Dell'Oro Forecasts $120 Billion Data Centre Market by 2030 as AI Reshapes Infrastructure Economics

SINGAPORE — The data centre physical infrastructure market is on track to hit US$120 billion in annual revenue by 2030, growing at a 22 percent compound annual rate, according to a new forecast from Dell'Oro Group. The projection, published August 20, marks a fundamental shift: AI-driven construction is no longer a demand surge but the baseline against which the entire supply chain is measured.

"What has changed in this forecast is where the risk sits," said Alex Cordovil, research director at Dell'Oro. "Demand is no longer the open question. Delivery is."

The forecast arrives as the three largest hyperscalers — Amazon Web Services, Google Cloud, and Microsoft Azure — collectively plan nearly $600 billion in 2026 capital expenditure, the majority earmarked for AI infrastructure. AWS lifted its 2026 cash capex guidance to approximately $220 billion from $200 billion, citing memory costs. Google raised its range to $195–205 billion from $180–190 billion. Microsoft expects roughly $175 billion in calendar 2026 capex, with the next quarter alone exceeding $50 billion.

Yet even at those levels, capacity falls short. "We will still not have enough capacity to meet all the demand we have in 2026, and I believe this pattern will also be true in 2027 too," AWS chief Andy Jassy told analysts in July. Google CFO Anat Ashkenazi echoed the sentiment: "While we have increased our capacity quite significantly over the past three years, the demand still outpaces that investment."

Cooling and Power Become the New Bottlenecks

Dell'Oro identifies thermal management as the fastest-growing segment of data centre physical infrastructure through 2030, with liquid cooling leading the charge. Rack densification — driven by GPU clusters that pack more compute into less space — is pushing air-cooled designs past their limits. Water-cooled chillers are expected to scale faster than air-cooled units, and liquid cooling is moving from optional to mandatory for high-density AI deployments.

Power infrastructure is undergoing a parallel transformation. Within the uninterruptible power supply segment, higher-rating three-phase systems are outpacing smaller units as AI clusters push deployments up the capacity curve. Medium-voltage designs are gaining traction, connecting UPS systems closer to the grid. Dell'Oro projects solid-state transformers will meaningfully affect UPS demand from 2029, initially targeting large AI facilities that have largely moved away from UPS-based architectures.

Modern server rack with blue LED indicators in a data center

Neoclouds Emerge as a Distinct Buyer Category

The forecast separates AI-specialised cloud providers — neoclouds and model builders — into their own category, expected to become one of the fastest-growing market segments. CoreWeave, the largest neocloud, illustrates the trajectory. The company operated 49 data centres worldwide with more than 1 GW of active power and 3.5 GW of contracted power as of March 2026. In August it announced a 360 MW deployment across three facilities in Greater Jakarta, its first Asia-Pacific footprint, targeting 2028 availability.

"CoreWeave is signaling that it intends to compete globally against legacy hyperscalers, transitioning from a North American and European player into a global AI infrastructure platform," said Dave McCarthy, research vice president for cloud and edge infrastructure at IDC. The Indonesia move reflects a broader shift: as inference workloads grow and governments impose data residency rules, enterprises need AI compute closer to users and data. Cushman & Wakefield research shows Asia-Pacific's data centre development pipeline hit a record 26.5 GW in the first half of 2026, with Southeast Asia accounting for nearly half of under-construction capacity.

Networking Enters a Supercycle

Beyond power and cooling, the network fabric itself is being rearchitected. AI clusters increasingly require multi-layer fabrics — scale-up, scale-out, and scale-across — delivering 10 to 100 times more bandwidth with ultra-low latency and reliable congestion control. As agentic AI and always-on inference expand east-west traffic, the network becomes part of the computer. Major upgrades in switches, optics, topology, telemetry, and resilience are underway.

Cisco's fiscal 2026 results underscore the trend. The company recorded $9.3 billion in orders from hyperscalers for the full year, with Q4 alone contributing $4 billion — a 4.9x year-on-year jump. Optical transceivers at 400 Gb/sec and 800 Gb/sec are in short supply; Cisco shipped 100,000 400 Gb/sec units and 35,000 800 Gb/sec units in Q4. The 1.6 Tb/sec generation is ramping for late 2026 and early 2027. If multi-rail networking takes hold on AI systems, transceiver and cable spend will climb further.

Modern server room featuring network equipment with blue illumination

Supply Chain Constraints Shift from Chips to Copper and Construction

The bottleneck has migrated. Eighteen months ago, GPU availability was the binding constraint. Today, Dell'Oro highlights equipment lead times, construction labour, grid interconnection, and community consent. The first statewide data centre moratorium is now in force in the United States, and community opposition is blocking or delaying a meaningful share of announced projects. EMEA is the only region revised down from Dell'Oro's January forecast, reflecting slower power availability and a more difficult permitting environment.

Critical minerals, especially copper for power distribution, are emerging as a chokepoint. Ore grades are declining, refining capacity lags, and geopolitical risks are growing. While hyperscalers' willingness to pay secures short-term needs, it tightens the market overall. Long-term offtakes, recycling, and diversified domestic refining are becoming essential to future-proof builds.

North America is forecast to lead capacity additions through 2030, followed by China. Annual net capacity additions are expected to peak in year-over-year growth terms in 2026 before moderating steadily, though growth remains in double-digit territory throughout the period.

Hyperscalers Stay Dominant but Growth Mix Evolves

Hyperscalers will remain the largest single contributor to data centre physical infrastructure revenue by 2030, though their growth rate will slow compared with the 2025–2026 pace. Colocation providers continue to play a key role as hyperscalers use third-party facilities, particularly outside the United States. Powered-shell developments — where tenants procure their own equipment — are shifting revenue between customer segments without changing underlying building occupancy.

Colocation giants such as Equinix, Digital Realty, and regional players across APAC are seeing unprecedented demand for powered shell and build-to-suit arrangements. The model lets hyperscalers move faster while transferring procurement risk to tenants, who increasingly specify custom cooling and power configurations.

What Comes Next

The Dell'Oro forecast frames a market where the question is no longer whether AI infrastructure will be built, but how fast the physical world can accommodate it. Liquid cooling, medium-voltage power, 800 Gb/sec and 1.6 Tb/sec optics, and sovereign compute deployments are no longer edge cases — they are the new standard. For the supply chain, the challenge is delivery at scale. For enterprises, the challenge is accessing that capacity before competitors do.

"The AI buildout has moved past the point where it can be treated as a surge," Cordovil said. "It is now the baseline against which the rest of the market is measured."

Sources: Dell'Oro Group Data Center Physical Infrastructure forecast (August 2026), Data Center Knowledge reporting by Shane Snider and Nathan Eddy, The Register reporting by Dan Robinson, IDC commentary by Dave McCarthy, Cushman & Wakefield H1 2026 APAC data centre review, Cisco Systems Q4 FY2026 earnings presentation.

Internal links: Semiconductors, AI

Outbound links:

Keywords: Dell'Oro Group, data centre infrastructure, AI capex, liquid cooling, hyperscalers, CoreWeave, Cisco, optical transceivers, sovereign compute, power infrastructure