Key Highlights
- HIVE Digital stock climbed more than 22% Monday following Columbia University validation of its Paraguay AI computing infrastructure.
- Columbia researchers confirmed HIVE’s legacy Nvidia A40 GPUs delivered performance comparable to H100 chips for specific LLM training tasks.
- AI training experiments were executed remotely from New York on GPU hardware positioned over 5,000 miles away in Asunción, Paraguay.
- Findings from the study have been submitted to NeurIPS, a prestigious machine learning conference.
- HIVE anticipates energizing a 100MW electrical substation in Yguazú, Paraguay by September 2026, followed by Tier III data center construction.
HIVE Digital Technologies (HIVE) stock experienced a remarkable surge exceeding 22% during Monday’s trading session, climbing above the $7 threshold in morning activity after Columbia University researchers confirmed the capabilities of its Paraguay-based artificial intelligence GPU infrastructure.
HIVE Digital Technologies Ltd., HIVE
The academic study, conducted in partnership with Columbia’s Department of Industrial Engineering and Operations Research, demonstrated that HIVE’s Nvidia A40 graphics processing units deployed in Paraguay achieved performance levels equivalent to Nvidia H100 systems when handling particular large language model pretraining tasks — specifically for models reaching up to 1.4 billion parameters.
Columbia’s research team spent two months refining their algorithms for the A40 architecture before conducting comprehensive evaluations of throughput, latency, and bandwidth metrics. When adjusted for baseline hardware specifications, the A40 infrastructure produced results consistent with H100 systems for their targeted applications.
The researchers additionally performed serving throughput and latency evaluations on the 1.4 billion-parameter model, alongside conventional benchmark testing using LLaMA models.
The complete research findings have been submitted for consideration at NeurIPS — the Conference on Neural Information Processing Systems — recognized as one of the premier academic forums in the machine learning field.
Remote Training Across Continents
Among the most notable aspects of the research project: Columbia’s team in New York City successfully conducted repeated training sessions on GPU hardware physically located in Asunción, Paraguay — spanning a distance exceeding 5,000 miles.
Executive Chairman Frank Holmes emphasized that the findings demonstrate “high-performance computing does not need to be limited by geography.” CEO Aydin Kilic characterized the A40-H100 performance equivalence as a “powerful result” that confirms the company’s engineering-focused strategy.
The experimental data collected during this initiative will establish the technical foundation for HIVE’s upcoming AI infrastructure campus in Yguazú, Paraguay.
Paraguay AI Facility Progressing
HIVE is advancing development of a substantial artificial intelligence and high-performance computing campus in Yguazú. Civil engineering work on a 100-megawatt electrical substation has reached completion, with commissioning activities planned for this summer and full operational status projected for September 2026.
The company has scheduled Tier III data center construction to commence during fall 2026. HIVE projects the facility will become operational during the latter half of 2027.
The organization positioned the Columbia University collaboration as independent third-party verification supporting its strategic transition from Bitcoin mining operations toward the AI infrastructure sector.
Holmes added that the project also marks a milestone for Paraguay itself: “Paraguay has the power, the strategic location, and now the proof point.”
The 100MW substation commissioning represents the company’s next critical near-term objective on HIVE’s infrastructure development timeline.
