Arizona Public Media · Data Centers
A Tucson entrepreneur wants to turn cannabis farms into mini data centers
By Kieran Hadley · · Listen 4:04
Read the full article on AZPM(opens in a new tab)Press
In the news, and the hard questions it raised.
Arizona Public Media profiled Moe Asnani, UNICANN’s founder, and asked an AI researcher and a Tucson council member what they made of the idea. Below are their words, verbatim, and our answers to each critique, including what those answers do not settle.
Three voices
A founder, a researcher, and a local official.
The founder
“It’s an easy transition to say, ‘We'll use this building for something else.’”
“I think making that easier and saying, ‘OK, is your power usage going to be more than before? Is your water usage going to be more than before?’ If those things are easy to answer and have the right answer, then I think it shouldn't be hard.”
The researcher
“Most large scale data centers have a power use efficiency just above 1.0, where most small scale data centers have it closer to 2.0 or 2.8. And the higher the number, the less efficient it is.”
“And it's what an entrepreneur and innovator can do and create that demand and show its advantages. So I think everything's worth exploring.”
The council member
“But if we have buildings that were using that amount of water and that amount of energy already in their previous function and [it] can be done without being invasive, I'm OK with exploring that option.”
Not an endorsement
“Councilman Cunningham says his support of adaptive reuse is not an endorsement of AI or data centers, but a choice to reduce inevitable impact.”
AZPM, reporting by Kieran Hadley
Where they agree, in AZPM’s words: “What Cunningham, Ebert and Asnani do agree on is the fact that data centers, whether people like them or not, aren’t going anywhere.”
As reported
What AZPM reported, in plain terms.
These are the article’s facts, attributed to AZPM. They are not UNICANN measurements for any site.
- Arizona cannabis sales peaked in 2024 and have declined since, a pattern AZPM reported experts attribute to an oversupply driving down prices.
- Arizona does not require growers to report their water use, so exact figures for cannabis farms are hard to find.
- D2, the founder’s former company, estimated that its 18,000-square-foot grow used about 1.7 million gallons of water in 2024. That is one company’s estimate, not a metered baseline for any site UNICANN screens.
- The idea is at an early stage. AZPM reported that the founder’s first step is to convince local government and the public that it is useful.
The critiques, answered
Five fair questions. Our answers, and their limits.
Each critique below comes from the article. Our answer cites independent sources where we have them and says plainly what it does not prove.
01
Aren’t small data centers less efficient?
“Most large scale data centers have a power use efficiency just above 1.0, where most small scale data centers have it closer to 2.0 or 2.8. And the higher the number, the less efficient it is.”
Our answer
Survey data agree on the direction. Uptime Institute’s 2025 global survey found that sites of 20 MW and above averaged a power usage effectiveness (PUE) of 1.44, against 1.54 for all respondents and 1.48 for facilities built in the past five years. Uptime adds that the largest sites tend to be newer, and that utilization, design, age, redundancy, and climate also shape the number.
Figures near 2.0 typically describe much smaller spaces. The Congressional Research Service notes that a PUE of 2 is typical of a small data center, one with less than 150 square feet of floor space. A 1.5–5 MW node is a different class of project: a new, engineered power and cooling design inside a converted industrial building.
UNICANN’s planning scenarios assume a PUE of 1.4 to 1.6. That is deliberately less efficient than the best large campuses, not a claim to match them.
What this does not settle
A converted building will not automatically reach any PUE. Efficiency has to be designed for the site’s climate and load, then measured after it is running.
Uptime Institute, 2025(opens in a new tab)Uptime Institute Journal, Feb. 7, 2024(opens in a new tab)Congressional Research Service, updated May 12, 2026(opens in a new tab)Compare cooling designs by climate02
Won’t a small node sit idle?
“Small data centers used by fewer clients, by contrast, could sit idle more often.”
Our answer
Idle capacity is a real risk for any data center, and a dedicated node with one main user carries it directly. Nobody else’s demand fills the gap.
That is why UNICANN asks buyers for a committed workload and a utilization case before a conversion is considered. It is also why the site’s scenario planner assumes 75–95% occupied capacity instead of billing every megawatt.
See what buyers are asked to proveWhat this does not settle
A node without durable, documented demand should not be built. Screening can test the case for demand; it cannot create it.
03
Is a smaller site really more secure?
“Ebert also pushed back on the idea that smaller data centers are more secure, saying that network safeguards and professional cybersecurity staff are more important than the number of users or connection points.”
Our answer
Ebert’s point is sound: network safeguards and trained security staff matter more than a site’s size.
The claim this site makes is narrower. A dedicated node lets one organization control the hardware and know which building its files are in. That is a question of custody and location, which some contracts and regulations care about.
“The whole idea is if you create these smaller nodes where people control their own data, where they're not at the mercy of other people's updates and things like that…”
Read the Verification StandardWhat this does not settle
Owning the hardware is not the same as being secure. A dedicated node needs the same network controls, monitoring, and professional staff as any other data center.
04
Won’t faster chips leave small sites behind?
“But he thinks that evolution will hurt small centers more, as they will need to constantly upgrade their hardware in order to follow the pace of evolution.”
Our answer
Chips do turn over quickly. Goldman Sachs Global Institute puts typical useful lives at four to six years for AI accelerators, about 20 years for buildings, and 25 years or more for power infrastructure.
That gap cuts both ways. Whoever owns the hardware must budget for refreshes. But the building and its power service can host several hardware generations, and each generation tends to do more work per watt. The vendor-reported comparisons in UNICANN’s efficiency explorer show about 5× more output per megawatt for one current rack-scale system than a prior generation on the same reasoning model.
“If the same laws apply in terms of evolution of microchips and computers, there’ll be something more power efficient by 2029 or 2030”
What this does not settle
Refresh costs are real and fall on the hardware owner. Vendor efficiency figures are claims for specific systems and workloads, not guarantees for any site.
Goldman Sachs Global Institute, May 1, 2026(opens in a new tab)UNICANN efficiency explorer, reviewed Sep. 27, 2026See what today’s models need05
Can a small operation afford efficient cooling?
“Technology that doesn’t use water or uses much less electricity or cooling is something that could be a higher cost at the beginning”
Our answer
Efficient cooling does cost more at the start. A grow’s existing air conditioning was sized for plants, not servers, so UNICANN does not assume it can be reused.
The anonymized case study carries a separate allowance for cooling. The water-and-power tool compares dry, closed-loop designs with water-using hybrids across six U.S. climates.
Open the anonymized case studyWhat this does not settle
Whether efficient cooling pays off depends on the site, the climate, and the load. It needs an engineering design and real quotes before anyone commits capital.
A fair test for any conversion.
“…repurposing buildings and putting restrictions on their water and energy use is way better than building brand new monstrosities on the outskirts that are totally invasive to our way of life.”
Comparable meters, utility documents, and engineering review answer whether a building can host a node without using more water or power than it did before. UNICANN’s Verification Standard is built around that evidence.
Sources and disclosures
Every quotation links to the original article. Read it in full for context.
Sources (reviewed Nov. 10, 2026)
- A Tucson entrepreneur wants to turn cannabis farms into mini data centers · AZPM (Arizona Public Media) · Nov. 9, 2026
- Uptime Institute Global Data Center Survey 2025 · Uptime Institute · 2025
- Large data centers are mostly more efficient, analysis confirms · Uptime Institute Journal · Feb. 7, 2024
- Data Centers and Their Energy Consumption (R48646) · Congressional Research Service · updated May 12, 2026
- Tracking Trillions: The Assumptions Shaping the Scale of the AI Build-Out · Goldman Sachs Global Institute · May 1, 2026
- Vendor-reported efficiency comparisons (sources named in the explorer) · UNICANN efficiency explorer · reviewed Sep. 27, 2026
AZPM’s coverage is independent journalism. Quotations are reproduced verbatim from the article published Nov. 9, 2026, and link to the original; read the full article for context. The article profiles UNICANN’s founder and does not name UNICANN. AZPM, its reporter, the University of Arizona, Dr. Ebert, and Council Member Cunningham have not reviewed or endorsed UNICANN, and Council Member Cunningham has said his support for adaptive reuse is not an endorsement of AI or data centers. Responses on this page are educational, not engineering, legal, or investment advice, and every site still needs utility, engineering, and permit review.