The Cloudcast Data Center Buildouts for AI
10 snips
Aug 14, 2024 
Scott Jarnagin, Founder and CEO of Caddis Cloud Solutions, dives into the evolving landscape of data centers in response to the AI boom. He discusses how current infrastructures must transform to meet rising power and cooling demands. Scott elaborates on the vital strategic decisions enterprises face while planning their AI needs, highlighting the importance of choosing the right data center. He also touches on the involvement of nuclear energy in data center expansion and the implications of government regulations. A must-listen for tech enthusiasts!
AI Snips
Chapters
Transcript
Episode notes
AI Is Radically Increasing Rack Power Density
- AI has dramatically increased per-rack power needs, jumping from ~10–15 kW to 120–140 kW in planning scenarios.
- This rapid jump creates intense demand for land, power and utility engagement that outpaces supply.
Design Data Centers For Mixed Cooling Methods
- Plan for flexible cooling and multiple cooling methods because rack heat loads vary and will increase rapidly.
- Combine chilled water, rear-door heat exchangers, in-row cooling or immersion where appropriate to future-proof sites.
Legacy Buildings Often Can't Host AI Racks
- Older buildings often lack floor load ratings, overhead structure and mechanical capacity for high-density AI racks.
- Retrofitting can be extensive; many operators must consider new greenfield sites to host modern AI workloads.