HIVE, AI Cloud First Proof of Concept Launched in Paraguay... Columbia University LLM Learning Initiative Launched

A research team from Columbia University in the United States has connected to Paraguay’s operational GPU AI cloud in South America to initiate large-scale language model training. This marks the first empirical phase of HIVE Digital Technology’s “AI Infrastructure” expansion strategy. It signifies that the attempt to combine renewable energy-based “AI Cloud” with high-performance computing has entered a stage of validation with actual data.

On the 18th, HIVE Digital Technology announced that its “BUZZ AI Cloud” in Asunción, Paraguay, is now officially operational. The infrastructure consists of a GPU cluster deployed within the largest local Tier 3 data center operated by a major telecom provider, designed for AI model training, inference, and high-load research tasks. The company plans to validate the performance of connecting GPU computing between New York and Asunción, laying the foundation for future large-scale AI infrastructure expansion.

The core of this project involves the participation of Columbia University researchers. They are conducting pre-training of large language models (LLMs) and foundational model training on the BUZZ Cloud, while developing algorithms to improve model performance and computational efficiency. Notably, by applying Muon and MuonClip technologies, they achieved approximately 1.3 times efficiency improvement over previous levels. The related research is scheduled to be published in the academic journal TMLR. Starting with models ranging from 200 million to 20 billion parameters, the project has now expanded to over 80 billion parameters and is conducting multi-GPU distributed training.

The accumulated data on latency, throughput, and workload management during this process will directly influence HIVE’s mid- to long-term “HPC Strategy.” The company plans to use this empirical data to gradually expand Tier 3 data centers within Paraguay before 2027, especially considering infrastructure additions centered around the Iguazu region.

Paraguay’s emergence as an “AI Infrastructure” hub is supported by its abundant hydroelectric resources and nationwide fiber optic network. HIVE has secured 300 MW of renewable energy and is planning an additional 100 MW expansion. This is said to provide a low-cost, high-efficiency computing environment capable of handling overseas demands, including those from North America.

Chairman Frank Holmes stated, “Paraguay’s business, which started with Bitcoin mining, is now at a turning point toward AI and HPC expansion,” adding, “Data from actual GPU workloads will become a core standard for future infrastructure investment decisions.” CEO Aiden Killick also commented, “The participation of Columbia University’s research team is an important signal of platform competitiveness,” and “The AI computation validation between New York and Asunción will officially expand the ‘GPU AI Cloud’ business.”

Market observers note that Paraguay, with its stable political environment and friendly foreign investment policies, is poised to grow into a South American “AI Hub.” Its combination of eco-friendly power and high-speed networks aligns with the global AI industry’s infrastructure diversification strategies.

Comment: HIVE’s empirical efforts can be seen as an early stage beyond simple data center operations, validating the profitability of a new business model based on renewable energy-powered “AI computing.” In the future, depending on actual customer demand and investment capacity, the landscape of the South American AI infrastructure market may undergo significant changes.