Designing Data Centers for 400kW GPU Racks | Researcher Conversations at GTC

Summary

Patrick Wohlschlegel, Head of Infrastructure at Radiant, discusses the company's newly launched AI cloud venture and its infrastructure roadmap. He details how the merger with Brookfield resolves key constraints such as land, power, and capital, enabling data center construction from greenfield to operational in 12 months. The conversation also covers behind-the-meter power generation, the Radiant Cloud OS software stack, and preparations for upcoming 400kW GPU racks and NVIDIA's Rubin architecture.

• Radiant is a newly launched Neo Cloud with a global infrastructure roadmap and access to several hundred megawatts of sites.
• The Brookfield merger provides behind-the-meter power, land, and capital, solving industry bottlenecks.
• They can go from greenfield site to fully productive AI environment in about 12 months.
• Radiant's software stack (Radiant Cloud OS) enables flexible provisioning from bare-metal to Kubernetes for varied customer needs.
• Data centers are being designed to accommodate 400kW GPU racks expected as soon as January next year.
• The team emphasizes reliability, observability, and high SLAs to keep enterprise customers satisfied.
• Discussion addresses the balance between serving large-scale training customers and smaller inference-focused clients with the same platform.