Today, we officially announce the launch of our next-generation high-density liquid-cooled server solution.

Built upon our self-developed patented “phase-change immersion liquid cooling” technology, this solution is designed to address the thermal bottlenecks and energy-consumption challenges of high-performance computing environments. It provides cloud computing, AI training, and supercomputing customers with a greener computing infrastructure featuring:

• Higher compute density

• Lower PUE

• Better total cost of ownership (TCO)

Why Liquid Cooling Has Become Inevitable

Over the past decade, data center computing power has increased dozens of times — but power consumption has surged alongside it.

Mainstream CPUs have evolved from consuming around 100W to 300W–400W today.

Meanwhile, GPUs designed for AI training now exceed 700W per card.

A single 8-GPU AI server can easily consume over 5kW of power — equivalent to running dozens of household air conditioners simultaneously.

This rapid rise in power density has created two major challenges.

Challenge One: Air Cooling Is Approaching Its Physical Limit

Air has fixed thermal conductivity and heat capacity limitations.

Once a single chip exceeds 300W of thermal output, traditional air cooling requires extremely high airflow speeds to dissipate heat effectively.

However, higher airflow introduces:

• Increased noise

• Higher fan power consumption

• Greater mechanical stress

At the 500W level and beyond, traditional air cooling simply cannot keep up.

Challenge Two: Electricity Costs Have Become an Operational Burden

Electricity expenses account for 30%–50% of total data center operating costs.

Importantly:

• IT equipment power consumption is only part of the equation

• Cooling systems consume a massive additional amount of energy

Traditional air-cooled data centers typically operate at a PUE of 1.5–2.0.

This means:
For every 1 kWh used for computing, an additional 0.5–1 kWh is consumed purely for cooling.

For hyperscale data centers operating tens of thousands of servers, this translates into tens or even hundreds of millions in additional annual operating costs.

Under these circumstances, liquid cooling has evolved from an optional technology into an inevitable one.

Liquids possess thermal conductivity approximately 25 times greater than air, enabling far more efficient heat transfer while significantly reducing cooling energy consumption.

Our Approach: From Passive Cooling to Active Thermal Control

Our exploration into liquid cooling began three years ago.

At that time, a customer approached us with a unique challenge.

They planned to deploy an AI training cluster inside an older building with:

• Limited structural load capacity

• Insufficient cooling infrastructure

• Restricted electrical availability

Traditional air-cooled deployment was simply impossible.

The customer asked:

“Can you create a more efficient thermal solution that allows us to deploy more compute power within limited space and power conditions?”

That question initiated our dedicated liquid-cooling R&D program.

Three years later, the result is our phase-change immersion liquid cooling solution.

Core Innovation: Zoned Phase-Change Cooling

The biggest difference between our solution and conventional immersion cooling systems lies in our “zoned phase-change” architecture.

Traditional immersion cooling submerges the entire server into cooling fluid.

While highly effective thermally, this approach introduces two major issues:

1. Low-heat components such as NICs and storage devices are unnecessarily immersed

2. Maintenance becomes difficult, since servicing a single component often requires removing the entire server from liquid

Our design separates the server into:

• High-heat zones

• Non-high-heat zones

High-Heat Zone

Components including:

• CPUs

• GPUs

• Memory modules

are enclosed within independent micro phase-change chambers.

Inside these chambers:

• Phase-change coolant absorbs heat

• The liquid evaporates into vapor

• Vapor rises to the condenser area

• Ambient cooling water removes the heat

• Vapor condenses back into liquid

• The liquid returns to the bottom and repeats the cycle

This continuous process delivers extremely efficient thermal transfer.

Non-High-Heat Zone

Components such as:

• SSDs

• NICs

• PCIe slots

remain in a conventional air environment with auxiliary low-speed airflow.

Unlike traditional servers, these components require only minimal airflow, resulting in:

• Very low fan power consumption

• Extremely low acoustic noise

Three Core Advantages

Advantage One: Exceptional Cooling Efficiency

The effective thermal conductivity of phase-change transfer is 5–10 times higher than conventional liquid cooling.

In laboratory testing, our solution maintained a 700W GPU core temperature below 65°C — nearly 30°C lower than traditional air-cooling systems.

Advantage Two: Easier Maintenance

Because the non-high-heat zone remains air-cooled:

• SSD replacement remains simple

• NIC installation remains convenient

• Maintenance procedures remain familiar

Although the high-heat zone is sealed, quick-connect interfaces allow fast servicing without draining coolant.

Advantage Three: Lower Retrofit Costs

Our solution does not require converting an entire data center into a “coolant pool.”

Instead, existing rack infrastructure can be upgraded by adding:

• Coolant distribution units

• Thermal piping systems

This dramatically lowers retrofit barriers for already operational facilities.

Real-World Test Results: PUE Below 1.15

Over the past three months, we deployed a full prototype system for continuous full-load validation.

Test environment:

• 32 liquid-cooled AI servers

• Latest-generation AI accelerators

• Continuous 30-day operation

The results were impressive.

Thermal Performance

All CPU and GPU temperatures remained stable between:

• 60°C and 70°C

Temperature variation between chips stayed within:

• 5°C

This significantly outperformed traditional air cooling, where thermal variation commonly reaches 15–20°C.

Energy Efficiency

The system achieved a PUE of 1.12.

This means:
For every 1 kWh consumed for computing, only 0.12 kWh was needed for cooling and auxiliary infrastructure.

Comparable air-cooled systems typically operate above 1.6 PUE.

Noise Reduction

By eliminating high-speed fans, operational noise dropped from approximately:

• 85 dB (air cooling)
  to:

• below 55 dB

One engineer participating in testing remarked:

“Standing next to the rack, you can barely hear anything.”

What These Results Mean for Customers

For customers, these results translate into major operational advantages:

• Deploy 40% more compute power within the same power budget

• Save millions annually in electricity costs

• Build data centers in locations previously unsuitable due to cooling constraints

• Even deploy high-density compute systems inside office buildings

Customer Success Story: From Impossible to Reality

The pilot customer for this solution was the same AI company that approached us three years ago.

Previously, because of cooling and electrical limitations inside their aging building, they could deploy only 32 servers.

After adopting our liquid cooling solution:

• They successfully deployed 64 servers in the same space

• Computing capacity doubled

• Electricity costs remained unchanged

Their technical director shared:

“With traditional air cooling, achieving this compute scale would have required building an entirely new data center — costing at least 50 million RMB and taking over a year. With your liquid cooling solution, we completed everything within three months at less than 10 million RMB.”

An additional surprise was the dramatic reduction in noise.

Because the liquid-cooled servers operate so quietly, the company placed racks directly beside office areas.

Engineers no longer needed to walk into remote machine rooms to debug AI models — they could operate systems directly from their desks, significantly improving development efficiency.

The Future of Green Computing

Today, carbon reduction goals have become a shared priority across industries.

For data centers — one of the world’s largest electricity consumers — green transformation is no longer optional.

Cities such as:

• Beijing

• Shanghai

• Shenzhen

have already mandated:

• New data centers must achieve PUE below 1.3

• Some core districts require PUE below 1.15

Under these standards, traditional air cooling is effectively no longer viable.

Liquid cooling has become the only sustainable path forward.

A Massive Retrofit Opportunity

Beyond new facilities, retrofitting existing data centers represents an enormous market opportunity.

China alone has more than 70,000 existing data centers, most of which still rely on air-cooling architectures with PUE values above 1.5.

If these facilities were upgraded to liquid cooling:

• Hundreds of billions of kWh could be saved annually

• Tens of millions of tons of carbon emissions could be reduced

Our liquid-cooling solution was created specifically for these two markets.

For new data centers:

• We provide complete liquid-cooling infrastructure design

For existing facilities:

• We provide retrofit solutions requiring no reconstruction of data halls or rack systems

helping customers achieve green upgrades at significantly lower cost.