# AI Infrastructure Pushes Data Center Capex Forecast Above $3 Trillion > Source: > Published: 2026-08-21 09:00:00+00:00 # AI Infrastructure Pushes Data Center Capex Forecast Above $3 Trillion AI infrastructure spending is forecast to push global data center capex beyond $3 trillion by 2030, driven by hyperscalers and AI-specialized clouds. Worldwide data center capital spending is projected to surpass $3 trillion by 2030 as hyperscalers, sovereign AI programs, and specialized cloud providers expand infrastructure to support growing AI workloads, according to a new report from Dell’Oro Group. The research firm [said](https://www.delloro.com/news/ai-buildout-maintains-momentum-as-data-center-capex-surpasses-3-trillion-by-2030/) its 2030 capex outlook has nearly doubled since its January 2026 forecast, reflecting higher hyperscaler spending guidance, increased estimates for global data center power capacity, and rising commodity costs. “AI accelerators are expected to account for about a third of the $3 trillion in data center capex,” Baron Fung, vice president at Dell’Oro Group, told Data Center Knowledge. “AI infrastructure is a major driver.” However, the cost of accelerators alone does not reflect the full expense of AI infrastructure. Operators also need servers to host the chips, specialized networking for AI clusters, and storage to support training and inference. Fung said the forecast assumes global data center power availability will increase to more than 200 GW. Dell’Oro estimates the four largest US cloud providers could account for about half of global data center capex. Its AI-specialized cloud category, which includes model developers and [neocloud providers](/cloud/neocloud-storm-gathers-as-data-center-deals-stall-over-credit-risk), is projected to grow at a compound annual rate of nearly 60%, while general-purpose server demand is expected to rise as inference, agentic AI and storage workloads expand. ## Hyperscaler Scale Could Tighten Supply The largest cloud providers are using their purchasing power to secure preferred pricing and capacity commitments through long-term supplier agreements. Fung said those arrangements could reduce component availability for other buyers, extending lead times and increasing prices. Their growing use of custom chips and architectures also lowers costs at scale while putting pressure on server manufacturers’ pricing. Enterprises may initially favor rented GPU capacity because it avoids major upfront investment while utilization and returns remain uncertain. “Ultimately, the scale and economics of the largest cloud providers could further favor cloud-based infrastructure, encouraging more enterprises to shift workloads to the cloud,” Fung explained. He said he expects many enterprises to take a hybrid approach – stable, heavily utilized AI workloads could eventually move on-premises when ownership becomes cheaper, while variable or incremental demand remains in the cloud. ## AI Spending Extends Beyond Compute Higher-density AI systems require changes throughout the physical data center, not only additional accelerators, Gordon Johnson, senior CFD manager at Subzero Engineering, a data center solutions company, told Data Center Knowledge. “AI workloads require more electrical power per rack than traditional computing, often requiring a combination of cooling strategies,” Johnson said. Accelerator deployments, therefore, drive related investment in power distribution, liquid cooling, containment and supporting architecture. Johnson suggested operators add capacity in stages, identify where high-density computing is required, and integrate new cooling methods with existing infrastructure. “The biggest risk is building too much too quickly, so operators must understand where high-density compute is required,” he said. ## Power Availability to Shape Expansion Power availability is becoming the biggest constraint on AI infrastructure expansion, according to Luke Edney, a partner at Norton Rose Fulbright. Modern data centers can require hundreds of megawatts, with some campuses now being designed at the gigawatt scale. “Power is also changing the economics of projects – the cost and timeline risk associated with power procurement is now a critical factor in site-selection decisions,” Edney told Data Center Knowledge. Grid connection timelines in established markets may take more than five years, pushing developers toward locations with surplus renewable generation, flexible connection regimes and utilities willing to form proactive partnerships. Demand for reliable, low-carbon electricity is also accelerating interest in on-site generation, fuel cells, and [small modular reactors](/energy-power-supply/nuclear-powered-data-centers-when-will-smrs-finally-take-off-). Edney said enterprises must balance the risk of falling behind against the financial exposure created by committing too early. “AI investment still needs to be tied to measurable business outcomes, not simply a response to market pressure,” Edney said.