Marijuana farms are becoming AI data centres in the US
TechRadar reports that falling cannabis revenues are pushing growers to consider converting cultivation buildings into small AI computing sites. The sites already have power, water and cooling, but upgrades, zoning and costs remain barriers.

Key takeaways
- Falling cannabis revenues are pushing some growers to consider converting cultivation buildings into small data centres.
- Existing grow sites can offer power, water, cooling and fibre connections that may reduce setup work.
- A Kalamazoo property has 1.5 megawatts of electrical capacity, with possible expansion to 15 megawatts.
- Zoning, upgrade costs, electricity prices and fast-changing AI hardware remain major obstacles.
In this article
What is changing
Marijuana farms data centres are being explored as a new role for cannabis cultivation sites, according to TechRadar. Data centres are buildings that house computing equipment, and AI refers to artificial intelligence. Falling marijuana prices and rising operating costs are making some growers reconsider what to do with their properties.
AI infrastructure growth is creating another possible use for buildings first built for large-scale indoor marijuana production. Some former and current operators are studying whether existing grow buildings could hold small data centres without new construction. The material describes proposals and assessments rather than completed conversions.
The reports describe this as a possible commercial reuse, not a completed industry shift. This kind of project is sometimes called adaptive reuse, meaning an existing building is given a new purpose. The trend is being driven by both falling cannabis revenues and rising AI demand.
Why grow sites have useful infrastructure

Indoor cannabis growing can require large amounts of electricity and water. Lights, fans, cooling and humidity control all add to demand. Indoor growing can therefore create demand for industrial-style utilities.
As a result, some sites already have power connections, industrial cooling gear and water systems. TechRadar says such features could lower the setup work needed for computing operations.
One cultivation property in Kalamazoo, Michigan, covers 32,850 square feet and has 1.5 megawatts of electrical capacity, a measure of power. The supply could rise to 2.5 megawatts within a year. Longer-term work could eventually lift capacity to 15 megawatts. The property also has fibre internet and a sophisticated climate-control system that can keep indoor temperatures steady.
The example shows how much power some grow sites already have. Such features could make old grow facilities appealing to firms that want smaller computing sites rather than new industrial complexes.
What could hold projects back
Conversion is not straightforward. A computing site may need two or three times as much power as many grow facilities, so upgrades can be crucial. Those upgrades could take up to five years, which may delay projects and reduce the appeal of buying low-cost agricultural land. The reports say such improvements could be expensive and time-consuming.
Before choosing a property, operators must weigh power costs, fibre links and taxes. Smaller projects may also face strict zoning rules, which control land use, building limits, local opposition and costly power-network improvements. They must also check whether local power networks can support computing loads.
The economics may become harder as AI processors, the chips used for AI computing, improve. Sites may need frequent hardware replacement to keep performance competitive. Large operators can buy equipment in large quantities, while smaller sites may pay more for each upgrade.
Even so, opposition to very large computing developments could push operators to look at existing industrial properties. That interest will depend on whether the infrastructure meets commercial needs.
Why the shift matters
AI demand is giving struggling cannabis sites an unexpected commercial role. Old marijuana farms already have infrastructure that may support smaller AI data centres, according to TechRadar. It also shows how AI infrastructure can seek sites with existing power and cooling.
For growers, conversion may offer a way to leave a weak cannabis market while keeping a property productive. For communities, reusing existing sites could be more acceptable than building new data centres from scratch. Tucson City Council member Paul Cunningham said existing industrial properties may be a better option than entirely new facilities. He said data centres are often viewed negatively, and few communities want a brand-new facility.
Moe Asnani, former owner of two D2 Dispensaries in Tucson, Arizona, said some cannabis facilities could become small data centre nodes. He said Arizona produces more cannabis than its residents consume, which is pushing some cultivators out of the business. The reports add that infrastructure and rules still create obstacles. The reports present the idea as a possible solution for underused cannabis properties.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



