AI Data Center Costs Reach $38B per GW, with Power and Location Driving Prices
A 1 GW AI data center in the US can cost about $38 billion when AI chips are included, according to a cost model by research group Epoch AI. Without chips, construction and related infrastructure average about $17.6 million per MW, or roughly $17.6 billion per GW. Servers and networking equipment account for nearly 70% of Epoch AI’s estimated total; buildings and electrical and mechanical systems account for about 30%.
AI data center costs vary widely by location. Turner & Townsend’s 2025–26 index puts traditional data center construction at $13.31 per watt in Silicon Valley, compared with $9.54 in Dallas and $9.50 in Charlotte. The firm also found that high-density, liquid-cooled AI facilities cost 7% to 10% more to build than conventional air-cooled sites. Overall costs per MW have risen 21% in less than two years, while prices for powered land in major US markets have climbed 51% year over year.
Electricity adds a substantial ongoing expense: US industrial power averaged 9.77 cents per kWh, with California at 25.11 cents and Texas at 7.07 cents. Developers are increasingly looking for locations with available power, land and skilled labor. These rising AI data center costs could influence where technology companies invest and how quickly new computing capacity comes online.
Neutral
The article concerns the cost, construction and power requirements of AI data centers, not cryptocurrency prices, blockchain projects or digital-asset regulation. Its direct trading impact on BTC, ETH and other tokens is therefore likely limited, and the appropriate market view is neutral.
In the short term, the figures could draw attention to the capital intensity of the AI boom and the constraints around power, land and skilled labor. Traders may monitor shares of semiconductor, cloud and data-center companies, as well as broader technology risk sentiment. Any movement in crypto would more likely come indirectly through changes in risk appetite or correlations with technology stocks than from the reported construction costs themselves.
Over the longer term, data center investment may support demand for computing hardware and electricity, while rising costs could slow or redirect some projects. That may influence sentiment around AI-related technology firms and, indirectly, crypto assets that trade alongside high-growth risk markets. However, the article provides no new information about crypto adoption, token demand or regulation. Unlike announcements of major crypto product launches or policy changes, which have at times prompted direct market repricing, this is an infrastructure-cost report. Traders should treat it as contextual information rather than a standalone trading catalyst.