NVIDIA Blackwell GPUs generate up to 1,000 watts per chip, over three times more heat than GPUs from seven years earlier. Traditional air cooling physically cannot dissipate heat at these densities. Above 50 to 100 kW per rack, liquid cooling is not optional. It is physics.

Per Blackwell chip
1,000 W
Current rack density
132 kW
Expected next generation
240 kW
Air cooling limit
50 to 100 kW

Sources: Lombard Odier, Tom's Hardware, Data Center Dynamics.

The Power Density Evolution

Understanding how we got here helps put the infrastructure challenge in context. In less than a decade, rack power density has increased nearly 10x for AI workloads.

YearRack densityWorkload
201715 kW per rackStandard enterprise workloads
202440 to 60 kW per rackAI workloads with H100 GPUs
2025132 kW per rackNVIDIA GB200 NVL72 systems
2026240 kW per rackNext-generation systems (expected)

Why Air Cooling Fails

Air has fundamental limitations as a heat transfer medium. Its thermal conductivity is roughly 25 times lower than water. At densities above 50 to 100 kW per rack, you cannot move enough air through the system to dissipate heat effectively.

The implications are straightforward: any facility planning to deploy current-generation or next-generation GPU infrastructure must plan for liquid cooling. This is not a feature preference. It is a physical requirement.

Liquid Cooling Approaches

Three primary approaches address high-density cooling requirements:

Rear-Door Heat Exchangers (RDHx)

Capacity: 30 to 50 kW per rack.

Retrofit solution for existing facilities. Captures heat at the rack exhaust. Suitable for moderate density increases but insufficient for current GPU requirements.

Direct-to-Chip Liquid Cooling

Capacity: 100 to 200 kW per rack and above.

Cold plates directly attached to CPU and GPU surfaces. Most efficient heat capture at the source. Required for high-density AI workloads. This is what NVIDIA recommends for GB200 deployments.

Immersion Cooling

Capacity: 200 kW per rack and above.

Servers fully submerged in dielectric fluid. Highest density support possible. Requires significant operational changes and specialized equipment.

What This Means for Planning

If you are planning AI infrastructure for 2026 to 2027, cooling strategy is not optional:

GPU generationRack densityCooling requirement
H100 and H20040 to 80 kWHigh-density air may work
GB200 (Blackwell)132 kWLiquid cooling required
Next generation (2026 and later)240 kWAdvanced liquid cooling mandatory