AI Cooling Is Moving Closer to the Chip
The rapid growth of AI infrastructure is changing the role of traditional HVAC companies.
Cooling is no longer limited to the mechanical room.
It is moving directly toward the GPU and CPU.
LG Electronics is one of the companies building this broader thermal architecture.
Its AI data center strategy increasingly connects several layers:
Cold Plate → CDU → Facility Cooling → Chiller
This “Chip-to-Chiller” direction reflects a major shift in data center design.
The thermal system is becoming part of the compute architecture itself.
LG Is Building the Single-Phase Foundation
LG’s currently publicized Direct-to-Chip systems are based on conventional liquid cooling.
Coolant flows through a cold plate mounted directly on high-heat components such as GPUs and CPUs, absorbs heat, and returns through the liquid loop to the CDU.
This architecture is becoming the mainstream approach for high-density AI servers because it fits relatively well with today’s rack, manifold and service infrastructure.
LG has also been developing large-capacity CDUs and integrated cooling control technologies.
This creates an important foundation.
Even if the industry moves toward two-phase DLC later, much of the surrounding infrastructure developed for single-phase cooling may remain relevant.
OCP Korea Tech Day 2026: From Single-Phase Toward Two-Phase
At OCP Korea Tech Day held in Seoul on August 21, 2026, a technical session discussed the evolution of AI data center liquid cooling and the possibility of moving from today’s single-phase DLC toward future two-phase systems.
The discussion reflected an increasingly important industry question.
Current AI infrastructure is being built around single-phase direct liquid cooling, including:
CDU → Rack Manifold → Hose → Quick Disconnect → Cold Plate
However, future AI racks are expected to operate at dramatically higher power densities.
As rack power moves from tens of kilowatts toward hundreds of kilowatts and eventually toward the megawatt class, simply increasing liquid flow may become less attractive.
This is where two-phase DLC becomes important.
The OCP Korea Tech Day discussion also emphasized that the transition will not depend only on the cold plate.
Quick disconnects, manifolds, hoses, refrigerant compatibility, service procedures and open interface standards will all become part of the next-generation cooling architecture.
Why Two-Phase DLC Is Different
In a conventional single-phase system, coolant remains liquid while carrying heat away.
Cooling capacity depends mainly on:
Flow Rate × Specific Heat × Temperature Rise
Two-phase cooling uses a different mechanism.
A refrigerant enters the cold plate as liquid.
When it absorbs heat from the GPU or CPU, part of the fluid boils.
The phase transition from liquid to vapor absorbs significant thermal energy through latent heat.
The basic loop becomes:
Liquid Refrigerant → Evaporative Cold Plate → Vapor → Condenser/CDU → Liquid Refrigerant
This can make two-phase systems attractive for very high heat-flux processors.
ZutaCore Is Already Commercializing the Concept
Israel-origin ZutaCore provides one of the strongest commercial examples of two-phase Direct-to-Chip cooling.
Its architecture uses a dielectric refrigerant circulating inside a sealed system.
Inside the cold plate, the fluid absorbs heat and undergoes phase change.
The vapor is then transported away from the processor, condensed and returned to the cold plate.
This approach is important because it demonstrates that two-phase DTC is no longer only a laboratory technology.
It is becoming a commercial thermal-management architecture.
The Cold Plate Is Only One Part of the System
The most important lesson from two-phase cooling is that the cold plate cannot be evaluated alone.
A complete two-phase loop requires:
Evaporative Cold Plate
Refrigerant-Compatible Hose
Low-Loss Quick Disconnect
Liquid/Vapor Manifold
Condensing CDU or Heat Exchanger
Pressure and Refrigerant Control
Leak Detection and Monitoring
This creates an entirely new engineering ecosystem.
The performance of each connection becomes more important because the system is no longer transporting only liquid.
It must manage both liquid and vapor under changing thermal loads.
ZutaCore Shows the Importance of Direct-to-Silicon Cooling
ZutaCore’s recent cold plate development also illustrates another trend.
The thermal interface is becoming thinner and closer to the silicon.
Future cold plates must handle higher heat flux while reducing thermal resistance.
They may also need to support new server layouts, rear-side power delivery and more compact GPU packaging.
In this environment, the cold plate becomes less like conventional plumbing and more like part of the semiconductor system.
Why LG Could Be Strategically Well Positioned
LG already has capabilities across several thermal layers.
It develops cold plates.
It develops CDUs.
It develops chillers.
It has extensive experience with refrigerants, heat exchangers, condensers and thermal controls.
This is strategically important.
A future two-phase CDU may require more than pumping liquid.
It may need to manage:
Vapor return
Condensation
Refrigerant inventory
Pressure stability
Liquid supply control
In that sense, a two-phase CDU begins to look more like a compact refrigeration system.
This is precisely the type of engineering area where major HVAC companies could have an advantage.
Refrigerant Selection Will Become Critical
Two-phase cooling also changes the importance of the working fluid.
The refrigerant must provide suitable:
Boiling Temperature
Dielectric Performance
Material Compatibility
Environmental Performance
Operating Pressure
Thermal Stability
Seals, hoses, quick disconnects and manifolds must all be compatible with the selected refrigerant.
A component that works perfectly with water-glycol may not automatically work with a two-phase refrigerant.
This is why future two-phase systems will require closer cooperation between cold plate suppliers, CDU manufacturers, refrigerant companies and fluid-interface suppliers.
Quick Disconnects May Become More Important
Quick disconnects are already important in current DLC systems.
In two-phase systems, their role may become even more critical.
The interface may need to control:
Very Low Leakage
Low Pressure Drop
Liquid and Vapor Conditions
Repeated Connection Cycles
Material Compatibility
Automated or Blind-Mate Connection
For future AI racks, these connectors may eventually need to be operated not only by technicians but also by robots.
That creates another connection between liquid cooling and Physical AI.
DATAAD View: Preserve the Existing DLC Infrastructure
DATAAD expects the transition toward two-phase DLC to be gradual.
Single-phase systems already have growing infrastructure around:
Rack Manifolds
CDUs
Hose Assemblies
Quick Disconnects
Server Cold Plates
OCP Interfaces
It would be economically inefficient to replace all of this infrastructure immediately.
A more realistic path may be to preserve as much of the existing single-phase rack architecture as possible while introducing two-phase cooling where heat density requires it.
This could produce hybrid systems.
For example:
Single-Phase DLC for lower-power components
combined with
Two-Phase DLC for the highest-power GPUs and CPUs
Such a transition would reduce disruption while allowing the cooling system to evolve with processor power.
The Megawatt Rack Changes the Equation
AI racks are moving rapidly toward much higher power density.
At sufficiently high rack power, traditional single-phase cooling may require increasingly large coolant flow rates, larger pumps and larger manifolds.
Two-phase cooling offers another path.
Instead of simply moving more liquid, it uses phase change to transport more heat.
This does not automatically make two-phase cooling simpler.
In fact, it introduces new challenges in refrigerant management, pressure control, sealing and serviceability.
But those challenges may become worthwhile when processor heat flux reaches new levels.
LG and ZutaCore Represent Two Different Starting Points
LG represents the large-scale HVAC and infrastructure side of the market.
ZutaCore represents the specialized two-phase cooling technology side.
LG brings:
Chillers
CDUs
Heat Exchangers
Refrigeration Expertise
Data Center Infrastructure
ZutaCore brings:
Two-Phase Direct-to-Chip Cooling
Evaporative Cold Plates
Dielectric Refrigerant Systems
High-Heat-Flux Cooling Architecture
The long-term AI cooling market may increasingly require capabilities from both sides.
From Chip-to-Chiller to Chip-to-Refrigerant
Today, the industry is talking about Chip-to-Chiller cooling.
Tomorrow, the architecture may become even more integrated.
The full system could evolve toward:
Silicon → Two-Phase Cold Plate → Refrigerant Loop → Condensing CDU → Chiller
At that point, the boundary between IT cooling and HVAC begins to disappear.
The complete data center becomes one thermal system.
DATAAD Outlook
Single-phase Direct Liquid Cooling will likely remain the dominant AI data center cooling architecture in the near term.
The infrastructure is already being standardized.
The ecosystem is expanding.
And operators are becoming familiar with liquid-cooled racks.
But the discussion at OCP Korea Tech Day 2026 also showed why the industry is already looking beyond the current generation.
As AI racks move toward several hundred kilowatts and eventually toward the megawatt class, two-phase Direct Liquid Cooling becomes increasingly difficult to ignore.
ZutaCore demonstrates that the technology is commercially possible.
LG demonstrates how major HVAC companies are moving closer to the chip through integrated cooling infrastructure.
The next major transition in AI data centers may therefore not be:
Air Cooling → Liquid Cooling
but rather:
Single-Phase Liquid Cooling → Hybrid and Two-Phase Liquid Cooling
And the companies that can connect the cold plate, refrigerant, hose, quick disconnect, manifold, CDU and chiller into one reliable system may become some of the most important infrastructure suppliers of the AI era.