SAN ANTONIO, June 22, 2026, 12:05 CDT
- HIVE jumped around 20% on the Nasdaq after it announced Columbia University researchers trained AI models remotely on the company’s GPUs in Asunción, Paraguay, from New York.
- AI demand isn’t the only story in the market. Tests show older Nvidia A40 GPUs can still do some lower-end training work if matched with the right software, offering a possible lower-cost compute option beyond the big U.S. cloud hubs.
- The rally is happening with a financing overhang. HIVE’s June 16 filing says there’s still about $214.7 million left on its at-the-market stock-sale shelf, a tool that could fund data-center growth but might dilute current holders if tapped.
HIVE Digital Technologies shares surged Monday as the company said Columbia University researchers had validated its Paraguay GPU cluster for AI research. The news gave another sign of HIVE’s move into high-performance computing. HIVE was at $5.12 on Nasdaq, up around 20% from its previous close. More than 54 million shares changed hands during the session.
HIVE wants to show its mining sites can bring in more value running AI compute than just mining bitcoin. The company said researchers in New York completed training runs using GPUs stationed over 5,000 miles away in Asunción. HIVE called the test an early example of cross-border AI work that doesn’t have to happen in Northern Virginia, Silicon Valley or Toronto.
Utilization is still the problem. AI infrastructure players are scrambling for Nvidia H100 and Blackwell chips, but a lot of enterprise and research teams don’t always need top chips if they can get the right mix of software, networking, and energy costs. Columbia researchers said HIVE’s A40 hardware delivered results on par with H100 once hardware performance was leveled out in pretraining runs on models up to 1.4 billion parameters.
That doesn’t turn an A40 into an H100. But it does show HIVE could market a layer of compute just below the high-demand systems for top-end AI training, where price and supply might be just as important as max performance. The company is basically trying to take older GPUs and low-cost renewable energy and package that into an AI cloud service.
HIVE said trial data—token-per-second, latency and bandwidth—will set a baseline for its planned high-performance computing and AI “Gigafactory” in Yguazú, Paraguay. The company is building a 100 MW substation there, set to go live in September 2026. Data-center work is slated to begin in the fall, aiming for service in the second half of 2027. HIVE Digital Technologies
Frank Holmes, executive chairman at HIVE, said the trial was “an important step” and proved that high-performance computing “does not need to be limited by geography.” CEO Aydin Kilic added that the Columbia project worked to make AI training “smarter, faster, and more efficient” by working on the neural-network math engine. HIVE Digital Technologies
Columbia-related news lands after HIVE on June 18 locked in a separate contract that put fresh revenue behind its AI push. Subsidiary BUZZ HPC agreed to a three-year, $220 million GPU cloud deal connected to Bell AI Fabric and Cohere, using 2,304 Nvidia Grace Blackwell GPUs at Bell’s Merritt, British Columbia, site.
Kilic said deployment is expected late 2026 or early 2027, which would bring in about $70 million in contracted annual recurring revenue for BUZZ HPC, on top of the current $35 million. For HIVE, that’s a clearer catalyst than the Paraguay study. Signed AI revenue lets investors see a number on HIVE’s shift from pure mining economics.
The field is already busy. Listed miners and power-heavy data-center players like IREN, Core Scientific and TeraWulf are pitching their grid connections, land, and cooling chops as ways to break into AI and high-performance computing. That’s the kind of heavy compute power needed for training models, running inference, and tackling scientific tasks.
HIVE posted fiscal 2026 revenue of $297.8 million, but most of that—$278.3 million—still came from digital currency mining. HPC hosting brought in $19.5 million. Bitcoin mining revenue dropped again in the fiscal fourth quarter as bitcoin prices fell and network difficulty increased. That cut into mining margins fast.
Investors may be pricing in HIVE’s expansion before the assets are even operational. In its own statement, HIVE said there are risks that the rollout takes longer than planned, could cost more, might miss performance goals, or that AI infrastructure demand fails to materialize. The company doesn’t expect its Paraguay data center to go live until the second half of 2027.
There’s another piece moving under the surface rally. HIVE’s June 16 registration switched its shelf to a Form S-3 and left a $300 million at-the-market equity program running. The filing showed HIVE already sold 29.2 million shares for $85.3 million gross, so $214.7 million remains. At Monday’s $5.12, that unsold part equals about 41.9 million shares, or 15.5% of the 270.4 million shares outstanding in the filing.
HIVE’s trade here is clearer. The company has picked up a stronger AI-infrastructure story, with research work shown in Paraguay and a signed GPU cloud contract in Canada. But the same run in the stock that backs up the shift could open the door to new shares. For shareholders, that puts the next question beyond AI operations—it’s how much upside they keep.
