US DeepInfra Launches AI Data Center in Toronto, Canada

en.Wedoany.com Reported - US-based DeepInfra recently launched an artificial intelligence data center node in Toronto, Canada. The facility offers 1.7MW of capacity and plans to deploy over 1,000 NVIDIA Blackwell B300 GPUs. This project marks DeepInfra's ninth data center node in operation globally and its first expansion of AI computing infrastructure outside the US market.

The Toronto facility is primarily focused on AI inference workloads. Unlike centralized training of large models, inference requires long-term, continuous responses to user requests, with applications including agents, generative AI services, real-time interactions, and large-scale API calls. By deploying computing resources closer to Canadian customers and data, the facility can shorten data transmission paths for certain operations and reduce network latency associated with cross-border calls to US computing nodes.

The announced 1.7MW represents the facility's total infrastructure capacity, while the over 1,000 B300 GPUs indicate the planned deployment scale for this node. Public information has not yet confirmed whether all these GPUs have been fully installed, nor has it disclosed the number of server cabinets, power density per cabinet, or phased deployment schedule. Therefore, the planned scale cannot be directly equated to the computing capacity currently in service.

US DeepInfra owns and operates its own GPU computing infrastructure. Its platform currently supports over 200 open-source models and processes nearly 5 trillion tokens per week. Once operational, the Toronto node will join its existing data centers to form a distributed inference network, supplementing regional computing resources for enterprises deploying open models, proprietary models, and agent applications. The company stated that other international data center nodes are also under evaluation.

From a functional perspective, the core of this project is building a high-density GPU cluster and its supporting data center environment. Continuous operation of B300 servers requires coordination of power supply and distribution, cooling, network interconnection, and storage systems within the facility. However, the project team has not yet disclosed the cooling method, network architecture, storage configuration, or data center operations partner for the Toronto node. Based on available information, US-based Data Center Dynamics speculates that the project may utilize third-party data center capacity, but DeepInfra has not officially confirmed this, so it should not be described as an independently constructed data center building by the company.

The commissioning of this node signifies DeepInfra's infrastructure expansion from the US market into Canada. Subsequent priorities will focus on completing GPU deployment, expanding actual available inference capacity, and integrating the new node into its global model service platform. At this stage, what has been confirmed is the official launch of the 1.7MW facility and the deployment plan for over 1,000 B300 GPUs. Specific production ratios and equipment rollout timelines still await further disclosure from the company.