Bitfarms is moving away from its bitcoin mining roots and repositioning itself as a high-performance computing provider for AI companies. The shift came after the company posted a $46 million loss in Q3 2025, pushing management to pursue a new operating model that it says could be in place by Q4 2026, the same period when Nvidia’s Vera Rubin GPU generation is expected to ship.
Losses pushed the company toward a different business model
Long known as one of the largest publicly listed bitcoin miners, Bitfarms is now redirecting its efforts toward AI infrastructure. Management said on its earnings call that the company is seeking a more stable path and is moving resources toward building out capacity for HPC workloads.
To fund that transition, Bitfarms has lined up more than $588 million through a convertible note offering, along with a $300 million debt facility. The capital is intended to help rework its infrastructure so it can support AI data center demand rather than traditional mining activity.
Latam exit leaves Bitfarms with a North American footprint
The company has also reshaped its geographic exposure. Bitfarms recently agreed to sell its last Latin American site, located in Paraguay, for $30 million. After that sale, the company will operate with a fully North American energy footprint and focus on markets where AI workload demand is stronger.
That is a clear break from its earlier structure, where Latin American mining sites were part of the business. The new plan concentrates operations in areas management sees as more relevant for large-scale AI deployments.
Why Bitfarms is waiting for Vera Rubin instead of building around Blackwell
Many companies adapting facilities for AI are targeting Nvidia’s current Blackwell generation, which is already widely used across the industry, including by OpenAI. Bitfarms is taking a different route. It wants to be ready for Vera Rubin, the next generation that Nvidia CEO Jensen Huang recently said was already in production.
According to unconfirmed reports cited in the source material, Vera Rubin GPUs could carry a thermal design power of 2.3 kW per GPU. That would place heavier demands on data centers built for Blackwell loads. By comparison, second-generation Blackwell chips draw 1.4 kW in total. The gap matters because future deployments may require deeper changes in power delivery, cooling, and site design.
Washington site is central to the plan
Bitfarms is using its Washington site as the main point for this buildout. CEO Ben Gagnon said the location could “potentially produce more net operating income than we have ever generated with Bitcoin mining, providing the company with a strong cash-flow foundation.”
On the company’s Q3 2025 earnings call, Gagnon argued that timing could become the main advantage. If Bitfarms has infrastructure ready when customers receive Vera Rubin chips, it may be able to attract companies seeking rapid deployment. He also said much of the infrastructure being built now may not be compatible with the next generation, while buyers that commit large amounts of capital to Vera Rubin GPUs will have strong economic incentives to put those assets to work quickly, even at a higher infrastructure cost.
Whether that thesis holds will depend on demand arriving as expected and on Bitfarms becoming one of the first operators able to support Vera Rubin deployments at data center scale.