Bleeding Edge launches Latin America's first Blackwell-based AI Factory in Mexico

This launch establishes a localized, high-performance infrastructure hub for Latin American AI development, reducing reliance on distant northern data centers. By integrating both the physical facility and the cloud service layer, Bleeding Edge bypasses the supply chain bottlenecks typical of traditional multi-vendor deployments. For the streaming and digital media ecosystem, this provides the specific GPU density required for real-time generative video and sophisticated recommendation inference at the local edge. Watch for the performance metrics of the Blackwell B300 systems at this scale to determine if modular 'AI Factories' become the preferred blueprint over traditional hyperscale builds in emerging regions.

The Querétaro region has emerged as Mexico's primary data center hub, currently hosting roughly 65% of the country’s installed IT capacity, including projects from AWS, Microsoft, and Google Cloud. According to the Mexican Association of Data Centers (MEXDC) in November 2025, the industry is projected to capture over $18 billion in direct investment by 2030, adding 1,516MW of power capacity to meet surging demand for AI and Edge computing. This growth is increasingly focused on high-density cooling; analysts at Precedence Research reported in July 2026 that the Latin American liquid cooling manifold market is expected to grow at a nearly 30% CAGR through 2035.

The deployment of Nvidia’s Blackwell B300, also known as 'Blackwell Ultra,' represents a significant leap in power density. Per internal tracking from Spheron in April 2026, the B300 delivers up to 15 petaFLOPS of dense FP4 compute—a 67% increase over the B200—but draws 1,400W per GPU. This intensity requires the specialized liquid-cooled infrastructure Bleeding Edge has implemented in its 'QRO1' campus. Furthermore, per Nvidia reports from mid-2026, the Blackwell architecture has become the standard for 'AI Factories,' as it allows for massive context windows and multi-step reasoning agents that were previously bottlenecked by the H100 generation's memory limits.