HIVE Digital surges 25% as Columbia validates Paraguay AI GPUs

Columbia University researchers trained neural networks on HIVE Digital's Paraguay GPUs, validating the crypto miner's pivot to AI infrastructure.

HIVE Digital Technologies' Nvidia A40 GPUs in Paraguay matched the performance of H100 systems on large language model pretraining tasks, according to a Columbia University study submitted to the NeurIPS 2026 conference.

"The findings demonstrate that high-performance computing does not need to be limited by geography," Executive Chairman Frank Holmes said in a statement.

The research team spent two months optimizing code for the A40 architecture before running training cycles on models up to 1.4 billion parameters. The experiments were executed remotely from New York City on hardware located more than 5,000 miles away in Asunción, Paraguay. When adjusted for baseline hardware capabilities, the A40's throughput and latency metrics corresponded closely with prior H100 benchmarks for the same application.

HIVE shares surged 25% to above $7 on Monday, pushing the stock to a four-month high. The company, which began as a Bitcoin miner, is targeting $140 million in annual recurring AI cloud revenue by the fourth quarter of 2026, with a broader goal of $225 million in total high-performance computing revenue.

Paraguay Infrastructure Buildout Accelerates

The Columbia validation comes as HIVE advances construction of its Yguazú, Paraguay AI computing campus. Civil work on a 100-megawatt electrical substation is complete, with full energization scheduled for September 2026. A Tier III data center facility is expected to begin operations in the second half of 2027.

CEO Aydin Kilic described the A40-H100 performance equivalence as a "powerful result" that confirms the company's engineering strategy. The experimental data from the Columbia collaboration will form the technical foundation for the Yguazú campus, according to the company.

HIVE has also announced plans for a separate 320-megawatt AI data center campus in Ontario, Canada, through its BUZZ HPC subsidiary. The project, located near the Toronto-Waterloo technology corridor, is expected to require an initial investment of about CAD $3.5 billion and support more than 100,000 GPUs for AI workloads.

From Bitcoin Mining to AI Cloud

HIVE was among the earliest publicly listed Bitcoin miners but has steadily repositioned toward AI and high-performance computing as mining margins compressed after the 2024 halving. The company produced 885 Bitcoin in its fiscal third quarter ended Dec. 31, 2025, while scaling installed hash rate to 25 exahashes per second.

Revenue reached $93.1 million in that quarter, up 219 percent from a year earlier, with gross operating margin expanding more than sixfold to $32.1 million. In February, HIVE signed a two-year, $30 million contract to deploy 504 Nvidia B200 GPUs for enterprise AI cloud services, a deal expected to contribute about $15 million in annual recurring revenue.

Other publicly traded miners including IREN and TeraWulf have made similar moves toward AI workloads, reflecting a broader view that data center infrastructure for artificial intelligence offers a more stable revenue base than crypto mining alone. HIVE's dual-engine model uses Bitcoin mining as a cash-flow generator while building recurring revenue from AI and high-performance computing contracts.

HIVE shares, which have more than doubled from their January lows, trade with a market capitalization of roughly $1.5 billion. The Columbia validation provides independent third-party support for the company's claim that its GPU infrastructure can compete with far more expensive hardware for specific AI workloads. The next milestone is the September 2026 substation energization in Paraguay, which will determine whether HIVE can scale its proof of concept into a commercially meaningful AI cloud business.

This article is for informational purposes only and does not constitute investment advice.