GPU Infrastructure Evolution

AI POD Evolution Roadmap

Caprelion AI PODs are designed to evolve with the GPU roadmap — from NVL72 through NVL144, NVL288 and NVL576-class density categories across Blackwell, Rubin and Feynman generations.

Caprelion prepares infrastructure before GPU clients are blocked by traditional hyperscale redesign cycles.

Built for GPU Generation Velocity

NVL72-class infrastructure

Current Readiness

GB200 / GB300 NVL72-class infrastructure. ~130–150 kW per AI rack. Liquid cooling dominant. Complete SU / AI POD block delivery.

GB200 NVL72 — ~130 kW/rack

GB300 NVL72 — ~150 kW/rack

Direct-to-chip liquid cooling

Complete SU/AI POD block

NVL144-class infrastructure

Next-Generation Planning

Rubin / Rubin Ultra NVL144-class planning. 480V AC power architecture. ~215–270 kW per AI rack. Enhanced CDU/XDU requirements.

Rubin NVL144 — ~215 kW/rack

Rubin Ultra — ~245 kW/rack

480V AC power architecture

Enhanced cooling plant

NVL288 / NVL576 planning

Future Density Horizon

HVDC / 800V-class awareness. NVL288 / NVL576 extreme density. 440–1,500+ kW-class racks. Facility-scale heat rejection and scalable power.

HVDC / 800V-class awareness

440–550 kW NVL288-class racks

~870 kW–1.5 MW NVL576-class

Facility-scale heat rejection

Full GPU Generation Matrix

Blackwell

GB200

Blackwell Ultra

GB300

Rubin

VR200

Rubin Ultra

VR300

Feynman

VF200

NVL72

Current readiness

GB200 NVL72

Current readiness

~130 kW / AI rack

GB300 NVL72

Current priority

~150 kW / AI rack

Vera Rubin NVL72

Transition planning

Planning horizon

—

NVL144

Next-gen planning

—

VR200 NVL144

H2 2026 planning

~215 kW / AI rack

VR300 NVL144

Late 2027 planning

~245 kW / AI rack

VF NVL144

2028–2029 planning

~270 kW / AI rack

NVL288

Advanced horizon

—

VR200 NVL288

Advanced planning

~440 kW-class rack

VR300 NVL288

Advanced planning

~500 kW-class rack

VF NVL288

Advanced planning

~550 kW-class rack

NVL576

Long-term density

—

VR200 NVL576

Long-term planning

~870 kW-class rack

VR300 NVL576

Long-term planning

~1 MW-class rack

VF NVL576

Long-term planning

~1.5 MW-class rack

Current Readiness

Near-Term Planning

Advanced Planning

Long-Term Planning

Planning Themes

Future-Ready AI POD Planning

Caprelion infrastructure is not only prepared for current GB200 / GB300 NVL72-class deployments. The platform tracks the evolution of each GPU generation and plans infrastructure upgrades in parallel.

480V AC planning for NVL144-class evolution

HVDC / 800V-class awareness for higher-density

Scalable heat rejection architecture

Liquid cooling readiness framework

External plant modular expansion

Flexible power upgrade paths

Upgrade-ready AI POD blocks

GPU generation handover planning

Power Architecture Evolution

NVL72 — Current

Standard AC power architecture. 130–150 kW per AI rack planning envelope.

NVL144 — Planning

480V AC planning. 215–270 kW per AI rack. Enhanced CDU/XDU requirements.

NVL288 — Advanced

HVDC / 800V-class awareness. 440–550 kW-class racks. Facility-scale heat rejection.

NVL576 — Long-Term

~870 kW to ~1.5 MW per rack. Extreme density planning.

Disclaimer: Future-generation names, dates, voltage assumptions and power figures are forward-looking infrastructure planning assumptions based on public roadmap discussions and remain subject to final OEM specifications. Caprelion does not claim official compatibility, certification, endorsement or partnership unless separately confirmed.