Bitcoin miner HIVE Digital Technologies has arranged an approximately $129 million, fixed-rate loan to help buy NVIDIA GB200 systems for a Canadian artificial-intelligence cluster. Its November 9, 2026 financial update describes 2,304 GPUs serving a three-year contract involving Bell AI Fabric and Cohere. The financing is signed, but HIVE expected hardware shipment before year-end; the planned cluster should not be counted as commissioned AI capacity or revenue already received.
HIVE and BUZZ HPC discuss sovereign AI infrastructure with VAST Data in an earlier official company session. Historical technical context, not footage of the new deployment. Watch on YouTube.
2,304 GPUs equal 32 complete GB200 racks
NVIDIA’s GB200 NVL72 combines 72 Blackwell GPUs and 36 Grace CPUs in a liquid-cooled rack connected through high-speed NVLink switching. At that reference density, 2,304 ÷ 72 = 32 rack-equivalents, or 1,152 Grace CPUs if delivered in full NVL72 configurations. HIVE describes GB200 NVL72 systems and Quantum InfiniBand interconnects; actual rack installation, electrical loading and cooling acceptance must still be verified.
One disclosure discrepancy deserves attention: HIVE’s October 27 financing release refers once to 2,303 GPUs, while its June customer announcement and November earnings release specify 2,304. The rack calculation uses the repeated 2,304-GPU contract figure, not an independently audited equipment inventory.

What the $129 million loan buys
The approximately $165 million hardware package is funded primarily through debt. HIVE’s October agreement specifies a 9% fixed annual rate, a three-year term and monthly principal-and-interest payments secured by the GPUs and related assets. The borrower is BUZZ Alpha Cloud; its parent provides a limited-recourse guarantee. HIVE says its contribution to the remaining cost comes from previously issued zero-coupon convertible financing.
Using the later SEC filing’s $128.9 million principal and conventional equal monthly amortization, the indicated payment is about $4.10 million monthly, with approximately $18.7 million of interest over 36 months. The calculation is illustrative; actual draw timing, fees and escrow provisions affect the contractual cash flow. HIVE’s filing says proceeds are released to the supplier after equipment acceptance, when repayments begin.
Proactive interviews HIVE executive chairman Frank Holmes about the company’s AI data-center strategy. The earlier interview does not verify November hardware delivery. Watch on YouTube.
$225 million contracted is not $225 million earned
HIVE now values its three-year agreement supporting Cohere at approximately $225 million total contract value, including GPU compute plus connectivity and storage. An even division implies $75 million per year, but contract value is not a forecast of equal quarterly receipts or operating profit. Hardware depreciation, financing, electricity, cooling, staffing, network transport and customer acceptance all affect profitability.
Its November results distinguish roughly $40 million of live GPU-cloud annualized revenue from $185 million of contracted annualized revenue, which includes capacity yet to start service. For the September quarter, BUZZ recorded only $8.8 million of actual HPC revenue. Our earlier HIVE earnings report documents the contrast with mining revenues and ASIC depreciation.
Why a Bitcoin miner is buying an AI factory
HIVE still runs large Bitcoin mining operations, including hydro-powered sites, while BUZZ builds a different product: reserved, high-bandwidth GPU compute for enterprise models. Experience operating power-dense facilities transfers across businesses, but AI customers also require specialized cooling, fast interconnects, resilient networking and stricter service guarantees. The QIT container proposal and Keel’s Washington rebuild show two different paths from mining infrastructure to AI halls.
The transition is not universal. HIVE previously discussed winding down Swedish miners as AI plans advanced, yet its Paraguay mining capacity remains central to the business. The next test is evidence of 32 configured racks actually delivered, powered, networked and accepted by the customer. Only then can the financed hardware become a measured AI-service operation.
Primary sources: HIVE November 9 results; October 27 financing announcement; SEC Form 10-Q, subsequent events; NVIDIA GB200 reference architecture.
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiative, please scan the QR code below.


Leave a Reply