Dell’Oro Group forecasts worldwide data-center capital spending could reach $1.7 trillion by 2030, with spending approaching $1 trillion in 2026. That is a forecast—not a committed industry budget—and its meaning depends on what is counted as “data-center capex.” Dell’Oro’s model includes data-center technology such as servers, storage and networking. Other analysts count mainly buildings, power and cooling, producing figures that are not directly additive.
What Dell’Oro’s $1.7 trillion forecast actually says
In a February 11, 2026 release, Dell’Oro said the multi-year artificial-intelligence expansion cycle could drive global data-center capex to $1.7 trillion by 2030. Its forecast covers worldwide data-center IT and related equipment across hyperscalers, neocloud providers, sovereign-AI initiatives, telecommunications companies and enterprises. Dell’Oro also expects global capex to approach $1 trillion in 2026.
The four largest U.S. hyperscalers—Amazon, Google, Meta and Microsoft—had raised combined data-center capex to nearly $600 billion entering 2026, according to Dell’Oro. The firm expects those companies to represent about half of global data-center capex by 2030. Accelerated servers used for AI training and specialized workloads could account for roughly two-thirds of data-center infrastructure spending in that year.
These are forward-looking estimates. They do not mean that $1.7 trillion has been approved, financed or spent, and the release does not make the figure a universal industry consensus.
Free tools Windows power users keep installed
One-click scans. No signup required.
#1 Best Overall
- Save valuable floor space: 6U wall mount server cabinet Dimensions: 13.78" H x21.65" W x17.72" D.Maximum mounting depth is 14.2"
- Keep critical network equipment secure: glass door and side panels are lockable to prevent unauthorized access. Front door can be installed on either side of the front of the cabinet to satisfy your door swing orientation preference
- Easy equipment configuration: Fully adjustable mounting rails and numbered U positions, with square holes for easy equipment mounting with top and bottom punch-out panels for easy cable access
- Durability: Made of high quality cold rolled steel holds up to 110lb (50kg) (Easy Assembly Required)
- PCI & HIPPA and EIA/ECA-310-E compliant
“Data-center capex” can mean several different spending pools
Before comparing market forecasts, separate the main categories:
- Physical-infrastructure capex: land preparation, buildings, electrical distribution, transformers, switchgear, UPS systems, generators, cooling plants, liquid-cooling equipment, fiber, security, utility interconnection and on-site power.
- IT capex: CPUs, GPUs and other accelerators, servers, storage, switches, network interface cards, optical equipment, racks and rack-scale systems.
- Tenant fit-out: GPUs, servers and networking installed by a cloud provider or enterprise inside a leased colocation facility. A property-focused estimate may exclude this spending even though it is essential to operating the site.
McKinsey’s more-than-$1.7 trillion estimate through 2030 explicitly excludes IT hardware such as GPUs and servers. JLL separately estimates that tenant IT fit-outs could add $1 trillion to $2 trillion through 2030. A colocation provider’s construction spending and a tenant’s equipment purchase can therefore describe the same project without being the same line item.
How the major estimates compare
| Source | Estimate | Scope | How to interpret it |
|---|---|---|---|
| Dell’Oro | $1.7 trillion by 2030 | Worldwide data-center capex, including IT-oriented categories | Primary source for the headline forecast |
| McKinsey | More than $1.7 trillion through 2030 | Global physical data-center infrastructure; excludes IT hardware | Similar headline number, materially different scope |
| JLL | Up to $3 trillion by 2030 | About $1.2 trillion of real-estate value plus $1 trillion–$2 trillion of tenant IT fit-outs | Broader combined framework; do not add mechanically to Dell’Oro |
| BCG | $1.8 trillion from 2024–2030 | Hyperscaler data-center-related capex in the United States | U.S.-focused comparison |
| CSIS | Up to $2.35 trillion by 2030 | Aggressive scenario for cumulative U.S. generative-AI infrastructure spending | Scenario, not a baseline forecast |
Sources: Dell’Oro, McKinsey, JLL, BCG and CSIS.
Why AI is making data centers more expensive
AI clusters are not simply larger versions of ordinary enterprise server rooms. Training models requires thousands of accelerators working in parallel, with high-bandwidth, low-latency connections between them. The result is greater power density, more networking and storage, and much more demanding thermal management.
- Compute density: GPUs and specialized accelerators consume far more power per rack than conventional enterprise servers.
- Networking: AI training depends on high-speed switches, optical links and network interface hardware to keep accelerators synchronized.
- Cooling: Direct-to-chip liquid cooling and other advanced systems are increasingly needed at high rack densities.
- Power delivery: Larger transformers, switchgear, UPS systems, generators and redundant distribution raise both construction cost and design complexity.
- Storage: Training datasets, checkpoints and model outputs require high-capacity, high-performance storage.
- Inference locations: Serving models to users can require capacity in multiple regions rather than one centralized training campus.
Dell’Oro identifies larger AI clusters, high-performance networking, storage, inference capacity, advanced power and cooling as principal drivers of the new cycle. McKinsey describes AI data centers as integrated power-and-thermal systems rather than merely rooms filled with servers.
Sources: Dell’Oro and McKinsey.
Training and inference create different infrastructure needs
Training
Training is concentrated in very large clusters that require tightly coupled accelerators, exceptional networking and substantial power at one site. McKinsey estimates AI training-data-center demand could rise from 31 gigawatts to 62 gigawatts by 2030.
Inference
Inference runs a trained model for users. It can be distributed across regions to reduce latency and may eventually require more total capacity as AI applications become widely used. McKinsey estimates inference demand could rise from 31 gigawatts to 93 gigawatts by 2030. JLL expects inference to become the dominant AI requirement around 2027, but says that outcome depends on applications that have not yet reached mass adoption.
AI is the fastest-growing demand source, not the only one. Traditional cloud services, enterprise digitization, edge computing, high-performance computing and ordinary internet workloads remain significant. BCG estimates generative AI could drive about 60% of data-center power-demand growth from 2023 to 2028 while representing roughly 35% of total demand in 2028; non-AI workloads would still account for about 55%.
Rank #2
- Save valuable floor space: 12U wall mount server cabinet Dimensions: 24.25" H x21.65" W x17.72" D. MAXIMUM MOUNTING DEPTH is 14.2".
- Keep critical network equipment secure: glass door and side panels are lockable to prevent unauthorized access; Front door can be installed on either side of the front of the cabinet to satisfy your door swing orientation preference
- Easy equipment configuration: Fully adjustable mounting rails and numbered U positions, with square holes for easy equipment mounting with top and bottom punchout panels for easy cable access
- Durability: Made of high quality cold rolled steel holds up to 110lb (50kg) (Easy Assembly Required)
- PCI & HIPPA and EIA/ECA-310-E compliant
Sources: McKinsey, JLL and BCG.
Who is funding the buildout?
Hyperscalers
Amazon, Microsoft, Google and Meta are the largest spending group in Dell’Oro’s forecast and are expected to account for about half of global capex by 2030. BCG expects hyperscalers to generate approximately 60% of industry growth from 2023 to 2028.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Neoclouds and AI model builders
Specialized GPU-cloud providers rent AI-optimized infrastructure, often growing faster than general-purpose cloud providers from a smaller base. Frontier-model companies may buy or lease dedicated capacity rather than rely entirely on standard public-cloud regions.
Sovereign-AI programs
Governments and state-backed entities are funding domestic compute for strategic, regulatory and national-security reasons.
Colocation providers
Colocation companies develop powered facilities and lease space to cloud providers, enterprises and AI customers. Their construction capex is often measured separately from the tenant’s GPU and networking purchases.
Enterprises
Enterprise spending outside hyperscalers is more constrained. Dell’Oro cites tariffs, monetary policy and uncertainty about AI returns as limiting factors.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minutePower—not just chips—is the central bottleneck
Projects can have land and financing yet remain unusable without an electricity connection. McKinsey projects global data-center electricity demand of about 1,400 terawatt-hours in 2030—roughly 4% of global power demand—and worldwide data-center capacity of approximately 220 gigawatts. Its U.S. estimate rises from 147 TWh in 2023 to 606 TWh in 2030, or 11.7% of U.S. power demand.
JLL says average waits for grid connections in primary data-center markets exceed four years. Developers are therefore considering behind-the-meter generation, on-site power and colocated batteries. Speed to power is JLL’s leading site-selection criterion, followed by community support, latency and customer proximity.
Rank #3
- Sturdy:4u server rack is construct from cold rolled steel, with a weight capacity of 110lbs(50kg); Electrostatic powder coat prevents rust and corrosion,quality finish
- Direct use:Open and use, not having to assemble it.Network rack can be placed flat or mounted on the wall,also can be installed vertically under the table
- Design Features:maximum mounting depth of 14 in,cables can be fixed on the side panel;Open frame server rack achieves effortless inspection, replacement and assemble
- Installation:wall mount network rack is easy to install,with instructions or videos for reference;Equipped with multiple accessories, suitable for different needs
- Application:EIA/ECA-310-E Compliant;wall mounted 4u rack fits all 19" racks and cabinets to hold various IT, network, and AV equipment;wall mount rack available in 4U, 6U, and 8U to choose
- JLL forecasts global capacity of about 200 GW by 2030, including roughly 100 GW added from 2026 through 2030.
- Average global construction cost rose from $7.7 million per MW in 2020 to $10.7 million per MW in 2025; JLL forecasts $11.3 million per MW in 2026.
- JLL says an AI tenant fit-out can cost as much as $25 million per MW, separate from shell-and-core construction.
Where new capacity is most likely to be built
Buildout will not be evenly distributed. Developers weigh power availability and interconnection speed, electricity cost and carbon intensity, land, fiber, cooling conditions, latency, permits, tax incentives, community acceptance and construction labor.
McKinsey identifies the United States as the largest data-center investment market, followed by China. Nordic countries are attracting growth because of lower-cost, lower-carbon electricity, cooler temperatures and room to scale. A project announcement should not be treated as operating capacity: land, permits, financing, equipment, grid connection and tenant hardware may still be outstanding.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Where investors and suppliers may capture spending
The investment opportunity extends beyond GPU manufacturers. Demand can reach transformers, switchgear, power semiconductors, UPS systems, generators, batteries, liquid cooling, racks, optical networking, storage, construction, engineering, utilities and data-center real estate. McKinsey says manufacturers of electrical, thermal and mechanical equipment face strong demand but also delivery and innovation constraints.
For an organization deciding how to obtain AI capacity, the practical choices are:
- Rent public-cloud or neocloud GPUs when flexibility and limited upfront capital matter.
- Lease colocation space when the organization needs control of hardware but not a full facility build.
- Buy servers and build privately when utilization is predictable and the organization can manage power, cooling, networking and refresh cycles.
Cloud and equipment prices vary by region, hardware generation, commitment, redundancy and utilization. A headline market forecast is not a quote for any particular project.
What could derail the forecast?
- AI adoption and revenue may grow more slowly than expected.
- More efficient models or chips could reduce compute per task.
- Lower inference costs could also increase total usage by making more applications economical, offsetting those efficiency gains.
- Rapid GPU replacement and depreciation could weaken returns on older clusters.
- Grid queues, transformer shortages, permitting and local opposition could delay projects.
- Interest rates, debt availability, power prices, water limits and emissions rules could change project economics.
- Hyperscaler concentration creates customer and credit risk for developers.
- A shift from centralized training to distributed inference could leave some sites poorly located for future demand.
CSIS notes that efficiency can lower the cost of individual workloads while stimulating more total demand. BCG likewise says technology advances could reduce power requirements, while more computationally intensive inference could push them higher. Dell’Oro has also highlighted increased scrutiny of returns on AI infrastructure.
Sources: CSIS, BCG and Dell’Oro.
How to evaluate any data-center spending forecast
- Check whether GPUs, servers, storage and networking are included.
- Check whether the geography is global, U.S.-only or limited to named providers.
- Check whether the number is an annual run rate or cumulative spending through a date.
- Identify treatment of leased facilities, tenant fit-outs and power infrastructure.
- Review assumptions for training, inference, utilization and hardware replacement.
- Compare projected spending with available grid capacity and interconnection timelines.
- Determine whether the estimate is a base case, upside case or unconstrained-demand scenario.
For personal investors, this framework matters because exposure can sit in different businesses: semiconductor and networking suppliers, electrical-equipment makers, utilities, power developers, colocation operators, construction firms or landlords. Each has different customer concentration, financing needs, cyclicality and sensitivity to AI demand.
The Bottom Line
Dell’Oro’s $1.7 trillion figure is a credible scale indicator for an AI-led investment supercycle, not a guaranteed pot of money or a universal industry total. The result depends on whether analysts count IT hardware, physical facilities, power systems, tenant fit-outs or only selected companies. AI is accelerating spending, but grid access, cooling, construction, financing, utilization and the economics of inference will determine how much of the forecast becomes real capacity.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
