Mr. Aydin Kilic reports

HIVE'S PARAGUAY AI INFRASTRUCTURE PERFORMANCE VALIDATED IN COLUMBIA UNIVERSITY STUDY, RESEARCH HEADS TO NEURIPS

Hive Digital Technologies Ltd. has completed its inaugural research project using Hive GPUs for artificial intelligence research purposes in Asuncion, Paraguay, in collaboration with the department of industrial engineering and operations research at Columbia University in New York.

This research has been submitted to the Conference on Neural Information Processing Systems, one of the leading machine learning and computational neuroscience conference held annually in December. Along with ICLR and ICML, it is one of the three primary conferences of high impact in machine learning and artificial intelligence research globally.

This establishes a proof of concept for intercontinental AI training, where researchers in New York ran iterative training runs on GPUs located in Asuncion, Paraguay. With these data, Hive has a reference point for the performance of AI workloads using GPUs in Asuncion. Using code optimizations developed by the Columbia team, the research found that Hive's A40 GPUs matched the performance of newer-generation H100 GPUs.

A researcher from the department of industrial engineering and operations research at Columbia University said: "We study neural network pretraining using optimization theory over general geometry and under large noise. We design and analyze an accelerated algorithm that matches the performance of the current leading method, Muon, in both theory and practice.

"As shown in the paper, our work on these nodes focuses on the performance of pretraining algorithms. Over the past two months, we optimized our code for the A40s and tested the throughput and latency of Muon and our variants. In our use case of pretraining LLMs of up to 1.4B parameters, our results match those of H100s after normalizing for each hardware's raw performance.

"Additionally, we tested the serving throughput and latency of our 1.4B model, which has not yet been heavily optimized, and included standard throughput and latency tests for LLaMA models as well."

Using the measured token-per-second, latency and bandwidth data as a baseline, Hive has established a foundation for an HPC/AI gigafactory in Yguazu, Paraguay, where the company has a 100-megawatt substation under construction. Civil works are complete, with commissioning expected this summer. The substation is expected to be energized in September, 2026, with construction on a new Tier 3 data centre beginning fall 2026 and a ready-for-service date in second half 2027.

Leadership commentary

Frank Holmes, executive chairman of Hive, stated: "This proof-of-concept is an important step in Hive's mission to help bring advanced AI computing infrastructure to Paraguay. Seeing professors from Columbia University in New York City remotely utilize our AI-optimized GPU cluster in Asuncion, more than 5,000 miles away, demonstrates the power of distributed AI infrastructure. It shows that high performance computing does not need to be limited by geography. Our vision is to bring the right power, data centre design, software stack and execution, so, through Hive, Paraguay can participate directly in the global AI economy. This milestone further validates the commercial potential of Hive's AI platform and our ability to deliver compute across borders. This is also an important moment for Paraguay. Paraguay has the power, the strategic location and now the proof point. Hive is proud to help bring that future on-line."

Aydin Kilic, president and chief executive officer of Hive, added: "At Hive, innovation is in our DNA. That is why this work with the AI researchers at Columbia University has been so exciting. Their research is focused on next-generation neural network pretraining. In simple terms, they are working on making AI training smarter, faster and more efficient by improving the mathematical engine behind how neural networks learn.

"Over the past two months, the Columbia team optimized their code for Hive's A40 GPU nodes in Asuncion across advanced AI workloads. In their specific use case, after normalizing for each hardware platform's raw performance, the results on Hive's A40 nodes matched those observed on H100 systems. That is a powerful result.

"For Hive, this validates a core belief: Great engineering can unlock significant value. Our history includes building the BuzzMiner in collaboration with Intel Corp. and also becoming one of the largest demand-response participants in Sweden, helping balance the national electrical grid. We will continue to invest in communities while bringing the advancements and innovations in our tech stack to our data centres globally. We see a bright future for AI in Paraguay."

An assistant professor in the department of industrial engineering and operations research at Columbia University added: "We are grateful for the support that made this research possible. This work advances our understanding of modern neural network optimization, including matrix-aware optimizers such as Muon and related scale-invariant methods. By clarifying their theoretical foundations and evaluating them in practical neural network training settings, this research highlights their potential relevance for future LLM pretraining. We look forward to sharing these findings with the broader research community."

About Hive Digital Technologies Ltd.

Founded in 2017, Hive was among the first publicly listed companies to prioritize mining digital assets powered by green energy. Today, Hive builds and operates next-generation Tier 1 and Tier 3 data centres across Canada, Sweden and Paraguay, serving both bitcoin and high performance computing clients. Hive's dual engine infrastructure -- driven by hash rate services and GPU-accelerated AI computing -- delivers scalable, environmentally responsible solutions for the digital economy.

We seek Safe Harbor.