AI infrastructure costs are driven first by accelerator-heavy servers, but a data centre’s full bill also includes the building, electrical systems, grid connection, networking, cooling and ongoing operations. Power availability and price, facility design, utilization and hardware lifetime can change the economics as much as the choice of GPU. Any estimate is useful only when its scope is clear: upfront capital expenditure (CapEx), recurring operating expenditure (OpEx), or annualized total cost of ownership (TCO).
What makes up the cost of an AI data centre?
The bill has two broad parts: the equipment and site needed to build the facility, and the recurring cost of running it. Accelerator servers often take the largest share in AI-focused models, but they depend on power delivery, cooling and high-speed networking. Those supporting systems are not optional extras; their design affects both construction cost and operating expense.
As an Amazon Associate I earn from qualifying purchases.
- Compute: Accelerator-equipped servers, memory and related components. Server type and count drive much of the initial hardware investment.
- Power infrastructure: Grid connection, substations, transformers, backup generation, uninterruptible power supplies (UPS) and power distribution, in addition to electricity bills.
- Networking: Switches and interconnect for communication among servers, as well as front-end connections for traffic entering and leaving the facility.
- Facility and site: Building shell, mechanical and electrical systems, land, utility works and external fiber cabling.
- Cooling: Cooling equipment and construction choices, plus the electricity used to remove heat.
- Operations: Energy, maintenance, labor, taxes and water, depending on what the estimate includes.
A capital-cost figure cannot be compared directly with an annual energy bill or an annualized TCO. TCO spreads capital investment over an assumed operating life and combines it with operating costs; the result therefore depends on assumptions such as useful life and financing.
How much can servers account for?
Epoch AI’s 2026 illustrative model for a U.S. hyperscale AI data centre reports $38 billion in upfront CapEx, $0.9 billion in annual OpEx and $8.5 billion in annualized TCO. It assigns $5 billion a year, or 60% of annualized TCO, to servers. The model assumes NVIDIA GB200 NVL72 systems and includes facility construction, substations and external cabling among its inputs. It is a modeled 1 GW facility, not an observed project or a universal price estimate; actual costs can differ by location, design and procurement.
#1 Best Overall
- 【Powerful Load-bearing】12U Network Rack Open Frame is constructed from durable cold rolled steel; Rack shelf supports enhance stability, wall-mounted capacity of 130lbs, the ground-mounted up to 260lbs
- 【Considerate Designs】Open-frame layout, including a top panel adding space, anti-slip shelf stops fixing devices and compatible racks for stack and expansion to meet requirements of home server rack
- 【Complete Accessories】A 12U open frame server rack, two ventilated shelves, four shelf stops, four velcro straps and a set of equipment mounting screws
- 【Versatile Application】Ideal for space-efficient multi-device setups in warehouses, retail, classrooms, offices and more; Excellent choices as AV Rack/IT Rack
- 【Effortless Setup】 Network Rack includes hardware, a comprehensive manual, mounting hole drilling template and an online assembly video to simplify setup
TrendForce’s 2025 discussion of a typical 125 MW hyperscale data centre also attributes roughly 60% of CapEx to servers. That is a report estimate from a public landing page, which does not expose the detailed cost table. It should not be treated as a directly comparable result to Epoch AI’s annualized TCO share: the facility scale and, crucially, the measure being reported differ.
What changes the server economics?
The number and model of accelerators matter, but so do memory and other server components, power density, useful life and utilization. A lightly used server spreads its purchase and operating costs across less useful work than a heavily utilized one. For that reason, an estimate of total facility spending does not by itself reveal the cost per unit of AI work.
Why is power both a cost and a constraint?
Electricity is a recurring operating expense, while delivering enough power to a site can require costly infrastructure and affect when the facility can be built. Grid connection, substations, transformers, backup systems, UPS equipment and internal distribution are distinct from the price paid per kilowatt-hour. Electricity rates and contract terms vary by location, so there is no single universal power cost for an AI data centre.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Rank #2
- Space Saving: Maximum depth: 14.8". Use the wall mount network cabinet to maximize available space for retail locations, classrooms, back offices, network cabinets, and other locations where space is limited.
- Fast Heat Dissipation: The server cabinet is designed with vents to optimize airflow and avoid critical IT equipment overheating. Heat sink holes in the top, bottom, and rear panels are more conducive to heat dissipation.
- Sturdy Construction: Robust welded frame construction for durability and long service life. With 100 lbs wall-mounted load capacity and 200 lbs ground-mounted load capacity, you can place multiple devices in the server rack cabinet as needed.
- High Security: The locked glass door ensures the security of data and equipment. Wall mount rack enclosure server cabinet is ideal for use in public places such as offices, effectively protecting the security of your devices.
- Hassle-free Installation: Fully adjustable square-hole mounting rails of the wall mount server cabinet facilitate device installation. Wiring holes on the top, bottom, and rear panels provide you with easy cable routing.
The scale of the wider data-centre electricity demand is rising, but the published figures are not AI-only measurements. The International Energy Agency (IEA) estimated global data-centre consumption at 415 TWh in 2024, about 1.5% of global electricity use, and projects about 945 TWh in 2030 in its base case. In that same base case, the IEA projects accelerated-server electricity consumption growing 30% annually, compared with 9% for conventional servers. These are estimates and projections for data centres and server classes, not a measure of the electricity used solely by AI.
For the United States, the Department of Energy’s 2024 announcement, reporting a Lawrence Berkeley National Laboratory (LBNL) study, gives data-centre electricity use of 176 TWh in 2023 and an estimated 325–580 TWh by 2028. LBNL’s 2026 central/reference estimate puts data centres at 11.8% of U.S. electricity in 2030, with a compounded uncertainty range of 521–843 TWh. These U.S. estimates use different forecast years and should not be read as AI-only totals or combined as if they were one measurement.
What do networking and cooling add?
Networking serves different kinds of traffic
Internal or back-end networks let servers exchange data inside the facility; front-end networks carry traffic to and from users and external systems. Their design and capacity depend on the workload and server topology, so the cost of networking is not captured by counting accelerators alone. The IEA estimates that networking equipment can account for up to 5% of data-centre electricity demand. That is an electricity-consumption share, not a networking CapEx share.
Rank #3
- Adjustable Depth: 23-40'' adjustable depth is used for servers and network equipment, ensuring enough space for AV equipment, components, and cabling, while allowing you to access ports and equipment from multiple sides.
- Strong Load Capacity: Ground-Mounted Load Capacity: 500 lbs, Wall-Mounted Load Capacity: 150 lbs. The av rack is made of carbon steel for better weldability performance and can help save space while meeting your need to place multiple devices.
- User-friendly Design: Ergonomic design makes the open frame av rack easier to use. The additional top panel is able to place other items with more available space. Roller design moves anywhere and anytime, is convenient, and is more energy-saving.
- Complete Accessories: We provide the accessories you need, including 2 x Pallets, 145 x M5*10 Cross Head Screws, 4 x Casters, 4 x M10*50 Expansion Screws,10 x M6*12 Cage Nuts, 1 x Grounding Wire, 1 x User Manual.
- Wide Application: The server rack wall mount maximizes the use of available space, suitable for retail venues, classrooms, offices, and other places where space is limited.
Cooling affects equipment and energy use
Cooling adds both construction and operating costs. The IEA estimates that cooling ranges from about 7% of electricity use in efficient hyperscale facilities to more than 30% in less-efficient enterprise facilities. Those figures describe electricity demand, not the share of the facility’s total cost. Cooling method and facility efficiency determine where a particular site falls.
In Epoch AI’s 1 GW model, the facility-construction input includes a 7–10% premium for liquid cooling. This is an assumption in that model, not a general premium applicable to every liquid-cooled data centre. Cooling equipment and construction choices also interact with the facility’s power and thermal design, so a single percentage cannot describe the total cost impact across designs.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why do AI infrastructure cost estimates differ?
Two estimates may both be reasonable and still describe different things. One may cover servers only; another may include site works and power infrastructure. A quoted total may be one-time CapEx, annual OpEx or annualized TCO. A forecast may describe a scenario rather than an observed facility. Before comparing numbers, check:
Rank #4
- An intelligent fan system designed for cooling audio video, DJ, server, network, and IT equipment racks.
- Protects rack-mount equipment from overheating, performance issues, and shortened lifespans.
- Programmable thermostat controller with automated speed control, alarm warnings, and backup memory.
- Premium anodized aluminum construction with CNC-machined detailing for a professional appearance.
- Size: 1U Rack Space | Design: Top Exhaust | Airflow: 60 to 300 CFM | Noise: 12 to 38 dBA | Bearings: Dual Ball
- Scale and scope: Facility size and IT load, and whether the estimate includes land, building, grid works, backup systems, fiber and cooling.
- Geography and power terms: Location, electricity rate and contract assumptions, as well as grid capacity and connection requirements.
- Hardware and utilization: Accelerator model and count, server configuration, network topology and expected utilization.
- Design: Cooling approach, power density and the facility’s electrical and mechanical systems.
- Time and accounting: Asset lifespan, financing or discount-rate assumptions, and whether a number is CapEx, OpEx or annualized TCO.
- Evidence type: Whether the figure comes from an observed project, a vendor or analyst estimate, or a modeled scenario—and what date and geography it represents.
For example, a server-only purchase estimate should not be presented as the cost to build a powered, cooled site. Likewise, the share of electricity consumed by cooling or networking cannot be converted into that equipment’s share of CapEx without separate cost data.
What do broad forecasts say—and what do they not say?
Data-centre electricity forecasts help show the scale of infrastructure pressure, but they do not establish the price of a particular AI facility. The IEA’s global figures are scenario-based, while LBNL’s U.S. estimate includes a wide uncertainty range. Outcomes depend on AI adoption, improvements in efficiency and constraints on infrastructure; neither forecast isolates an all-in AI-only cost.
Free tools Windows power users keep installed
One-click scans. No signup required.
McKinsey’s 2025 article cites $6.7 trillion in cumulative worldwide data-centre capital outlays by 2030. This is a broad forecast for data-centre investment, not a realized expenditure total and not an AI-only tally. It should not be compared directly with the cost of one facility or the annualized TCO of a modeled site.
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.
