Frank Holmes on HIVE's $350 Million AI Deal and Why Bitcoin Miners Built the Rails for It
Bulldog Holmes, known to regulators as Frank Holmes, executive chairman of Hive Digital Technologies, joined Simply Bitcoin the day after his company announced a $350 million AI cloud contract. The conversation covered how a Bitcoin miner ends up building AI infrastructure, why the chips behave nothing like Bitcoin's ASICs, and what a "minister of AI" actually does in Canada.
THE $350 MILLION DEAL
HIVE's wholly owned subsidiary BUZZ HPC signed a five-year GPU cloud services agreement worth approximately $350 million in total contract value, adding about $70 million in annualized revenue. The dedicated cluster of 2,016 Nvidia Blackwell Ultra GPUs goes live later this year at the Bell AI Fabric facility in Merritt, British Columbia, powered by renewable hydroelectric energy.
The deal lifts BUZZ HPC's total annualized revenue to roughly $180 million, with about $35 million already realized and $145 million contracted and expected online through the fourth quarter of 2026. It is HIVE's second large GPU cluster deal in two months, following a three-year, $220 million sovereign AI cloud contract at the same Merritt facility for Bell and Cohere signed back in June.
Holmes framed the moment against Hive's Bitcoin mining roots. Last year, he said, the company's team focused on expanding its ASIC computing power and its tier-one data center footprint in Paraguay, work that grew Bitcoin operations several times over and lifted revenue by roughly 300%. This year that same discipline is going into the GPU buildout across Canada, coast to coast.
FROM BITCOIN MINER TO AI HYPERSCALER
Holmes credited HIVE's GPU expertise to an earlier, less glamorous chapter: mining Ethereum before it moved to proof of stake. When that ended, the company started renting out its idle compute power to wholesalers, a business that grew to roughly a million dollars a month before ChatGPT's public debut turned AI into the obvious next act. Holmes brought in Craig Devaris to scale that into what BUZZ HPC is now, alongside President and CEO Aydin Kilic.
The distinction Holmes drew is between renting out land and power to a hyperscaler under a long lease, which he called a REIT-style model many of HIVE's peers use, and running the compute business directly. HIVE chose the second path, which comes with thinner margins per megawatt on paper but a bigger share of the upside if GPU demand keeps compounding.
He pointed to the market's reaction as proof the thesis is landing. H.C. Wainwright reiterated a Buy rating and a $7.00 price target on HIVE following the announcement, noting the new deal brings the company's total contracted annual recurring revenue from its AI and HPC subsidiary to approximately $180 million. Shares rose about 13% in premarket trading the morning the deal was announced.
THE PARAGUAY BLUEPRINT, NOW IN CANADA
Holmes leaned on HIVE's Paraguay build-out as the proof of concept. The country exports beef and grain, he said, but has a surplus of hydroelectric power that cannot be shipped anywhere over a wire. A GPU chip solves that problem differently than an ASIC does: instead of turning electricity into Bitcoin that settles into a treasury account, it turns electricity into compute power that can be sent over ocean cable to wherever the demand sits. He described a University of Columbia researcher in New York City using GPU compute from a facility in Asuncion 5,000 miles away as the clearest example of that idea working in practice.
Canada gets the same treatment, he said, but with a different customer base. Where the obvious knock on the door in the US comes from OpenAI, in Canada it is Bell Canada as strategic partner and Cohere building the models on top of the infrastructure. Universities in Toronto and Waterloo sit at the center of HIVE's new data center plans, a deliberate bet on proximity to research talent rather than just cheap power.
WHY THE CHIPS DON'T LAST AS LONG
Asked about lifespan, Holmes said an ASIC chip is rated for roughly four years but HIVE has found it closer to two in practice, because Bitcoin's difficulty keeps rising and forces miners onto more efficient hardware faster than the rated life would suggest. GPU chips run the opposite way: HIVE wrote off its Nvidia chips from its Ethereum mining days to zero, only to see them regain most of their original value once the AI compute business took off, with GPU hardware now getting pushed toward five to seven years of useful life. That difference, he said, is what makes GPU infrastructure easier to finance with credit, since lenders can underwrite a shorter payback window against a longer-lived asset.
Holmes closed on the framing he keeps coming back to: Bitcoin miners spent years building out access to stranded, cheap power in places most of Wall Street ignored. The AI boom did not create that infrastructure. It found it already built.
This story comes from the Simply Bitcoin Live show. Watch the full episode.
