At OCP Global Summit 2025, Flex executives Rob Campbell and Chris Butler presented a roadmap for rearchitecting data centers by breaking down the organizational and physical walls between power delivery, cooling, and IT hardware. They argued that AI-driven rack densities — scaling from 20 kW to 1 MW — make the existing siloed structure untenable, and demonstrated Flex's full-stack approach (power, liquid cooling, and IT rack hardware) as the solution. The presentation walked through three generations of data center architecture: today's 48 V air-cooled, the 2027 consolidated liquid-cooled 800 V DC, and the 2030 next-gen using solid-state transformers with medium voltage brought directly into the data hall.
Key Takeaways
- Rack power has escalated from 20 kW to 50 kW to 1 MW in roughly one year, making air cooling obsolete at high density and forcing a rethink of the entire power-compute-cooling stack.
- Moving from 48 V to 400–800 V DC (targeted 2027) delivers 20% efficiency gain, $240k annual energy savings per rack (Florida baseline), and an 80% reduction in IT floor space (from 160 to 32 sq ft).
- The 2030 'Next Gen' architecture uses solid-state transformers to bring medium voltage directly into the data hall, eliminating the separate electrical room, yielding 27%+ efficiency gain, $330k energy savings per 1 MW per year, and 90% space savings.
- NFPA 70E electrical safety standards do not address DC voltages above 600 V; the industry shift to 800 V DC requires a new safety framework that does not yet exist.
- Flex claims to be the only company offering the complete vertical stack — utility power pods through rack-level power modules, IT hardware, rack integration, and liquid cooling CDUs — and launched a pre-fabricated, pre-commissioned integrated platform at OCP 2025.
- Flex's integrated platform can cut data center deployment time by one-third (from a baseline of 18 months), with pre-optimized design, pre-commissioning, and on-site final acceptance by Flex personnel.
- Organizational silos persist at data center operators — power, cooling, and IT teams often do not converge until four levels up in the org chart — and Flex argues this must change for AI infrastructure to scale.
Sign in to read the full article.
Sign In