HIVE Digital Technologies Secures $220M GPU Cloud Deal as Bitcoin Miner Pivots to AI

HIVE Digital Technologies Closes Landmark AI Infrastructure Deal

HIVE Digital Technologies announced a three-year $220 million GPU cloud contract with Bell Canada and Toronto-based AI firm Cohere, marking the company's most significant move yet in its transition from Bitcoin mining to AI infrastructure 1

1

. HIVE stock climbed more than 7% on the TSX following the announcement, with shares jumping 11.8% in premarket trading 5

5

.

The deal is delivered through HIVE's subsidiary, BUZZ High Performance Computing, and involves deploying 2,304 NVIDIA Grace Blackwell GPUs at Bell's purpose-built data center in Merritt, British Columbia 1

1

. These chips are designed specifically for frontier AI model training and inference, positioning the infrastructure to handle demanding enterprise and government workloads.

Cohere, which builds large language model systems for enterprises and governments, will use this compute layer to run its platform for Canadian clients 1

1

. The contract aligns directly with Canada's Sovereign AI Compute Strategy, which has committed over $2 billion to domestic AI compute, including $240 million invested directly into Cohere 1

1

. Sovereign AI refers to AI that runs on infrastructure inside a country's borders using locally controlled data, which matters significantly when clients are government agencies. Bell Canada and Cohere had an existing partnership dating to July 2025, and this contract provides the compute infrastructure underneath that relationship 1

1

.

Bitcoin Miner Transition to AI Accelerates Revenue Targets

Once the deployment goes live between late 2026 and early 2027, HIVE expects approximately $70 million in new annual recurring revenue on top of the $35 million it already books from existing GPU operations 1

1

2

2

. The company's contracted high-performance computing revenue target now exceeds $100 million 2

2

. HIVE is funding the purchase of the NVIDIA Grace Blackwell systems using proceeds from its $115 million convertible note financing completed in April 2

2

.

For HIVE, which reported $278.3 million in Bitcoin mining revenue in its last quarter, this represents the latest chapter in a pivot underway since 2022 1

1

. The company began its AI shift by redirecting GPU capacity from crypto mining, landing a deal with Dell for new GPUs last November 1

1

. Earlier this month, HIVE reported that revenue from its HPC division increased to $19.5 million in fiscal 2026, nearly doubling from a year earlier 2

2

. Meanwhile, the company's Bitcoin treasury holdings fell to 150 BTC from 481 BTC a quarter earlier 2

2

.

Why Bitcoin Miners Are Racing Toward AI Data Center Infrastructure

Crypto mining returns are volatile and get harder as more miners compete for block rewards, with profitability falling to record lows 1

1

. Bitcoin mining difficulty fell 10.09% on June 14, one of the largest downward adjustments in the network's history, attributed to weaker mining economics, Bitcoin's price decline, and miners dedicating power to AI and high-performance computing projects 2

2

. In contrast, AI compute demand is growing fast and clients, especially government agencies, sign multi-year contracts at locked-in rates 1

1

.

HIVE is not alone in this strategy: Keel Infrastructure, formerly Bitfarms, sold off its last Paraguay mining facility in April and is following the same playbook 1

1

.

HIVE also has a larger project in development: a 320-megawatt AI data center campus in the Greater Toronto Area designed to house more than 100,000 NVIDIA GPUs at full build-out 1

1

. The company expects this facility to generate roughly $360 million in annualized recurring revenue at full operation and has set a broader target of $660 million in annualized HPC revenue by the end of 2028 1

1

.

Columbia University Research Validates GPU Performance in Paraguay

In a separate development, HIVE reported that its GPU cluster in Asunción, Paraguay supported its first academic research project through a collaboration with the Department of Industrial Engineering and Operations Research at Columbia University 3

3

. The work has been submitted to NeurIPS, considered one of the three most influential global conferences in machine learning and AI 3

3

. According to researchers, HIVE's older-generation NVIDIA A40 graphics processors in Paraguay delivered performance comparable to NVIDIA's flagship H100 chips for specific AI training workloads after two months of code optimization 4

4

.

Researchers successfully trained large language models with up to 1.4 billion parameters remotely on HIVE's GPU cluster while operating from New York, more than 5,000 miles away 4

4

. "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," a Columbia University researcher said 4

4

. The research provides an important proof point for HIVE's planned 100-megawatt AI and high-performance computing campus in Paraguay, where civil construction on a substation has been completed with energization expected later this year 4

4

.