HIVE Digital Technologies completed a cross-continental AI training project between its Nvidia A40 cluster in Asunción, Paraguay and Columbia University researchers based in New York, roughly 5,000 miles away, the company said.

The collaboration ran from March 2026 to June 2026 and paired HIVE's high-performance computing infrastructure with Columbia's Department of Industrial Engineering and Operations Research. The work focused on pre-training large language models and produced results the partners say were strong enough to support a submission to the NeurIPS conference, one of the field's most selective venues.

HIVE, which trades on the TSX and Nasdaq under the ticker HIVE, originally built its infrastructure for Bitcoin mining. The Paraguay deployment sits on 300 MW of hydroelectric capacity across the Valenzuela and Yguazú sites, drawing on the country's Itaipu Dam-fed grid.

What did the researchers actually accomplish?

Columbia's team spent roughly two months optimizing workloads for HIVE's A40 hardware. According to the company, the older-generation GPUs then delivered performance comparable to Nvidia's H100 accelerators under specific conditions tied to pre-training runs of up to 1.4 billion parameters.

That comparison carries commercial weight. The H100 dominates frontier AI training and trades at multiples of the A40's price. If software optimization can close part of that gap on selected workloads, operators holding older silicon gain additional runway before forced upgrades.

What did the project measure?

The Columbia tests generated empirical data on long-distance AI training. Latency and throughput decide whether compute transmitted over fiber remains viable at scale. The findings give HIVE and its peers measured numbers rather than theoretical estimates for cross-continent workloads.

HIVE framed the work as a transition from energy transmission to compute transmission. Executive Chairman Frank Holmes described the project as a milestone in that shift.

Why Paraguay, and what is HIVE building next?

The company operates 300 MW of hydroelectric capacity in Paraguay. A new 100 MW substation at Yguazú is set to energize by September 2026, and construction on a new data center at the same site is projected to begin in Fall 2026.

HIVE plans to use the Columbia results to guide commercial scaling of its high-performance computing business in Paraguay through 2027.

How does this change HIVE's market position?

HIVE is repositioning itself as a sustainable digital infrastructure operator rather than a pure cryptocurrency miner. AI training demand has absorbed much of the data-center industry's available capacity, pushing hyperscalers and specialist operators to look for power-rich, lower-cost jurisdictions.

Paraguay's hydroelectric surplus and HIVE's installed capacity fit that brief. The September 2026 substation energization and the Fall 2026 data center groundbreaking are the next operational markers for the thesis.

The NeurIPS submission adds a credibility signal. Publication or peer review of the latency and throughput findings would give HIVE third-party validation of its long-distance training approach, a meaningful asset when pitching enterprise HPC contracts.