An estimated $30 billion cost for an AI data center in Texas highlights infrastructure and cost challenges
2–4 minutes
A Bloomberg projection estimates that building and equipping a 1-gigawatt artificial intelligence data center in the Abilene region of Texas could cost around $30 billion. This figure represents a development simulation and does not reflect actual expenses incurred. The main cost components are chips and servers, electrical infrastructure, and cooling systems.
The analysis combines construction estimates from consulting firm Cushman & Wakefield with equipment calculations from Bloomberg Intelligence. To illustrate the scale of such a project, the upper limit of these estimates was used. This amount does not correspond to the actual budget of any specific facility.
In the scenario outlined, chips and servers account for $17.1 billion. Electrical infrastructure represents $3 billion, while cooling costs $1.66 billion. A contingency reserve for unforeseen events and cost increases adds another $2.79 billion.
This breakdown helps explain why announcing an investment in a data center can have different meanings. There is the property and its support systems, the computing equipment, and the operating expenses. Not all of these stages are always financed by a single participant.
Construction cost per megawatt has increased by 21%
A September report by Cushman & Wakefield shows that construction costs per megawatt have increased by an average of 21% since the previous edition, released in the fourth quarter of 2024. The study covers markets in the United States and Canada.
For modern facilities built from scratch in these countries, the consultancy calculates an average development cost of $17.6 million per megawatt, excluding chips and GPUs. This average has a different scope than the Abilene simulation and should not be added to it.
According to the consultancy, supply chain restrictions, wages for specialized professionals, and more expensive components are putting pressure on projects. Electrical infrastructure accounts for an average of 21% of the development cost of the facilities analyzed.
The location of a data center depends on more than just finding a large area. It is necessary to verify whether electricity can reach the facility in the required volume and within the required timeframe. Cushman & Wakefield notes that land with available energy in the main American markets had an annual appreciation of 51% in 2026, up to the date of the survey.
The problem also exists in Brazil. According to Bruno Porto, manager of Industrial, Logistics and Data Center at JLL, electrical infrastructure is the main obstacle to the expansion of the sector, due to the complexity of the approvals and works required.
Demand is affecting suppliers such as WEG, which announced $165 million to expand generator production in North America. The first phase is scheduled to start operating in the second quarter of 2027.
Cushman & Wakefield estimates a global project portfolio of approximately $2.3 trillion. Of this total, $492 billion was under construction, while about $1.8 trillion corresponded to planned projects with prior commitments.
The International Energy Agency projects that global electricity consumption by data centers will increase from 485 terawatt-hours in 2025 to about 950 terawatt-hours in 2030. The agency notes that bottlenecks along the chain reduce the possibility of faster expansion in the short term.
Frequently Asked Questions
Does every AI data center cost $30 billion?
No. This figure is an estimate for a 1-gigawatt project in Abilene, Texas, with specific equipment and assumptions.
What does 1 gigawatt mean?
It is a measure of power, equivalent to one thousand megawatts. It does not represent an amount of energy consumed over a year.
Does the cost include only construction?
No. The simulation presented by Bloomberg also includes computing equipment and other project components.
Has the $2.3 trillion already been invested?
No. Cushman & Wakefield’s estimate includes projects under construction, planned projects, and projects with prior commitments.
