HIVE and Columbia University Test Long-Distance AI Training

HIVE Digital Technologies has successfully demonstrated the feasibility of long-distance AI research, connecting its GPU cluster in Paraguay with researchers at Columbia University in New York. The project, conducted over roughly two months from March to June 2026, tested whether AI training could effectively occur when the hardware and researchers are separated by about 5,000 miles.

HIVE's cluster of Nvidia A40 GPUs in Asunción, Paraguay, was used remotely by Columbia's Department of Industrial Engineering and Operations Research to run pre-training workloads for large language models (LLMs). After optimizing their code, the team achieved performance comparable to Nvidia's H100 chips under specific conditions, particularly for workloads involving up to 1.4 billion parameters. The project also yielded sufficient data for a submission to the prestigious NeurIPS conference.

Paraguay's abundant renewable hydroelectric power, primarily from the Itaipú Dam, supports HIVE's operations. The company currently runs 300 MW of hydroelectric capacity and plans to energize a new 100 MW substation in Yguazú by September 2026. Executive Chairman Frank Holmes highlighted the shift from energy transmission to compute transmission, marking a significant milestone.

Originally a Bitcoin mining company, HIVE has evolved into a sustainable digital infrastructure provider. The remote research project is intended to guide the commercial scaling of its high-performance computing (HPC) business in Paraguay through 2027. Construction on a new data center in Yguazú is expected to begin in Fall 2026.