Why Modern AI Chips Are Forcing Data Centers to Switch to Liquid Cooling

Data centers are changing fast because of AI. The biggest reason is not just software, but the chips inside servers. Modern GPUs are becoming extremely powerful, and with that power comes a lot of heat. Earlier, cooling was manageable with air. But today, heat levels are growing so fast that traditional methods are failing. This is why liquid cooling is no longer optional—it is becoming necessary. In this blog, we will understand how chip power (TDP) is driving this shift in simple words.

What is TDP (Thermal Design Power)?

TDP means Thermal Design Power, which is the amount of heat a chip generates that needs to be cooled. It is usually measured in watts (W).

In simple terms:👉 Higher TDP = More heat = More cooling needed

For example:

• Older CPUs (before 2015): 80W – 150W

• GPUs around 2018–2020: 200W – 300W

• Modern AI GPUs: 700W – 1200W+

This shows how fast heat generation has increased.

How Chip Power Has Changed Over Time

Before 2020, chip power was growing slowly. Each new generation added a small increase in heat. Data centers could handle this using better fans and airflow.

But after 2020, things changed dramatically. AI and machine learning workloads started demanding much more computing power. Companies began designing GPUs specifically for AI tasks.

This caused TDP to grow almost exponentially, not gradually. Instead of small increases, we now see massive jumps in power consumption with each new chip generation.

The Rise of High-Power AI GPUs

Modern GPUs used for AI training and inference are extremely powerful. Some of the latest chips are reaching 1.0 kW to 1.2 kW per GPU.

To understand this:

• A single GPU today can consume as much power as a small household appliance

• A server may contain multiple GPUs

• A rack can hold dozens of such servers

👉 This leads to rack power densities of 50kW to 100kW+, which is far beyond what traditional data centers were designed for.

Why Air Cooling Cannot Handle This

Air cooling works by moving cold air across components and removing heat. But air has very low heat capacity.

As chip power increases:

• Fans need to spin faster

• More air needs to be moved

• Cooling systems consume more energy

At very high densities (like 1kW+ GPUs), air simply cannot remove heat fast enough. Even if you increase airflow, it becomes inefficient and expensive.

👉 This is the “wall” that air cooling has hit.

The Turning Point: Around 1kW Per Chip

Industry experts often say that once chips cross ~500W–700W, air cooling starts struggling.

At 1kW+ per chip, air cooling becomes almost impossible in dense environments.

This is exactly where modern AI GPUs are today. For example:

• New generation GPUs can reach 1.2 kW TDP

• Multiple GPUs in one server multiply the heat problem

👉 At this level, only liquid cooling can handle the heat efficiently.

Why Liquid Cooling Becomes Necessary

Liquid cooling works differently. Instead of cooling the air around the chip, it removes heat directly from the source.

Because liquids transfer heat much better than air:

• Heat is removed faster

• Less energy is wasted

• Systems remain stable even at high power

This makes liquid cooling ideal for high-density AI workloads.

Impact on Data Center Design

The rise of high-TDP chips is changing how data centers are built.

Earlier:

• Designed for 5kW–10kW per rack

• Air cooling was enough

Now:

• Designed for 50kW–100kW+ per rack

• Liquid cooling is required

This shift is forcing companies to rethink infrastructure from the ground up.

Statistics That Show the Shift

• Cooling can take 30–40% of total data center energy

• Liquid cooling can reduce this by 20–50%

• AI workloads are expected to grow 10x in the next few years

• High-end GPUs now exceed 1kW TDP, compared to <300W a few years ago

These numbers clearly show why cooling technology must evolve.

What This Means for the Future

The growth of AI is not slowing down. As models become larger, the demand for powerful chips will continue to rise.

This means:

• Even higher TDP in future chips

• Even more heat in data centers

• Greater need for efficient cooling

Liquid cooling is not just a trend—it is becoming a requirement for modern infrastructure.

Conclusion

The biggest driver behind the shift to liquid cooling is the rapid increase in chip power. Thermal Design Power has grown from a few hundred watts to over 1kW per chip in just a few years.

Air cooling cannot keep up with this change. It is inefficient at high densities and consumes too much energy. Liquid cooling, on the other hand, provides a direct and efficient way to remove heat.

As AI continues to push the limits of computing, data centers must adapt. The rise in TDP is a clear signal that the future of cooling is liquid.