Rapid AI factory scaling is reshaping data center infrastructure. In the near term, independent power racks stabilize power delivery, while the mid-to-long-term roadmap points to centralized HVDC distribution and solid-state transformers. Meanwhile, next-gen AI architectures are fully adopting liquid cooling, elevating it from optional to essential and driving deep cross-domain supply chain collaboration.

Key Highlights

• Scale Shift: Massive AI expansion drives complete infrastructure overhauls and modular designs.
• Power Evolution: Interim independent power racks will transition to centralized high-voltage DC grids.
• Liquid Cooling: Next-gen AI platforms fully adopt liquid cooling, making it essential.
• Synergy: Total ecosystem collaboration is vital.

Table of Contents

1. Introduction
   • NVIDIA estimates that clustered GW-level data centers will accommodate up to 300,000 next-generation GPUs
2. 2026-2027 Short-Term Solutions: Power Rack Architectures from Delta and LITEON
   • Delta's and LITEON's Respective Power Rack Designs for NVIDIA's Architectures
3. Mid-to-Long-Term Evolution Post-2028: HVDC Architecture Blueprints from Schneider and Eaton
4. Liquid-Cooling to Gradually Shift from an Optional Configuration to an Indispensable Specification as NVIDIA Looks to Surpass 50% in Penetration Rate of AI Chips in 2026 alongside Releases of Vera Rubin and LPU
   • Penetration Rate of Liquid-Cooling among NVIDIA's AI Chips to Surge in 2026
5. Conclusion

<Total Pages: 8>