The AI boom has caught us all and, whether we want to be or not, it has become inextricably entwined with our lives. At first we embraced it as a loyal assistant. Then, as the race for AI domination hit full speed, it became a hungry industry gobbling land, power and water.

And our enthusiasm began to shift.

What was once a race for GPU compute capacity and the power to support it has become something far more complicated: a race for permission.

That’s because a data centre is more than just a warehouse with pretty flashing lights that generate funny cat videos.

Take a 100 MW data centre. It admittedly isn’t small, but the International Energy Agency uses 100 MW as representative of a hyperscale facility. Running continuously, it can consume up to 876 GWh of electricity per year. With traditional evaporative cooling, it could theoretically consume close to a billion litres of water annually.

Then there’s the land, substations, transmission infrastructure, backup power and cooling equipment. Suddenly, constructing an AI data centre turns into a local infrastructure decision, not a simple technology investment.

And yet, by the standards of the AI arms race now underway, even 100 MW is beginning to look modest. Meta’s new Alberta campus is expected to draw as much as 1,800 MW at full build-out.

Believe me, the surrounding communities are beginning to notice.

Which brings us to HIVE Digital Technologies.

HIVE is positioning itself right at the intersection of Canada’s demand for more AI infrastructure and the growing demand that somebody explain where all the land, electricity and water required to build it is going to come from.

The question is whether HIVE actually has an answer.

Okay, HIVE wasn’t always HIVE. The shell company goes back to 1987 and has operated as various ventures mostly within the resource sector going through several incarnations including Carmelita Petroleum, Carmelita Resources, Pierre Enterprises and eventually Leeta Gold Corp.

Not striking gold as any one of these entities, the company pivoted from its latest mining mandate to acquiring GPU-based cryptocurrency data centres from Genesis Mining. And in 2017, the company opened its doors once again as HIVE Blockchain Technologies Ltd.

HIVE, at that point, ran a data centre in Reykjanes, Iceland. It started with only 2.05 MW, containing roughly 2,300 GPU mining rigs. HIVE immediately expanded it by another 1.75 MW and then began building additional GPU capacity in Sweden. Cryptocurrency was hot and HIVE wasn’t going to be left behind.

Then it happened.

Ethereum went from proof-of-work mining to proof-of-stake in September 2022, basically eliminating Ethereum GPU mining. Suddenly HIVE had GPU infrastructure and expertise that needed more things to do. It had to find another revenue stream.

Instead of tossing the whole thing into the bin, HIVE began repurposing GPUs toward high-performance computing and cloud applications. So now HIVE lives in two worlds: Bitcoin mining and GPU cloud computing for AI/HPC.

Now Bitcoin hasn’t gone the way of the dodo. Cryptocurrency mining still makes up a large portion of the company’s revenue.

In fact, fiscal 2026 saw HIVE expand aggressively in Paraguay, adding 18 EH/s across 300 MW of capacity through its Valenzuela and Yguazú developments. That put the company’s installed Bitcoin mining capacity at approximately 25.5 EH/s.

So Q1 2027 revenue was split thusly:

• Digital-currency revenue: US$72.1 million
• BUZZ HPC revenue: US$7.1 million

And even though HPC revenue made up only 9% of the HIVE total quarterly take, it was still up almost 47% year-over-year.

What does this mean?

Well HIVE isn’t an AI company yet, but its secondary HPC segment is becoming increasingly more material to the company’s bottom line. Enough so, the company split off its HPC activities into a wholly owned subsidiary, BUZZ High Performance Computing. This means HIVE isn’t satisfied with just renting out spare GPUs, it wants to develop BUZZ into a full blown AI cloud and data centre infrastructure operator.

The foundation has already been set. HIVE’s Swedish operation has been around since 2018 and its 32 MW Boden facility is now being repositioned toward Tier III infrastructure. HIVE announced in June that the municipality had approved its acquisition of the previously leased property. It also signed a non-binding LOI contemplating approximately 25 MW of critical IT load there for up to ten years.

Which brings us back to Canada.

HIVE acquired a Toronto facility with existing power access in September 2025 and used it for Bitcoin hashrate services while putting together an upgrade toward Tier III standards that would allow for HPC/AI workloads.

In February 2026, HIVE turned on the jets, signing customer agreements  covering 504 liquid-cooled NVIDIA B200 GPUs located at Bell Canada’s AI Fabric facility in Manitoba. That’s a total contracted value of approximately US$30 million over two years.

When the units became operational in May, management estimated that deployment at about US$15 million of annual recurring revenue, bringing its then-expected annualized HPC business from approximately US$20 million to roughly US$35 million. Approximately 50% of their expected Bitcoin business.

Then the afterburners kicked in.

In May, BUZZ announced plans for a 320 MW AI gigafactory in the Greater Toronto Area and the numbers are eye-watering:

• ~320 MW utility capacity
• 100,000+ GPUs at full build-out
• ~C$3.5 billion estimated capital investment

According to HIVE, this brought their control to more than 850 MW of global power with approximately 450 MW operational plus an additional 400 MW in their development pipeline. The Toronto project made up 320 MW of that pipeline.

We’re still not done.

In June, BUZZ inked a deal with Bell Canada and Cohere to develop Canadian sovereign AI infrastructure at Bell’s purpose-built facility in Merritt, British Columbia. BUZZ would deploy the sovereign AI cloud/GPU infrastructure; Cohere would run its models and enterprise AI products on that compute for Canadian government and corporate customers.

This means that BUZZ is now offering to help government organizations and mission critical companies execute their sensitive AI operations on domestic soil using Canadian infrastructure. This is a big deal in terms of privacy and security.

The business is scaling as HIVE announced in August that its contracted GPU-cloud ARR had reached approximately US$110 million. That doesn’t mean this cash has come through the door, it’s just on the other side, the actual quarterly revenue still remains $7.2 million.

In August, BUZZ announced a US$350 million AI cloud-services agreement with an unnamed investment-grade enterprise customer. Now there’s some meat on the bone here. Five-year deal involving 2,016 NVIDIA Blackwell Ultra GPUs. Capex is expected to come to approximately US$185 million. A US$35 million customer deposit. A Q4 deployment timeframe. HIVE also says the contract adds approximately US$70 million in annualized revenue.

But no reason to dance yet, so just stop it.

We still don’t know who the customer is, what the termination provisions look like, what happens if the customer doesn’t take the expected capacity or what margins HIVE expects to generate. Nor do we know enough about the underlying economics to independently determine how “awesome” this US$350 million contract really is.

So there’s considerably more meat on the bone than there was before, but we’re still missing the steak.

Now, the reason I wrote this piece.

Today, BUZZ announced a partnership with ProCogia. Vancouver-based ProCogia will procure dedicated GPU capacity from BUZZ’s Canadian data centres and put its own applied-AI products and services on top of that infrastructure. The companies also intend to jointly pursue customers in Canada, the United States and Europe.

Before you break out the balloons and kazoos, selling all this compute is only half of HIVE’s challenge.

Remember the 100 MW facility we talked about at the beginning of this article? HIVE intends to build one more than three times that size in the Greater Toronto Area.

Where does 320 MW of power come from? Who pays for the infrastructure required to deliver it? How much land does it consume? How much water does it use? And perhaps most importantly, what does the community hosting the thing get in return?

This isn’t just a thought exercise anymore. It’s a very real concern.

Turns out Canada has been asking exactly the same questions.

On September 3, the federal government revealed its Responsible Data Centre Development Principles, a national framework designed to establish what communities should expect from the companies building this infrastructure.

And wouldn’t you know it, BUZZ HPC is a signatory. Along with Amazon Web Services, Anthropic, Bell, Cohere, Google, Meta, Microsoft, OpenAI, TELUS and a pile of the other companies driving the AI infrastructure boom.

The rules aren’t particularly complicated.

Build something that benefits the community. Don’t stick Canadian ratepayers with the electricity bill. Minimize water consumption and environmental damage. Tell communities what you’re actually building. And make sure Canada gets something meaningful out of the deal.

Pretty reasonable.

Now let’s see how HIVE measures up.

On the power front, HIVE appears to have cleared one of the biggest hurdles.

The company says its 25-acre GTA property has a full 320 MW power allocation, something CEO Aydin Kilic says HIVE spent more than a year working to secure.

But let’s make something clear.

HIVE describes 320 MW as the site’s utility capacity, while the completed facility is anticipated to provide approximately 200 MW of IT load. This means HIVE isn’t claiming 320 MW worth of GPUs. The balance accounts for the broader facility infrastructure required to support the IT load.

Also, HIVE hasn’t publicly identified the utility counterparties, detailed the transmission or substation work required to deliver that 320 MW, or disclosed exactly who will pay for those upgrades.

Doesn’t mean nobody asked.

During HIVE’s June earnings call, an analyst asked management to clarify precisely what the “full allocation” meant, where the project stood in the interconnection process and what permitting remained.

Management declined to go into that level of detail.

So the power appears to be allocated. Turning that allocation into an operating 320 MW data centre is another matter entirely.

What about the billion litres of water?

This is where it gets interesting.

Remember, the hypothetical 100 MW data centre could consume close to a billion litres of water annually using traditional evaporative cooling.

HIVE says its proposed GTA facility, which is more than three times that utility capacity, is being designed around closed-loop liquid cooling with a “no-water-use approach.”

But what does that mean?

Closed-loop cooling doesn’t really mean no water consumption. You can circulate coolant around your GPUs in a closed loop and still have to ultimately dump the heat through an evaporative cooling tower that consumes enormous quantities of water.

However HIVE seems to be looking at something different.

BUZZ  described the use of closed-loop and dry-cooler designs that transfer heat from the computing equipment into circulating coolant and ultimately pump that heat outside rather instead of evaporating water to get rid of it. At its Merritt, B.C. facility, HIVE explicitly states its closed-loop direct liquid cooling eliminates ongoing water consumption for cooling.

If HIVE can replicate that architecture at 320 MW, the billion-litre water problem we laid out at the beginning of this whole thing, largely disappears.

There’s a but…

HIVE hasn’t publicly disclosed the GTA project’s cooling architecture in enough detail to verify exactly what they mean when they say “no-water-use”. No published WUE target, projected annual water withdrawal or explanation of whether dry cooling will take precedence or require evaporative assistance during extreme temperatures.

So for now, we are still at design objective, not operating result.

However, this one isn’t technologically far-fetched. Closed-loop direct liquid cooling paired with dry heat rejection can operate with virtually no ongoing water consumption.

HIVE just has to prove it can do it at 320 MW.

What does the community get?

The company estimates its C$3.5 billion GTA project will create more than 800 construction jobs and hundreds of permanent high-skill positions. This bodes well. Canada’s Responsible Data Centre Development Principles give jobs and infrastructure investment as examples of the lasting benefits these projects should offer their host communities.

HIVE’s sovereign-compute argument works with Ottawa’s want that data centres strengthen Canadian digital resilience, security and domestic compute capacity. HIVE says it intends to put more than 100,000 GPUs under Canadian control and has already begun building relationships with Bell, Cohere and ProCogia to put Canadian AI workloads on Canadian infrastructure.

But Canada’s definition of community benefit is broader than this.

It outlines skills training, apprenticeships, local and Indigenous procurement, tax contributions, research partnerships and access to compute for local businesses and institutions. Ottawa is also of the mind that developers should engage with host communities and Indigenous rights holders early enough that those communities get a say in what the benefits actually are.

That’s where HIVE’s public disclosure gets hazy.

At least HIVE says the project is taking a “Community First Approach,” but hasn’t been defined in any real way as yet.

The jobs are real commitments on paper. The C$3.5 billion investment is substantial. The sovereign-compute argument works too.

But HIVE will have to go further to demonstrate that it is not just building in a community and is actively seeking to do it for the benefit of that community.

So where does that leave HIVE as a stock?

Still interesting.

Yes, HIVE is a Bitcoin miner. The AI headlines shouldn’t detract from that. Crypto brings in the majority of its cash, so the world of Bitcoin still materially affects its stock performance.

However, the BUZZ side-hustle is gaining steam and shouldn’t be ignored out of hand.

The company has real GPUs, real customers, real revenue and increasingly large multi-year contracts. Its latest US$350 million agreement comes with a customer deposit, a defined deployment of NVIDIA Blackwell Ultra GPUs and a five-year term. HIVE says its GPU-cloud business is already generating around US$100,000 per day and that it has signed more than US$600 million in GPU-cloud contracts so far this year.

That’s a real business.

But how big will it grow? That is the question.

It isn’t about what HIVE is doing right now that will trigger this AI transformation. It’s that proposed 320 MW GTA gigafactory on the horizon.

If the company actually builds it and hits numbers anywhere close to what management is projecting, the revenue mix of this company could look radically different from the Bitcoin miner investors own today.

But this is also the risk.

We’re talking about financing billions of dollars of infrastructure, turning a 320 MW power allocation into an operating facility, securing the remaining permits and interconnections, packing that facility with extraordinarily expensive hardware, while bringing in and keeping customers. And making all this happen without colliding with community resistance that increasingly appears when faced with massive data-centre projects.

HIVE does appear to understand that last part.

But promises aren’t infrastructure.

Look, HIVE spent the better part of a decade learning how to turn cheap electricity into computing power. Bitcoin brought it this far. AI may take it to the next level.

Now it has to prove it can build what it has promised and solve AI’s latest challenge: data centre acceptance.

 

–Gaalen Engen

*the author has no connection to or relationship with the companies named in this article