The global data centre physical infrastructure (DCPI) market is expected to grow steadily between 2025 to 2030, surpassing $80 billion, according to a new report by Dell’Oro Group.

Although demand signals remain “exceptionally strong”, the report suggested that power scarcity has become an increasingly “binding constraint” on near-term expansion. This comes as on-site generation evolve from a contingency option to a practical necessity for large AI campuses.

“The DCPI market is undergoing a period of profound structural transformation,” said Alex Cordovil, research director at Dell’Oro Group. “Technology disruption is cascading across the stack as rising chip thermal design power and the approach of the one-megawatt rack force fundamental changes in facility design.”

Despite power constraints, the report suggested operators are deploying creative mitigation strategies to bring capacity online, which Cordovil said supports Dell’Oro Group’s “decision to revise the DCPI growth outlook upward for the 2025-29 period”.

The forecast report also indicates that thermal management is projected to grow at a 20% CAGR by the end of the decade, suggesting that cooling is expected to evolve from a supporting function to a “primary determinant of data centre design”.

Indeed, cooling has been touted as absolutely critical to enabling AI as data centres are seeking infrastructure that is high-performing, scalable and efficient. In fact, the GPUs that power AI Factories generate 20 to 50 times more heat than traditional CPUs (Central Processing Units).

Speaking with Capacity at the start of the year, Motivair by Schneider Electric CEO Rich Whitmore said: “The reality is, if you don’t have the best cooling, the chips don’t perform where they actually need to be.

“The current conversation of power and cooling is absolutely critical to enable AI infrastructure worldwide.”

The Dell’Oro report also found that direct liquid cooling (DLC) is expected to surge to $8 billion by 2030, as it transitions from an enabler to a foundational technology for AI Factories. This comes amid the launch of Nvidia’s Vera Rubin compute tray designs that could eliminate fans entirely and enable fully liquid-cooled configurations.

Nvidia CEO Jensen Huang made comments at CES 2026 about the company’s Rubin chips, which can be cooled using water at temperatures that don’t require traditional water chillers. His remarks caused traditional cooling companies like Trane Technologies and Johnson Controls to see their stock dramatically fall, as fears were sparked over demand for their products this year.

As global demand for AI computing capacity continues to put pressure on cooling supply chains and power infrastructure, Dell’Oro also found that power availability has become a “binding constraint”.

This is as grid interconnection delays and transmission bottlenecks are expected to drive rising demand for natural gas turbines, as operators deploy on-site generation to bring large gigawatt-scale campuses online.

“Emerging DC power distribution architectures are expected to influence deployment choices toward the end of the decade,” the report said. “Service providers – including cloud, colocation and neocloud operators – are forecast to grow DCPI spending at roughly five times faster than enterprise, reflecting the increasing complexity and capital intensity of AI-ready infrastructure.”

Related stories

Cisco unveils powerful chip to scale up gigawatt AI data centres

Google enlists TotalEnergies for solar power at Texas data centre

HAPPENING NOW: Middle East’s AI and data centre surge spotlighted at Datacloud market address