HIVE completes AI research project with Columbia University using Paraguay GPUs

HIVE shares climbed about 25% to around $5.20, a seven-month high, after Columbia University-backed AI results and a separate C$220 million, three-year cloud deal tied to Bell Canada and Cohere boosted its pivot beyond Bitcoin mining.

BTC

35d ago

Fact Check

HIVE's official press release confirms the completion of an AI research project with Columbia University using A40 GPUs in Asunción, Paraguay, with A40 GPUs matching H100 performance when normalized for raw hardware, focused on LLM pretraining up to 1.4B parameters and submitted to NeurIPS. Independent reporting from The Block and CryptoBriefing corroborates the substance, including the stock surge. The Block reports a ~25% surge and other coverage notes 18%; the claimed 'more than 22%' falls within the reported range. The only minor nuance is the H100 comparison being 'when normalized for raw hardware capabilities' rather than absolute, which the claim approximates with 'neared H100-level results on some AI pretraining tasks.' Core claim is well-supported.

Summary

HIVE Digital Technologies shares rose about 25% on June 22 to around $5.20, their highest level in seven months, as investors responded to Columbia University-backed research on the company’s Paraguay GPU infrastructure and a recently announced C$220 million, three-year AI cloud contract linked to Bell Canada and Cohere. The Columbia study found HIVE’s software-optimized Nvidia A40 GPU cluster in Paraguay approached H100-class performance on certain roughly 1.4 billion-parameter AI pretraining tasks, reinforcing the Bitcoin miner’s effort to reposition itself as AI computing infrastructure provider. The rally also reflects growing market support for HIVE’s broader diversification strategy, including its planned Yguazú high-performance computing campus targeted to begin operating in the second half of 2027.

Terms & Concepts

GPU cluster: A group of graphics processors linked together to handle computing workloads such as AI model training.
Pretraining tasks: Early-stage model training runs in which an AI system learns broad patterns from large datasets before later fine-tuning.
High-performance computing campus: A large site built to host advanced computing infrastructure for intensive workloads such as AI and data processing.