Tuesday, July 14, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeCompute & CloudReport
Compute & Cloud · Report

DeepInfra opens 1.7 MW Toronto AI inference data center with 1,000+ NVIDIA B300 GPUs.

Neocloud GPU capacity expansion in North America; adds competitive supply outside hyperscaler control.
Trade pressSlicast · July 14, 2026 · US · Source: Google News
importance 70

DeepInfra, a Palo Alto-based GPU inference platform, has opened a 1.7MW data center in Toronto housing more than 1,000 NVIDIA Blackwell B300 GPUs. Announced July 8, the facility marks the company's first deployment outside the United States and its ninth data center overall.

The Toronto cluster occupies leased capacity at an undisclosed third-party colocation facility. DeepInfra did not name the colocation partner or disclose specific cooling or networking configurations. The deployment is designed to reduce inference latency for Canadian and international enterprise customers while serving enterprises with data residency requirements that mandate processing within Canadian borders.

Founded in 2022, DeepInfra operates a managed inference platform supporting more than 200 open-source AI models and processing nearly five trillion tokens per week. The company holds SOC 2 and ISO 27001 certifications and maintains a zero data retention policy for enterprise customers. The expansion comes two months after DeepInfra closed a $107 million Series B funding round in May 2026, building on an $18 million Series A raised in April 2025. The company has not disclosed its total valuation or annual revenue.

The B300 is NVIDIA's latest Blackwell-architecture accelerator, designed for higher inference throughput and improved energy efficiency compared to previous Hopper-generation hardware. DeepInfra owns and operates its GPU infrastructure but leases the underlying data center space from its colocation partner.

The Toronto deployment reflects a broader industry shift from training-dominated GPU demand toward inference workloads. McKinsey research cited in DeepInfra's announcement projects inference will account for more than 40% of total data center demand by 2030, growing at a roughly 35% compound annual rate. For GPU suppliers, this trend is significant—NVIDIA's Blackwell architecture is positioned as the primary platform for next-generation inference infrastructure. NVIDIA shares traded at $203.53 on July 11, giving the company a market capitalization of approximately $4.9 trillion.

DeepInfra competes in the managed inference market alongside players like Together AI, Fireworks AI, and Groq. The company indicated that additional international data center deployments are under evaluation, though no specific locations or timelines were disclosed.

Read the original
DeepInfra opens 1.7 MW Toronto AI inference… · Slicast