HIVE's stock jumped 13% after its BUZZ HPC subsidiary signed a five-year, $350 million AI cloud deal with an unnamed investment-grade enterprise customer, the biggest AI contract in the compa
HIVE's stock jumped 13% after its BUZZ HPC subsidiary signed a five-year, $350 million AI cloud deal with an unnamed investment-grade enterprise customer, the biggest AI contract in the company's history.
The agreement, announced Aug. 17, covers a dedicated cluster of 2,016 Nvidia Blackwell Ultra GPUs at HIVE's facility in Merritt, British Columbia, which runs on hydroelectric power with closed-loop liquid cooling. It adds roughly $70 million in annualized revenue and lifts BUZZ's contracted run rate to about $180 million, HIVE's second major GPU deal at the site in two months, after a three-year, $220 million sovereign AI contract signed in June.
Executive chairman Frank Holmes returned to TheStreet Roundtable to explain how a Bitcoin miner beat the hyperscalers to it.
For Holmes, the deal is the payoff of last year's unfashionable grind, when HIVE quadrupled its Bitcoin mining operation so the cash flow could carry an AI pivot.
"Last year was about scaling up to 2% of the Bitcoin ecosystem. This year is to scale to be a significant high-performance computing player, the largest hyperscaler in Canada," he said.
Asked why a large enterprise would pick HIVE over an established hyperscaler, Holmes pointed to HIVE's record in an entirely different, yet related, business.
A year ago, HIVE's focus was Bitcoin mining, where it expanded at breakneck speed, from 6 to 26 exahash, roughly a fourfold increase in a single year, at a time when much of the industry was retreating from mining in part or in whole. Hash rate is the total computing power on the Bitcoin network; a higher share means more rewards for the operator.
The expansion shows up in the numbers. HIVE reported fiscal 2026 revenue of $297 million, up 158% year over year, an operating cushion that can sustain an investment-heavy high-performance computing buildout.
"That economies of scale has allowed us to make money every month," Holmes said. "Even the worst month this year, (which) was February — we still made money every day."
That consistency is the pitch to an AI customer. A company renting a dedicated GPU cluster needs the machines running around the clock, and Bitcoin miners who have operated at scale have, in effect, performed a years-long public audition for exactly that. Miners live and die on uptime and energy efficiency, and the metrics are visible to anyone who cares to look.
"Other service providers look at your uptime and your efficiency as a Bitcoin miner. We're always one and two," Holmes said. "But we function over nine time zones, three continents, five languages — and we still are more efficient than a Bitcoin miner only in Texas."
Data center construction has always been a capital problem. Until recently, only a handful of banks would lend for these type of projects, and in Canada, Holmes says the market “basically didn’t exist.”
In the last year, investors have piled in en masse. Private credit especially have begun writing huge checks to almost anyone asking for it.
"All these pools of capital, from BlackRock, Blackstone, that are creating these credit funds. Goldman Sachs has credit funds, ten billion dollars," he said. "And they're lending against GPUs."
Lending against GPUs has become the basis for HIVE’s AI pivot. ASIC chips, which are specialized chips designed for Bitcoin mining, are wasting assets with almost no salvage value.
“After four years they're worth nothing," Holmes explained.
This is because the Bitcoin network’s mining difficulty rises relentlessly and only the most efficient chips, paired with cheap energy, can keep margins up. For that reason, lenders don’t want these kinds of chips as collateral.
GPUs are different. They are general purpose, in global shortage, and productive across a variety of different workloads. A Bitcoin miner that once financed depreciating ASICs out of its own cash flow can not borrow against Nvidia hardware the same way a landlord borrows against a building.
HIVE has seen this play out in its own business before. The Nvidia chips it bought four years ago recouped about 80% of their cost mining Ethereum, and when Ethereum changed to proof-of-stake, it ended the business overnight but the hardware simply changed jobs.
The $350 million deal is what that trajectory looks like when it lands a marquee customer: roughly $185 million of capital spending against $70 million a year in contracted revenue implies the hardware pays for itself in under three years of a five year contract. After that, HIVE keeps the GPUs and whatever revenue generating life remains in them.
The structural shift Holmes is describing outlasts any single deal. When the world’s biggest asset managers decide GPUs are bankable collateral, the companies holding powered sites and long operating records become borrowers on terms Bitcoin mining couldn’t achieve.
HIVE spent nearly a decade learning to run computers cheaply in difficult places, and credit markets just decided that skill is worth financing.
