Data centers are often described as the unseen backbone of our digital economy – and increasingly, of the AI revolution. McKinsey estimates that by 2030, companies will invest almost $7 trillion in capital expenditures on data center infrastructure globally. It sees 70% of that demand will come from hyperscalers, a trend already evident in the recent announcements of tech giants partnering with various U.S. states to invest billions in constructing large-scale data center sites.
While data centers are catalysts for unprecedented economic opportunity, they are simultaneously exerting mounting strain on energy systems already in the throes of transition. Per World Resources Institute figures, North American data center power demand is surging due to AI expansion, with Goldman Sachs projecting U.S. consumption to double from 31 GW in 2025 to 66 GW by 2027. By 2030, these facilities could account for 6% to 12% of total U.S. electricity demand, drastically straining the continental power grid.
To date, much of the sustainability dialogue has focused on the responsibilities of data center operators – from improving hardware efficiency and cooling innovations, to increasing renewable energy sources. But there is another critical lever: how data center enterprise customers design the systems within these environments.
By shifting away from legacy, always-on models to real-time, reactive architectures, enterprises can dramatically reduce compute and data movement demands, easing pressure on energy infrastructure while still fueling AI-driven growth.
Ultimately, unlocking the full potential of North America’s data center boom will depend not only on scaling clean energy but also on enabling customers to adopt smarter, more efficient digital foundations that align growth with sustainability at the application and architecture level.
From ‘Always On’ to ‘Always Available’: Shedding Architectural Dead Weight
In the AI economy, speed is strategic, but the traditional “request-response” model is failing the sustainability test. In legacy environments, systems continuously poll one another for updates, staying “awake” simply to wait for a request.
The result? A staggering waste: per the U.S. Department of Energy, 60% to 80% of server power is typically wasted during idle periods. This is the architectural equivalent of leaving a car engine running for hours, just in case it’s needed for driving.
As AI workloads intensify, these inefficiencies create a “distributed monolith” – a tightly coupled system that appears modular but remains functionally inseparable. When one system in the chain experiences latency, the entire resource stack remains engaged, burning compute cycles and electricity while waiting for a handshake.
This “always-on” dependency drives a costly cycle of over-provisioning for customers. To avoid crashes during peak demand, organizations are forced to “overbuild”, sizing their infrastructure for the busiest hour of the year. The consequence is massive amounts of hardware powered on but underutilized, leading to a drain on both capital and the environment – one that tightening sustainability mandates can no longer accommodate.
Smarter Flow, Leaner Compute
Breaking this cycle of waste requires an architectural pivot, away from rigid, overprovisioned systems toward a more responsive, event-driven foundation that optimizes how data moves. At the core of this shift is the event mesh – a dynamic network of interconnected event brokers that allows data to flow continuously across systems and in real-time, without the overhead of legacy polling. Instead of systems constantly requesting updates, information is shared as it changes, delivered instantly to wherever it’s needed.
This means digital infrastructure components like applications, services, edge devices and AI agents publish events as their state changes – whether a workload completes, a threshold is breached, or a failure occurs. These events trigger immediate, decoupled responses across systems, enabling real-time orchestration without tight dependencies.
This transforms how compute resources are utilized in two critical ways:
Reduces the idle polling: Systems are triggered only when a relevant event occurs – such as an AI inference request or operational signal – allowing unused resources to scale down or be redeployed instead of sitting idle.
Decouples system speeds: Event brokers absorb spikes in demand, enabling backend systems to process workloads at a steady, optimized pace, rather than maintaining excess capacity for short-lived peaks.
For AI workloads, which are notoriously spiky and unpredictable, this precision ensures that processing power is applied precisely when and where it matters, reducing both infrastructure strain and overall energy consumption.
Orchestrating Intelligence Efficiently
As data centers evolve to support increasingly complex AI workloads, the challenge extends beyond moving data efficiently to coordinating how that data is acted upon. This is where an agent mesh comes into play.
Building on an event-driven foundation, an agent mesh introduces a distributed layer of intelligence that operates with shared, real-time context to automate AI agent workflows and maintain situational awareness across data center operations.
By ensuring events are delivered reliably and in the correct sequence, customers can avoid misaligned decisions, reduce redundant processing, and maintain stable, energy-efficient operations.
This shared intelligence allows them to optimize how AI resources are used, rather than simply scaling infrastructure by throwing more GPUs at a problem. It’s the difference between a hundred people shouting in a room and a well-coordinated team sharing a single, real-time script.
In essence, event mesh delivers the real-time flow of information to drive operational efficiency, while the agent mesh determines how best to act on it with full, real-time context. Together, they support enterprise customers with a more coordinated, adaptive system that improves utilization while minimizing wasted energy, supporting both performance and sustainability objectives.
A Priority for North America’s Digital Future
Efficient resource management is no longer a “nice-to-have” – it is quickly becoming a license to operate. center
The future of data centers will not be defined solely by infrastructure or what happens within the server room. Instead, their enterprise customers will increasingly be measured by the efficiency of their own IT systems, particularly on the resilience, precision, and intelligence of the logistics systems that support everything beyond it.
We must move away from the brittle, energy-wasting “plumbing” of the past and toward a digital nervous system that is responsive, distributed, and resilient.
With the right approach, North America has the opportunity to set a global benchmark for how efficiently customers use data centers that are not only high-performing but also efficient, resilient, and built for a more sustainable future.
