Editor's note:George Sakellaris is the founder and CEO of Ameresco.

As global leaders gather in Davos, the topic dominating the World Economic Forum is familiar: artificial intelligence is the next powerful engine of productivity, competitiveness, and economic growth. "Industry in the Intelligent Age" is one of this year's key themes.

However, we pay insufficient attention to the less glamorous reality behind it: the physical infrastructure that powers AI. We can build the best models in the world, write the best code, and design the fastest chips, but none of it matters if we cannot power the data centers. Today, the grid cannot expand fast enough to keep up with AI-driven demand growth, potentially affecting reliability and affordability.

It is projected that U.S. data center grid electricity demand will grow by 22% in 2025 and nearly triple by 2030, driven primarily by AI workloads, putting pressure on existing grid capacity. In fact, PJM Interconnection, the largest U.S. grid operator, is nearing its limits due to surging data center demand—generation capacity may peak under extreme weather, and consumer dissatisfaction over rising electricity prices is growing. The question is how we respond.

If we handle the AI boom the wrong way, grid stability and public trust will clash with tech-driven growth. If we handle it right, we can accelerate AI competitiveness while avoiding household electricity bills becoming collateral damage.

In the coming years, more large data center developers will reach the same conclusion: they cannot rely solely on the grid. They must bring their own power. AI workloads require enormous computing power, which means more servers, more cooling systems, and higher electricity consumption.

We are actually building a highly electricity-intensive new industry in real time, but the grid still operates on assumptions designed for a different era. Utilities and grid operators across the country are warning that demand is growing faster than expected. Transmission and interconnection queues are getting longer. Even when generation capacity is available, delivering power to the right place at the right time has become a bottleneck.

For most Americans, resilience is only noticed when it fails. But for hyperscale data centers, power outages mean lost revenue, customer disruption, equipment damage, and reputational risk. If you invest billions in a facility that must run 24/7, the answer is not to hope the grid catches up, but to design for resilience.

This leads to an uncomfortable truth: the power system cannot meet every new load on the timeline the AI economy demands. This is not about blaming anyone. Utilities are working hard, and regulators are moving faster than before. But even with the best intentions, grid expansion and modernization take time. Permitting, transmission planning, material constraints, and labor availability do not disappear because of urgency.

Meanwhile, placing huge new electricity loads on an already strained system endangers everyone. If large customers treat the grid as an unlimited and instant resource, ratepayers will bear the consequences. That is why the next phase of U.S. AI competitiveness must include a new operating model—treating the grid as a partner. Future data centers will no longer be powered like traditional buildings with a single power line, but more like self-contained energy ecosystems.

This model already exists in other sectors of the economy. Hospitals, military bases, emergency shelters, and critical government facilities do not assume external power is always available. They build redundancy, invest in backup generation, deploy microgrids, and plan for outages.

Data centers should be treated the same, as they increasingly perform functions vital to society and the economy. The difference now is scale. Some of these investments will be driven by economic factors, others by reliability requirements and customer expectations. In Washington, policymakers often talk about winning the AI race.

That goal requires realism and, more importantly, energy. Energy is foundational. If the U.S. cannot rapidly expand its power infrastructure, it risks turning the AI boom into a self-imposed constraint. Projects will be delayed, investment will flow elsewhere, costs will rise, and resilience will be undermined when it is needed most.

This means modernizing the grid, accelerating new generation and storage, reforming permitting and interconnection timelines, and recognizing that large energy customers need to invest directly in resilience.

Developers and operators are increasingly exploring behind-the-meter power solutions because they understand the stakes. The future requires a collaborative model: the grid provides connection and market coordination, while large customers bring distributed resilience and flexibility.

The grid cannot do it all alone. Therefore, the most reliable data centers of the future—those that protect uptime, control costs, and support national competitiveness—will not just plug in; they will bring their own power.