
The Power Grid Is Redrawing the Cloud Map
- Cloud networking
- November 16, 2026
- RSS Feed
By Todd Wenzel, Solutions Architect
AI is driving data center power demand into new territory, changing where infrastructure gets built and how businesses connect to it.
Table of Contents
- In summary
- Why is power such a constraint for AI infrastructure?
- How is power demand changing data center site selection?
- Are established data center hubs running out of capacity?
- Where will the next generation of data centers be built?
- What does a more distributed cloud map mean for enterprises?
- Why does connectivity matter when compute moves?
- How should businesses prepare for a more distributed infrastructure landscape?
In summary
Rising AI power demand is making electricity availability a deciding factor in data center site selection. Developers are looking beyond established hubs to regions with available grid capacity, reliable energy supplies, suitable cooling conditions, and room to expand. As compute becomes more distributed, organizations need flexible connectivity that can link new locations to their clouds, data, applications, and users.
Not long ago, data center site selection was still a real estate conversation, with the network as an afterthought. Developers followed the pull of established cloud regions and fiber routes, then worked through the obstacles of turning a chosen site into live infrastructure. Electricity mattered, of course, but it was usually treated as a checkbox in the plan.
Now, AI has moved it to the front of the conversation.
The question now comes earlier and lands harder: Can the local grid deliver enough power, and on a timeline the business can work with? A site can look viable on paper — with the right location, a plausible permitting path, and proximity to major fiber. But if the electricity won’t arrive for years, the project (and the cloud growth behind it) will quickly stall.
That constraint has already begun to redraw the cloud map, and compute is moving outward from established hubs and into locations that once sat on the edge. Here, we revisit one of our key predictions for how the data center has evolved in 2026: Digital infrastructure is being shaped not only by where data needs to go, but by where electricity can actually be delivered.
Why is power such a constraint for AI infrastructure?
Because accelerated computing concentrates far more processing into each rack than conventional enterprise IT. Graphics processing units (GPUs) and other accelerators chew through electricity when training models and running inference, while the surrounding cooling and power systems must work harder to support their density.
The International Energy Agency (IEA) estimates that data centers consumed about 415 terawatt-hours (TWh) of electricity in 2024. It expects that figure to reach around 945 TWh by 2030, with AI as the most important driver of the increase. Electricity use by accelerated servers is also projected to grow by 30% a year over the same period.
Most data center energy consumption goes directly into IT. According to the IEA, servers account for around 60% of electricity use in a modern data center, while cooling can range from about 7% in efficient hyperscale facilities to more than 30% in less-efficient enterprise sites. Higher rack density can therefore raise both direct data center power consumption and the energy needed to remove the resulting heat.
This is why AI data center power consumption cannot be reduced to the wattage of a GPU. The facility around the hardware is just as important.
How is power demand changing data center site selection?
Rising data center energy demand is making power availability and connection time key factors in site selection. A suitable plot is of limited value if the local grid cannot energize the facility at the required scale for several years.
Developers now need to consider a range of criteria before committing to a site:
- Available capacity: The local grid must support the initial load and provide a credible path for expansion.
- Connection timeline: Grid studies and transmission upgrades determine when a facility can begin operating; lead times for critical equipment can add further delays.
- Cooling conditions: Local climate and water availability affect which cooling designs are viable, along with the facility’s total energy demand.
- Network access: Diverse fiber routes and interconnection options determine whether the new site can reach cloud environments and data stores, thus serving users effectively.
The last point is easy to underestimate. A region may offer abundant energy and affordable land, but it still needs the connectivity to function as part of a wider infrastructure environment. Power makes a site possible; the network makes it useful.
Are established data center hubs running out of capacity?
Established hubs are not simply running out of capacity, but many are struggling to add it at the pace AI demand requires. Crowded connection queues and grid constraints are slowing new projects. Long equipment lead times add another source of delay, pushing some developments farther from traditional clusters.
The IEA estimates that around 20% of planned data center projects could face delays unless grid risks are addressed. New transmission lines can take four to eight years to build in advanced economies, while data centers themselves may become operational in two to three years.
Concentration of facilities only adds to the pressure. Half of the data centers under development in the U.S. are located in existing large clusters, according to the IEA. CBRE also reports that limited power availability is inhibiting growth in some core markets, with operators in Europe adding supply farther from city centers and development in Asia Pacific shifting toward secondary locations.
The established hubs will remain important because they already have rich connectivity ecosystems and proximity to major cloud regions. But their borders are stretching. In some cases, the next viable data center may be tens of miles from the location an organization would once have chosen by default.
Where will the next generation of data centers be built?
The next generation of data centers will be spread across expanded core hubs and emerging regional markets. Locations with grid capacity, reliable energy, suitable cooling conditions, and strong fiber access are likely to attract more development.
This does not point to one new global capital for AI infrastructure. The geography will vary by region and workload. CBRE has identified growing opportunities in markets including Richmond, Santiago, Mumbai, and Johor as power constraints affect established hubs. Other locations will emerge as energy projects and transmission upgrades come online, supported by new network routes.
Workload types will also influence the map. Some model training jobs can run in regions with greater power availability because they are less sensitive to their distance from end users. Inference, on the other hand, often benefits from being closer to the people or systems requesting a response. Organizations may therefore use a combination of large regional facilities and smaller distributed sites rather than concentrate every AI workload in one place.
What does a more distributed cloud map mean for enterprises?
A more distributed cloud map gives enterprises greater choice, but it also makes infrastructure planning more complex. The best location for compute may no longer match the organization’s primary cloud environment or data store. It may also sit farther from users.
That can affect application latency, data transfer costs, regulatory requirements, and resilience design. An AI workload placed where power is available may need to retrieve data from another region before using services in a public cloud. Its results may then travel to users somewhere else. Every extra leg becomes part of the workload’s performance and cost profile.
Infrastructure teams will need to evaluate the whole path rather than the facility alone.
Why does connectivity matter when compute moves?
Connectivity matters because moving compute does not automatically move the clouds or data that feed it. Applications and users may remain somewhere else too. When power availability pushes infrastructure into new regions, the network has to bridge the distance without turning it into a bottleneck.
AI makes that especially important. Training and inference can combine large data sets with distributed storage, while supporting services run across several environments. Poorly planned routes can introduce latency or unpredictable performance. They can also increase exposure to data transfer charges and leave a workload dependent on a single path – a resilience disaster waiting to happen.
Flexible private connectivity gives organizations more freedom to place compute where power and capacity are available while keeping it connected to the rest of their environment. Data Center Interconnection can link infrastructure across facilities, while Global WAN solutions can connect distributed physical and virtual points of presence.
For AI environments, AI Exchanges provide a way to connect data centers and clouds with storage services or GPU as a Service providers across a private network. Infrastructure teams can’t make distance disappear, but the next best thing is to gain more control over how that distance is crossed.
How should businesses prepare for a more distributed infrastructure landscape?
Businesses should prepare by treating site and power choices as inseparable from network architecture. Committing to compute capacity first and working out the network later can leave an organization with infrastructure that is difficult or expensive to use.
A stronger planning process includes the following steps:
- Classify workloads by location sensitivity: Separate jobs that can run where power is available from those that need to stay close to data or users.
- Add energy questions to due diligence: Ask providers about available capacity, expansion plans, grid connection timelines, and cooling constraints.
- Design connectivity before deployment: Model traffic paths, bandwidth requirements, latency, redundancy, and data transfer costs alongside the site decision.
- Keep the architecture portable: Avoid unnecessary dependence on one facility or region so workloads can move as capacity and business needs change.
The cloud was never weightless, and now AI is making its physical constraints even harder to ignore. As electricity availability redraws where infrastructure can be built, organizations that can connect across that changing map will have more options for where their workloads run next.
To see how companies are successfully tackling this shift, Michael Reid spoke with Ben Baldieri, Founder of The GPU, to explore how Europe’s power constraints, fragmented regulation, and limited grid capacity are forcing a new generation of AI infrastructure strategies. Join the conversation on the Uplink Podcast.
Explore Megaport Enabled Locations






