, and operations – and coordinating decisions across disciplines – operators can make better use of available capacity, manage operating costs, and build more resilient infrastructure.
Ultimately, successful AI infrastructure will be measured not simply by how much capacity it can deliver, but by how efficiently and reliably it can support evolving workloads.