Table of Contents
Dell’Oro’s research director Alex Cordovil said that neoclouds help diversify GPU demand away from a handful of hyperscalers
In sum – what to know:
Rack densities accelerating toward 600 kW require higher-voltage power designs – Operators must adopt new electrical architectures, tools and training to manage efficiency, safety, and costs.
North America leads expansion thanks to AI labs and power availability – Policy support, secondary-market capacity, and on-site generation options are sustaining growth, with Europe, Asia, and the Middle East starting to catch up.
Neocloud providers will drive demand but face consolidation – Diversified GPU access, enterprise workloads, and colocation partnerships will fuel growth, though only a handful are likely to emerge as long-term leaders.
As AI workloads push computing requirements to unprecedented levels, data center power architecture is becoming the industry’s next major disruption. In an interview with RCR Wireless News, Dell’Oro research director Alex Cordovil explained how rising rack densities, evolving regional dynamics and the growth of Neocloud providers will shape global demand through 2026.
Cordovil said that liquid cooling dominated headlines over the past two years because it represented a major technological break. But the next shift may be even more consequential: power distribution. Densities approaching 600 kW per rack are forcing the industry to rethink electrical systems from the ground up, the analyst said.
“If we go to very higher power, we cannot operate at lower voltages unless we compromise on the amount of material,” he explained. Lower voltages at these loads also mean higher currents, driving up losses and reducing efficiency. As a result, data centers will increasingly rely on high-voltage distribution brought “closer to the rack,” whether through AC or DC architectures.
This shift will change equipment layouts and introduce new operational risks. “As we increase voltages, it means it increases danger,” Cordovil said. Operators will need new training, protective equipment and procedures to safely manage higher-voltage environments. As with liquid cooling, widespread adoption will follow necessity. “These workloads… can only be liquid cool because there’s no way around it,” he said. “Higher voltages are going to be on a very similar path.”
Regional demand patterns are also evolving. North America grew 23% in Q1, driven by hyperscalers and the location of leading AI labs. “All the major AI labs are based in the US,” Cordovil said, and secondary markets such as Wyoming, Louisiana, and Tennessee still offer available energy capacity. Government support is another accelerant. “There’s only been doubling down on the strategic importance of AI,” he said, noting efforts to ease permitting and invest in power solutions, including SMRs and on-site gas turbines.
Growth is broadening outside North America as well. Europe is seeing new momentum around sovereign AI, supported by EU funding and national-level interventions to accelerate data center approvals. The Middle East, particularly the UAE and Saudi Arabia, recognized AI sovereignty early and continues to invest aggressively. Meanwhile, China’s domestic chip ecosystem is strengthening, and once new AI processors reach maturity, “we expect the Chinese growth to really be unleashed again,” Cordovil noted.
A final force reshaping the market is the rise of neocloud providers such as Scaleway and Nebius. These companies help diversify GPU demand away from a handful of hyperscalers. Cordovil referenced Nvidia’s goal of reducing dependency, noting that “three customers made more than 50% of their data center revenue.” Neoclouds also serve enterprises seeking flexible GPU access without massive upfront investment.
However, their rapid proliferation won’t last. “We expect to have a few winners … and a lot of losers in the market,” he said. Most neoclouds rely on colocation providers rather than building their own data centers, allowing them to focus on algorithms, models, and services.
