AI Data Centers Are Eyeing Struggling Cannabis Grow Facilities

August 24, 2026

Photo by wal_172619 via Pixabay.

Introduction

You could forgive a casual observer for thinking the cannabis industry had finally found its white knight. After years of brutal price compression, oversupply, and the slow-motion collapse of the boom-era cultivation bubble, a new savior appeared on the horizon: artificial intelligence. Not the kind that helps growers optimize their lighting schedules or predict harvest yields, but the kind that devours electricity by the megawatt and demands vast, climate-controlled spaces to house its ever-hungry servers.

The narrative writes itself. Distressed cannabis grow houses, built with heavy-duty electrical infrastructure and industrial HVAC systems during the heady days of market projections that never materialized, now sit half-empty or fully dark. Meanwhile, AI data center developers can't build fast enough to keep up with demand, waiting years for utility interconnection queues and fighting zoning battles in power-strapped markets. A match made in heaven? Not quite.

Why cannabis grow sites are suddenly on data center shopping lists

Photo by Pixabay via Pexels.

Walk into any large-scale cannabis cultivation facility built during the boom years and you'll find something that looks nothing like a typical warehouse: three-phase electrical service rated for hundreds of amps, industrial HVAC and dehumidification systems sized to handle thousands of watts of grow light heat, and often a dedicated substation or transformer upgrade the operator paid six or seven figures to install. That infrastructure was expensive to build and, for a growing number of operators, has become expensive to keep running at a loss. Now it's exactly what AI data center developers are hunting for.

The timing isn't a coincidence. Cannabis cultivation has spent the last three years absorbing a brutal supply-and-demand correction. Too many licensed grows came online chasing a market that never materialized at the size projections promised, wholesale flower prices in mature markets have fallen to a fraction of what they fetched five years ago, and cheap, loosely regulated hemp-derived THC products have siphoned off price-sensitive consumers who used to buy from licensed dispensaries. The result is a wave of cultivators downsizing their canopy, subletting space, or shutting down entirely.

Colorado's numbers tell the story plainly. The state posted a record $2.23 billion in licensed cannabis sales in 2021, at the tail end of the pandemic-driven buying surge. By 2025, that figure had fallen to roughly $1.32 billion — a decline of nearly 41% in a market widely viewed as one of the country's most mature and best-regulated. Denver cannabis attorney Tom Downey, of the firm Ireland Stapleton, has pointed to this gap as the predictable outcome of licensing and construction decisions made when everyone assumed demand would keep climbing indefinitely. It didn't, and the buildout that outpaced real consumption is now sitting there, often underutilized, with a power bill still attached.

That's the opening data center developers are stepping into. Building a new AI-ready data center from bare land means years of utility interconnection queues, substation construction, and zoning fights that can stall a project indefinitely in markets where power capacity is already stretched thin. A former cannabis grow, by contrast, often comes with heavy-duty electrical service already energized, industrial zoning already approved, and climate-control systems built from day one to move large volumes of conditioned air. For a developer trying to get compute capacity online fast, that head start can be worth more than the real estate itself.

The power and size gap that makes conversion hard

Photo by Pixabay via Pexels.

The math doesn't work as cleanly as the headlines suggest. Niki Krear of New Lake Capital Partners, one of the larger cannabis-focused real estate investment trusts, has pointed out that the average data center now needs well over 10 megawatts of power to run — and that's before accounting for the redundancy and cooling capacity that hyperscale operators build in on top of that base load. Most cannabis cultivation buildings simply weren't designed anywhere near that scale.

Grow facilities were engineered around a different set of loads entirely: double-ended grow lighting, dehumidification, chilled water for HVAC, and CO2 injection systems. That typically nets out to somewhere in the 3-to-5-megawatt range for a sizable indoor operation. It's a lot of power by warehouse standards, which is exactly why cannabis operators gravitated toward industrial parks with existing utility infrastructure in the first place. But it's a fraction of what a modern AI-training or inference facility draws, and utilities don't upgrade substations overnight just because a new tenant shows up with a bigger appetite.

Square footage is the other mismatch. Cultivation buildings tend to top out around 30,000 square feet — plenty of room for flowering rooms, veg space, drying and a processing area, but nowhere close to the footprint data center developers are chasing. Hyperscale campuses routinely run into the hundreds of thousands of square feet, sometimes over a million, because server halls, cooling plant and backup generation all need room to sprawl. A 30,000-square-foot building with 3-inch dead-loaded concrete floors and eight-foot clear heights just doesn't pencil out for that kind of buildout.

Even where a conversion looks feasible on paper, it's not a quick reposition. Rezoning, permitting, and the utility work needed to bring in adequate power and fiber connectivity can stretch the timeline to as long as five years. That's a long runway for a landlord to sit on a vacant or underused asset, which is part of why New Lake, despite fielding interest from data center players, still had three cultivation facilities on the market for lease as of June 30 — and is marketing them to cannabis operators first, not converting them speculatively.

Where the trend is taking hold

Photo by Taylor Vick via Unsplash.

Krear's ranking puts Texas at the top of the list for conversion potential, with Virginia close behind at number two. That pairing makes sense given Virginia's existing status as the densest data center market in the world, and Texas's aggressive push to build out grid capacity and court hyperscale developers with tax incentives. Georgia comes in next, followed by a cluster of Midwest states, the Rocky Mountain region and Nevada -- markets that share cheap land, available power, and a glut of cannabis real estate built during the boom years that never found a stable tenant base.

Michigan deserves its own mention. The state's cannabis market cratered under oversupply and price compression, leaving behind a stock of grow houses and processing buildings that are often sitting half-used or fully dark. At the same time, Michigan's industrial power rates run noticeably cheaper than what operators pay in California or Massachusetts, two states where cultivation costs got squeezed hard by electricity pricing. That combination -- distressed cannabis buildings plus below-average power costs -- is exactly the setup data center brokers look for.

Harbor Farmz, a Kalamazoo-based operator run by CEO Michael Ward, is testing that thesis directly. The company is marketing its roughly 32,850-square-foot cultivation and processing facility for data center conversion rather than continuing to run it purely as a grow. The site currently draws about 1 megawatt of power, with a near-term path to 1.5 megawatts, and Harbor Farmz has laid out plans to scale the property up to 15 megawatts if a data center tenant or buyer comes through. That's a fifteenfold jump in electrical capacity, which gives a sense of how much headroom these old cannabis buildings can have once you stop thinking of them as grow rooms and start thinking of them as raw industrial shells with power access.

One detail gives the Kalamazoo property a real edge: it's already zoned for data center use, something local outlet Second Wave Media flagged as unusual and valuable. Zoning fights are often the slowest, costliest part of any data center project, so a site that's cleared that hurdle already skips months or years of local government back-and-forth.

The same dynamic is playing out north of the border. A shuttered 130,135-square-foot facility in Saskatoon, Canada, formerly run by Prairie Plant Systems and later CanniMed, has been listed for conversion. No buyer has stepped forward yet, which is a useful reminder that not every distressed cannabis building finds a second life -- location, grid access and local incentives still decide who wins.

Not every grow becomes a server farm

Photo by Borys Zaitsev via Pexels.

The headlines make it sound like every dark grow warehouse in Denver or Coachella is one phone call away from becoming a server farm. That's not how the math works. A full data center conversion needs three-phase power in the multiple megawatts, fiber connectivity that either already runs to the building or can be trenched in without a permitting fight, and enough clear-span square footage to justify the capital it takes to gut a facility and pour new electrical infrastructure. Most cannabis buildings, especially the retrofitted warehouses and converted tilt-up industrial boxes that make up a huge share of licensed cultivation space, simply don't clear that bar. They were sized and wired for HVAC and LED lighting loads, not for racks of GPUs running around the clock. Brokers who've actually walked these deals say the number of cannabis properties that qualify for true hyperscale or even mid-size colocation use is small, maybe a handful in any given state with a mature cultivation footprint. What's happening more often, and more quietly, is a partial play. A landlord sitting on a 60,000-square-foot facility that's only running flower in 20,000 square feet because of price compression carves out the unused portion and leases it to a modular data center operator running containerized compute units. These are shipping-container-style pods, self-contained with their own cooling, that can be dropped onto a slab or parking pad without touching the licensed cultivation space at all. The operator gets power and a roof, the landlord gets a second rent check, and the cannabis tenant keeps growing in the footprint they still need. Nobody's tearing out grow lights to make room for servers; the two uses just sit side by side under one roof or one parcel. Real estate attorneys working distressed cannabis portfolios are careful to frame this as one tool in a bigger kit, not a rescue plan for the industry. Sale-leasebacks, tenant swaps between operators consolidating licenses, and straight-up conversions to non-cannabis industrial use are all on the same table, and which option makes sense depends entirely on the specific building, its power infrastructure, and how much debt is sitting on top of it. The data center angle tends to come up specifically for properties with underused acreage or excess power capacity that a grow operator overbuilt years ago and never fully utilized. That last point is worth sitting with. This trend is less a story about AI's insatiable appetite for compute than about how much cannabis cultivation capacity got built during the boom years and never found a tenant willing to pay for it. Wholesale flower prices in mature markets like California, Colorado, and Oregon have been depressed for years, and a lot of that excess grow space was already sitting half-lit before anyone mentioned AI. Container-based data centers are simply the first buyer to show up with a check for the parts of these buildings that cannabis never needed in the first place.

Conclusion

If there's a takeaway from the sudden interest in cannabis-to-data-center conversions, it's this: the real estate that got built for a speculative cannabis market is now being repurposed for a speculative AI market, and neither industry should mistake this for a stable long-term solution. The buildings that work for conversion are the exception, not the rule. The ones that don't will sit empty or find other tenants entirely.

What makes this moment interesting isn't the handful of deals that actually close. It's what they reveal about the underlying economics of both industries. Cannabis overbuilt because capital was cheap and projections were rosy. AI is currently doing the same thing, with developers racing to secure power and land before the next wave of demand materializes. When that wave inevitably crests, it's worth asking who will be left holding the bag on all those half-built data centers and whether some future industry will find them just as convenient as data center developers now find these old grow houses.

For now, though, the math is what it is. A cannabis facility with excess power capacity and underused square footage has a shot at a second life. A facility built on spec with maxed-out electrical infrastructure and no room to expand doesn't. The operators who built with headroom are the ones getting calls from data center brokers. The ones who built to the minimum are getting calls from their lenders. In both cases, the phone is ringing, but the conversations sound very different.

The cannabis industry spent the last decade learning that supply doesn't always create its own demand. The AI industry, for all its current momentum, would do well to remember that lesson. The grow houses that data center developers are circling today were built for a future that never came. The server farms going up now are being built for a future that may or may not arrive. History has a way of repeating itself, even when the hardware changes.