A Tucson entrepreneur wants to turn cannabis farms into mini data centers

For Tucson, a proposal to convert former indoor cannabis farms into small data centers could offer a way to reuse buildings that may otherwise sit empty as the marijuana industry contracts. The idea also enters a contentious local debate over how much water and electricity data centers should consume. Rather than constructing large new facilities, entrepreneur Moe Asnani wants to adapt existing cultivation sites into smaller computing hubs, a concept that has drawn cautious interest from at least one Tucson City Council member.

Asnani has worked in the cannabis industry for more than 13 years. He formerly owned two D2 Dispensaries in Tucson and created iLAVA, a cannabis brand that produced gels containing THC for pain relief. He has since moved away from cultivation and is developing a business concept that would place computing equipment inside former grow facilities. His argument is that the buildings already have substantial electrical, cooling and water infrastructure, making them possible candidates for a new use.

The proposed facilities would fall between a server rack in an office and a massive hyperscale data center. Asnani envisions sites capable of serving several hundred thousand users while keeping their water and power needs roughly comparable to those of an indoor cannabis farm. David Ebert, the University of Arizona’s chief AI and data science officer, describes a data center in basic terms as a room or facility where computing takes place. Under that broad definition, the scale can range from a desk-sized installation to major projects such as Project Blue, the controversial facility planned near the Tucson fairgrounds.

The timing reflects a broader change in Arizona’s cannabis market. State cannabis sales reached a high point in 2024 and have declined since then, a trend experts attribute in part to an oversupply of marijuana that has pushed prices lower. As businesses leave cultivation, Asnani believes some of their facilities could be preserved through what planners call adaptive reuse. Arizona does not require growers to report their water consumption, so the industry’s precise statewide footprint is difficult to calculate. However, Asnani’s former company estimated that its 18,000 square foot growing operation used about 1.7 million gallons of water during 2024.

That figure is far below the possible annual consumption associated with Project Blue, which could use as much as 31 million gallons. The comparison does not establish that a converted grow facility would be environmentally preferable in every case, particularly because a data center’s needs would depend on its equipment, cooling system and workload. Still, Tucson Council Member Paul Cunningham said he is willing to examine conversions that avoid the construction of entirely new industrial sites. His position is not an endorsement of artificial intelligence or data centers, he said, but an effort to limit impacts from facilities that he expects will continue to be built.

Asnani also sees small facilities as a foundation for what he calls sovereign AI. In that model, companies and other users would retain greater control over where their data is processed instead of depending entirely on large technology operators. He believes distributing computing among smaller nodes could reduce exposure to cybersecurity problems and give customers more control over software updates and data handling. Other major AI companies are pursuing smaller data centers as well, including OpenAI and Anthropic, which have recently invested in facilities on a scale similar to the type Asnani is considering.

Ebert questions whether smaller operations would necessarily be more efficient or secure. Large data centers can distribute computing demand across thousands of users, keeping equipment busy and spreading the cost of advanced systems. He said large facilities generally have a power usage effectiveness rating just above 1.0, while smaller centers may be closer to 2.0 or 2.8. A higher figure indicates that more energy is being used beyond the computing itself. Ebert also said security depends more on network protections and qualified cybersecurity personnel than on whether a facility serves a small or large number of customers, noting that banks and credit card companies commonly rely on large data centers.

The two men also disagree about what improving computer chips will mean for the industry. Both expect chips and other hardware to become substantially more efficient within the next five years. Asnani believes those gains could allow modest facilities to handle much more computing, weakening the case for enormous projects. Ebert sees another possibility: smaller centers could be forced to replace equipment more often to keep pace with rapid technological changes. He added that advanced cooling systems and energy efficient hardware may require considerable upfront spending, with large operators better positioned to benefit from economies of scale.

No conversion has been announced, and Asnani remains at the beginning of the process. His immediate goal is to persuade local officials and Tucson residents that the idea has practical value. He wants the approval process to focus on whether a proposed reuse would increase water or power consumption compared with the building’s previous operation. Cunningham supports making adaptive reuse easier when those impacts can be limited, especially compared with building new facilities on undeveloped land around Tucson. For residents watching the Project Blue debate, the proposal offers a different question: whether some of the region’s future computing demand can be absorbed by buildings already designed for intensive industrial use, rather than by creating more large developments from scratch.