The Lens · The Pivot Being Mispriced
The Miners Quietly Becoming AI Landlords
A bitcoin miner just signed a $220 million deal to run AI for Bell Canada and Cohere. That's not a crypto story — it's the clearest sign yet of a quiet rotation: the miners sitting on cheap power and ready-built data centers are turning into AI compute landlords, while the market still prices some of them like the hash machines they used to be. Hover or tap any underlined term.
Dragonfly Lens · June 19, 2026 · The rotation from hash machines to AI landlords — who's real, who's pumping.
The short version
- What's real: HIVE Digital's compute arm signed a 3-year, ~$220M deal to run Cohere's AI on Bell Canada's network — 2,304 NVIDIA GB200 GPUs, adding ~$70M of recurring revenue. Stock jumped ~10%.
- The rotation: bitcoin miners have the two things AI is starving for — cheap power and existing data centers. The smart ones are converting from mining a volatile commodity into renting contracted compute.
- Why it matters: a miner that pulls this off re-rates — from a boom-bust hashprice stock into a company with predictable, contracted revenue. Different multiple, different business.
- The opportunity: a small basket is making this turn. What separates the winners is boring and measurable: contracted revenue, power cost, cooling, and balance sheet.
- The honest caveat: the press releases are pumped, and the GPUs are often funded with dilutive debt. Lead with the revenue number, not the rhetoric.
What actually happened
On June 18, 2026, HIVE Digital — a company most people still file under “crypto miner” — announced through its BUZZ HPC subsidiary a three-year, roughly $220 million contract to provide AI compute for Cohere (Canada's enterprise AI model company) running on Bell Canada's network. The hardware: 2,304 NVIDIA GB200 chips, liquid-cooled, deployed at Bell's facility in Merritt, British Columbia, expected live late-2026 into 2027. HIVE says it adds about $70 million in annual recurring revenue, pushing its contracted compute revenue past $100 million. The stock rose roughly 10% on the news.
Read the press release with one eye closed. The contract is real and verifiable. The framing around it — executive quotes about “a decade of asymmetric returns” — is stock promotion, not analysis. The discipline: anchor on the number you can check ($70M recurring revenue), and treat the adjectives as noise.
Why miners are becoming AI landlords
This isn't a gimmick — it's economic gravity. The thing throttling the AI buildout is power and the data centers to put it in. Guess who already has both:
They sit on cheap, often stranded power. Miners spent years hunting the cheapest electricity on Earth — the single biggest input AI compute needs. That power contract is suddenly worth far more pointed at GPUs than at hashing.
They already own the hard part: the building. Permitted sites, grid connections, cooling, and electrical buildout take years. Miners have them now — a head start measured in years over anyone breaking ground today.
The day job got harder. Bitcoin's halving keeps squeezing mining margins. Repurposing that power and real estate into contracted AI revenue is the obvious escape hatch.
The re-rate is the whole prize. A pure miner trades on the bitcoin price and hashprice — volatile, commodity, feast-or-famine. A company with contracted recurring revenue trades like an infrastructure business. Same company, very different valuation.
The honest split: who's really pivoting vs. who's just saying it
“We're pivoting to AI” is the most overused sentence in the sector right now, so the Lens job is to separate the ones with real contracts from the ones with a press release. A basket is in motion — HIVE/BUZZ (this deal), plus IREN, Core Scientific, Cipher, Applied Digital, TeraWulf — and four boring questions sort the winners from the noise:
| The question | Why it decides everything |
|---|
| Contracted revenue? | Real signed AI revenue (with named, creditworthy customers) vs. a roadmap slide. This is the scoreboard — check it every quarter. |
| Power cost & efficiency? | Cheap, clean power and a low PUE are the durable moat. The whole pitch is “convert cheap energy into intelligence.” |
| Cooling capability? | AI racks run hot — liquid cooling is table stakes now. A miner that can't cool a GB200 rack can't play. |
| Balance sheet? | GPUs cost a fortune. Who can fund them without crippling dilution? This is the hidden separator (HIVE funded part of this deal with a convertible note). |
The trap: a soaring stock on a vague “AI partnership” with no dollar figure, no named customer, and a balance sheet that needs constant share issuance to buy chips. The announcement pumps the stock; the economics never show up. Demand the contract, not the headline.
The Lens angle: it's the same bottleneck, one layer down
Step back and the miner pivot is the same trade as every other piece of the AI buildout, just wearing a crypto costume. It sits on four of our coverage spines at once:
- Sovereign AI. This exact deal is a Canadian stack — Bell's network, Cohere's models, kept on Canadian soil. The same “own your compute, don't rent it from a foreign giant” logic driving sovereign supply chains.
- The compute buildout. GB200 rack-scale capacity, supply-constrained. Every megawatt a miner converts is a megawatt of demand someone was scrambling to find.
- The thermal bottleneck. Liquid-cooled, efficiency-obsessed — the intelligence-per-watt thesis in the wild.
- Energy → intelligence. The literal business model is “turn cheap (often renewable) power into AI compute.” The miners with the cleanest, cheapest electrons win.
The insight that ties it together: don't buy the “crypto miner” label — price the
asset. The durable thing a miner owns isn't a hashrate; it's
cheap power plus a ready data center, which is the single scarcest combination in AI right now. The market is still valuing some of them on the old business while the new one signs contracts. That gap — old multiple, new revenue — is the entire opportunity. Same
picks-and-shovels lesson: own the bottleneck, not the brand.
The risks — named, not buried
- Dilution. GPUs are bought with debt and stock. A “great” AI contract funded by relentless share issuance can leave existing holders worse off. Watch the share count, not just the revenue.
- Execution. Mining a coin and running enterprise-grade AI uptime for a Bell or a hyperscaler are different sports. Not every miner can deliver the reliability a serious customer demands.
- Customer quality. A contract with Cohere/Bell is worth far more than one with a speculative crypto-adjacent tenant. Read who is paying.
- The pump. This is a hype-saturated corner of the market. The honest names will be the ones publishing contracted revenue numbers quarter after quarter — the fakers won't.
The viral take and the true take are rarely the same trade
Don't price the “crypto miner.” Price the cheap power and the ready data center.
Dragonfly Lens separates the miners with real AI contracts from the ones with a press release — and names the risks out loud. Plain English, every claim sourced and flagged.
Join the Lens →
Quick answers
Why are bitcoin miners pivoting to AI? Because they already own the two scarcest inputs in AI: cheap power and ready-built, grid-connected data centers. As bitcoin mining margins get squeezed by the halving, repurposing that power and real estate into contracted AI compute is far more valuable — and it re-rates the stock from a volatile commodity play to an infrastructure business.
Is the HIVE / Bell / Cohere deal real? Yes. HIVE's BUZZ HPC arm signed a ~3-year, ~$220M contract for AI compute (2,304 NVIDIA GB200 GPUs) serving Cohere on Bell Canada's network, expected to add ~$70M in annual recurring revenue, going live late-2026 to early-2027. The contract is real; the promotional language around it should be discounted.
Which miners are turning into AI compute companies? A basket is in motion — HIVE, IREN, Core Scientific, Cipher, Applied Digital, TeraWulf among them. What separates winners from press-release pivots: real contracted revenue with named customers, low power cost, liquid-cooling capability, and a balance sheet that can fund GPUs without heavy dilution.
Sources: HIVE / BUZZ HPC ~$220M, 3-year AI compute deal with Bell Canada + Cohere; 2,304 NVIDIA GB200 GPUs; ~$70M ARR; ~10% stock move; Merritt BC; live late-2026/early-2027 — CoinDesk, The Block, HPCwire.
Educational research, not personalized investment advice. Dragonfly Lens is not a registered investment advisor. Figures are as reported by the sources above. Company names illustrate a sector rotation, not buy recommendations — miner-to-AI pivots carry real execution, dilution, and hype risk; verify contracted-revenue claims against primary filings before acting. Past performance does not guarantee future results.