LG Electronics Targets Liquid Cooling Growth in AI Data Centers

LG Electronics is expanding beyond conventional chillers after securing about KRW 600 billion in AI data center cooling orders during 2026’s first half. The figure, surfaced through a google news report, mainly covers chillers and air-side equipment rather than its newer rack-side liquid cooling products.

That distinction creates the real tension. LG already knows how to remove heat from buildings, but AI infrastructure increasingly requires cooling inside densely packed server racks. The company must now prove it can convert its HVAC position into a trusted place beside NVIDIA GPUs.

According to the original cooling orders report, LG expects additional orders worth several trillion won by the end of 2026. That target combines a large commercial ambition with an unsettled technology transition.

NVIDIA’s rack-scale systems increasingly depend on liquid cooling, creating an opening for suppliers that can connect cold plates, coolant distribution units, chillers, controls, and facility infrastructure. Yet operators still care about compatibility, leak management, maintenance, and retrofit costs. Certification starts that conversation, but real deployments decide it.

What the Google News Report Actually Changed

LG has moved from presenting a cooling portfolio to attaching near-term orders, certified hardware, and a capacity roadmap to that portfolio.

The report describes an expansion beyond chillers and air-side systems. Those products remove heat from rooms or facility water loops. LG is adding rack-side equipment that operates much closer to processors.

Its central rack-side product is a coolant distribution unit, or CDU. A CDU regulates the temperature, pressure, and flow of liquid traveling between facility cooling systems and server components.

Cold plates form another part of that chain. These metal assemblies sit against heat-generating processors and transfer thermal energy into circulating coolant. They replace some work previously assigned to fans and cold air.

LG reportedly secured about KRW 600 billion in purchase orders for AI data center cooling equipment during the first six months of 2026. The company said chillers and air-side systems represented most of those orders.

That composition matters because it establishes demand without proving rack-side adoption. A chiller order shows that LG can participate in a data center project. It does not automatically validate LG’s CDU economics, service model, or long-term reliability.

The company also expects several trillion won in additional orders before 2026 ends. That forecast is much larger than its reported first-half order total, although the public report does not provide a precise amount.

It remains unclear how much of the expected backlog will involve rack-side liquid cooling. The report says LG expects rack-side revenue to grow as power density and heat increase. It does not disclose signed CDU revenue or customer identities.

LG’s hardware roadmap makes the shift more concrete. NVIDIA has validated a 600-kilowatt CDU, while LG is developing higher-capacity systems for larger clusters.

A 1-megawatt unit was reportedly approaching the end of certification. LG also planned 2.5-megawatt and 4-megawatt units for the second half of 2026.

Those capacities suggest that LG wants to support more than isolated racks. Larger CDUs can serve wider groups of liquid-cooled systems, depending on facility design and operating conditions.

The progression also signals a broader commercial strategy. LG wants to sell cooling as a connected stack rather than a collection of unrelated mechanical products.

That stack can include cold plates, CDUs, room-level equipment, chillers, monitoring software, and power infrastructure. The approach gives LG more influence over system integration, where many liquid-cooling problems emerge.

The google news headline therefore captures only part of the event. LG is not simply increasing production of another cooling appliance. It is attempting to move closer to the compute layer and capture more value from every AI facility.

That move creates higher technical and commercial expectations. Rack-side systems interact directly with expensive processors, network equipment, manifolds, sensors, firmware, and facility water loops. A failure can affect computing availability rather than room comfort.

The first-half order figure gives LG credibility. The product roadmap establishes intent. Neither one settles whether hyperscalers and colocation operators will standardize around LG’s complete architecture.

AI Racks Are Forcing Cooling Suppliers Closer to the Chips

The opportunity exists because rack-scale AI systems concentrate more computing power and heat than traditional air-cooled layouts were designed to handle.

NVIDIA’s GB200 NVL72 illustrates the change. The system connects 72 Blackwell GPUs and 36 Grace CPUs inside a rack-scale design that uses liquid cooling.

NVIDIA documentation lists approximate rack power consumption of 120 kilowatts for a GB200 NVL72. It also describes liquid-cooled processors, switches, manifolds, power shelves, and leak-detection components in the broader architecture.

The GB200 rack design shows why cooling has become part of computing architecture. Operators cannot treat it as a separate building service after selecting their servers.

Liquid carries heat more effectively than air in compact spaces. Direct-to-chip cooling sends coolant through cold plates attached to processors, while fans continue cooling components that remain suitable for air.

This hybrid arrangement preserves some existing facility infrastructure. It also allows the hottest components to receive more targeted thermal management.

LG describes its own approach as chip-to-chiller coverage. The phrase connects cold plates at the processor with CDUs, air-side systems, facility controls, and chillers outside the rack.

That coverage can appeal to operators who want fewer integration boundaries. Every handoff between vendors introduces questions about pressure, coolant quality, controls, warranties, alarms, maintenance, and responsibility during failures.

A broader portfolio does not eliminate those boundaries. It can reduce them when the equipment has been engineered and tested as one operating system.

LG has been preparing for that role since before the latest report. At Data Center World 2026, the company displayed a 1.4-megawatt CDU alongside direct-to-chip equipment, computer room air handlers, chillers, and management software.

Its integrated cooling lineup also included immersion cooling partnerships. Immersion cooling places servers or components in a nonconductive fluid rather than circulating coolant through individual cold plates.

The two methods address related heat problems but require different operating models. Direct-to-chip systems retain familiar rack structures. Immersion designs change hardware access, service procedures, and facility layout more substantially.

LG does not need immersion cooling to displace direct-to-chip systems for its strategy to work. It needs enough flexibility to serve customers with different rack densities, building designs, and deployment timelines.

The pressure extends beyond cooling specialists. Server makers, chip vendors, construction firms, power suppliers, and data center operators must coordinate earlier in a project.

An operator buying a high-density AI system cannot wait until installation to determine whether pipes, pumps, heat exchangers, and monitoring controls are compatible. Cooling choices influence building design, commissioning, and computing capacity.

This creates a procurement advantage for companies that can join projects early. LG already sells large mechanical systems and can potentially enter discussions before individual racks arrive.

It also creates pressure on established data center cooling suppliers. Vertiv, Schneider Electric, Johnson Controls, Carrier, and other vendors bring their own engineering relationships, service networks, and liquid-cooling portfolios.

LG’s challenge is not recognizing the trend. Competitors recognize it too. The contest concerns which suppliers can deliver repeatable, maintainable systems across multiple regions and compute generations.

NVIDIA Validation Gives LG Entry, Not Control

NVIDIA validation lowers one procurement barrier, but it does not guarantee that customers will deploy LG equipment at scale.

LG announced in July 2026 that NVIDIA had validated its 600-kilowatt CDU against more than 100 evaluation criteria. The company positioned the result as confirmation of compatibility with NVIDIA AI infrastructure requirements.

The 600kW validation applies to one defined product and capacity. It should not be interpreted as certification of every planned LG unit or every possible deployment configuration.

LG says the validated unit can maintain coolant temperature within plus or minus 0.25 degrees Celsius. It also includes real-time monitoring, leak detection, virtual sensors, and centralized management software.

Those functions address important operational requirements. Temperature instability can reduce computing performance or trigger protective behavior. Undetected leaks can damage equipment and interrupt workloads.

The certification also places LG within NVIDIA’s infrastructure partner network. That position can help buyers narrow vendor lists when they design systems around NVIDIA reference architectures.

However, the cooling supplier does not control the entire specification. Chip platforms determine much of the heat load, flow requirement, allowable coolant temperature, and rack layout.

Server manufacturers add another layer. Their designs determine cold-plate placement, internal tubing, quick disconnects, manifolds, leak detection, and service access.

Facility operators then add water conditions, redundancy requirements, maintenance practices, monitoring systems, and local regulations. Each layer can alter how a CDU performs outside a validation environment.

LG operates a chip-to-chiller testing site in Pyeongtaek, South Korea. The facility lets the company test interactions among rack-side devices, facility cooling, and control software.

That lab is strategically important because capacity alone says little about system quality. A megawatt rating does not explain efficiency across changing loads, response during pump failures, or behavior during maintenance.

AI workloads can change quickly. Training jobs, inference traffic, and maintenance events can produce shifting thermal loads across a cluster.

A cooling system must adjust without creating unstable temperatures or wasting energy. Controls, sensors, and software therefore matter alongside pumps and heat exchangers.

LG’s planned 1-megawatt, 2.5-megawatt, and 4-megawatt units raise the scale of this challenge. Larger systems can consolidate capacity, but they can also increase the consequences of poor redundancy or control decisions.

Operators will examine whether LG’s architecture supports service without shutting down critical computing. They will also test how the system responds when sensors, pumps, valves, or network connections fail.

The company’s manufacturing background can help. LG produces compressors, heat exchangers, motors, controls, and HVAC equipment across a broad global operation.

Still, rack-side cooling requires credibility with a different technical audience. Data center reliability teams judge equipment through uptime, maintainability, interoperability, and field evidence.

This is why the NVIDIA relationship is best understood as an entry credential. It gives LG a serious place in procurement conversations. It does not give the company ownership of those conversations.

The primary contest is LG’s integrated-stack promise against the operational reality of mixed-vendor data centers. Customers rarely replace every layer of infrastructure at once.

LG must make its products work inside facilities containing equipment from multiple manufacturers. That includes older chillers, existing building controls, third-party servers, and different liquid loop specifications.

If the company handles those interfaces well, its broad portfolio becomes an advantage. If customers encounter compatibility or service friction, the same breadth can produce more points of dispute.

Liquid Cooling Demand Is Real, but Adoption Remains Uneven

Rising power density supports LG’s thesis, while slow retrofit cycles prevent the market from becoming an automatic hardware boom.

The International Energy Agency projects that global data center electricity consumption will reach about 945 terawatt-hours in 2030. That represents roughly twice the level recorded for 2024.

Its electricity demand outlook projects annual data center electricity growth of about 15 percent between 2024 and 2030. Accelerated servers represent a major share of that increase.

Cooling accounts for part of the load. The agency estimates that cooling ranges from about 7 percent of consumption in efficient hyperscale facilities to over 30 percent in less-efficient enterprise sites.

These figures strengthen the business case for more efficient thermal management. Energy used by cooling systems cannot run processors, storage, or networking equipment.

Yet a large addressable problem does not produce immediate adoption. Data centers contain long-lived mechanical systems, complicated service agreements, and conservative reliability practices.

Uptime Institute’s 2025 survey found that direct liquid cooling remained a minority technology. Among respondents answering its cooling-use question, 22 percent reported direct liquid cooling, while 75 percent used perimeter air cooling.

The cooling adoption survey included 1,033 respondents. It identified high rack density as the main adoption driver.

The survey also found that 46 percent of respondents considered retrofit ease among the most important factors determining viability. A lack of standards, high costs, and failure concerns remained barriers.

These results explain why LG continues selling chillers and air-side systems while developing rack-side products. Most facilities will not switch every workload to direct liquid cooling at once.

Operators may divide a campus into different thermal zones. Conventional servers can remain air-cooled, while dense AI racks use direct-to-chip systems.

That mixed environment favors suppliers with hybrid portfolios. LG can support room cooling while adding liquid loops where customers need greater heat removal.

However, hybrid infrastructure can also become complicated. Teams must monitor two cooling methods, coordinate control systems, and maintain different equipment types.

Retrofits make the problem harder. Existing buildings may lack suitable pipe routes, floor loading, water systems, or space for distribution equipment.

A customer also needs confidence that future server generations will fit the chosen cooling architecture. Proprietary interfaces can create expensive lock-in or force another redesign.

Standards development can reduce that risk, but implementations still differ. Coolant chemistry, materials, connectors, pressure levels, temperature ranges, and maintenance procedures all require careful alignment.

The google news report presents larger CDU capacity as evidence of progress. Capacity matters, but customers will judge the surrounding deployment package.

They will ask how quickly LG can commission equipment, train local technicians, replace parts, and diagnose problems. They will also ask which company carries responsibility when integrated systems underperform.

This is a harder test than a product demonstration. It requires documented operating history across different climates, facility types, and server configurations.

LG’s several-trillion-won order expectation should therefore be treated as a company outlook, not a completed result. Public information does not identify the customers, contract stages, or rack-side share behind that projection.

The underlying demand signal remains strong. The conversion of that demand into recognized revenue remains less certain.

The Real Contest Is Integration Versus Operational Risk

LG wants customers to buy a coordinated cooling chain, while operators need proof that integration reduces risk instead of concentrating it.

A chip-to-chiller portfolio sounds simple when shown as a diagram. Coolant collects heat from processors, passes through a CDU, and transfers that heat into facility systems.

Real facilities add pumps, filters, valves, redundant paths, controls, alarms, water-treatment requirements, maintenance bypasses, and emergency procedures. Each component affects performance elsewhere.

LG’s ability to design across those layers can reduce guesswork. It can tune controls and hardware together while providing a clearer escalation path.

The company has also expanded partnerships rather than trying to manufacture every component alone. In late 2025, LG signed an agreement with Flex covering integrated modular cooling systems for AI data centers.

The modular cooling partnership combines LG’s chillers, air systems, CDUs, and monitoring products with Flex infrastructure and power technologies.

LG also reported completing a liquid-cooling proof of concept with LG Uplus. Such projects can help translate laboratory performance into operating procedures.

Partnerships broaden market access, but they also create responsibility boundaries. A data center operator needs to know who supports the system when hardware from several suppliers interacts unexpectedly.

Cooling failures are not limited to visible leaks. Incorrect temperatures, inadequate flow, contamination, trapped air, control faults, or blocked channels can reduce reliability.

Leak detection remains important because liquid operates close to costly processors. NVIDIA’s documentation includes tray-level and rack-level detection, showing that risk management is part of the reference design.

Serviceability presents another issue. Technicians need safe methods to disconnect equipment, replace server trays, drain loops, and return racks to operation.

A system that performs efficiently but takes too long to maintain can be commercially unattractive. Downtime costs often outweigh small energy savings for critical workloads.

Supply-chain capacity also deserves scrutiny. LG plans several larger CDU models during 2026, while pursuing a sharp increase in cooling orders.

Scaling engineering prototypes into repeatable products requires qualified components, production testing, documentation, field technicians, and replacement inventory. Public capacity ratings do not reveal readiness across those areas.

Customer concentration creates another uncertainty. A few large projects can produce impressive order totals while exposing a supplier to delayed construction or specification changes.

AI data center projects already face constraints involving electricity, grid connections, land, financing, networking equipment, and accelerators. A cooling order can move when another part of the project slips.

Revenue timing may therefore differ from purchase-order timing. Investors should distinguish signed equipment demand from shipment, installation, acceptance, and recognized sales.

Competition will pressure margins as well as market share. Established infrastructure suppliers are investing in direct-to-chip products, CDUs, controls, and service capacity.

Some server and rack integrators package cooling with computing hardware. That model can move part of the buying decision away from traditional HVAC vendors.

LG’s strongest response is its ability to connect facility-scale equipment with rack-side hardware. Its weakest point is the limited public evidence showing that customers have adopted that full stack at scale.

The reported KRW 600 billion order figure mainly reflects LG’s established cooling categories. It demonstrates a route into projects, not the final economics of liquid cooling.

This does not invalidate the company’s strategy. It defines the evidence that still needs to appear.

What to Watch After LG’s Liquid Cooling Push

Three signals will show whether LG is building a durable AI infrastructure business or extending an HVAC narrative into a promising market.

The first signal is certification and commercial availability for higher-capacity CDUs. LG reportedly placed its 1-megawatt model near final certification and planned 2.5-megawatt and 4-megawatt products during 2026’s second half.

Completed validation would strengthen the case that LG can serve larger AI clusters. Delays would suggest that scaling capacity introduces additional engineering or compliance work.

Certification details also matter. Buyers should look for supported server platforms, operating temperatures, redundancy designs, coolant requirements, and integration standards.

The second signal is the composition of LG’s order backlog. The company expects several trillion won in additional cooling orders by the end of 2026.

A rising order total would support management’s demand forecast. However, investors need separation between traditional chillers, air-side equipment, CDUs, cold plates, software, and services.

Growth led only by chillers would still benefit LG. It would not prove the company has secured a major position inside liquid-cooled racks.

Named customers or deployment case studies would provide stronger evidence. They could show system size, operating duration, workload type, energy performance, and maintenance experience.

The third signal is field performance across mixed infrastructure. LG must show that its controls and hardware operate reliably with third-party servers and existing facility systems.

Metrics should include commissioning time, temperature stability, pump efficiency, leak events, service interruptions, and maintenance duration. Consistent results would strengthen LG’s integration argument.

Repeated interface problems would weaken it, even if standalone components meet their specifications. Data center operators ultimately purchase availability, not capacity labels.

Competitor responses will add context. New certifications, bundled rack systems, acquisitions, or large customer wins could narrow LG’s opportunity.

The market does not need a single winner. AI infrastructure is growing across regions and facility types, creating room for several cooling architectures and suppliers.

Still, vendor position matters because early design wins can shape later expansions. Operators often prefer familiar components, procedures, and service relationships after a successful deployment.

For developers and AI product teams, this infrastructure contest can appear distant. It directly affects how quickly compute capacity becomes available and how much sustained performance operators can deliver.

For enterprise buyers, cooling maturity affects deployment schedules and provider selection. A facility unable to support dense systems may offer fewer accelerator options or require larger physical footprints.

Knowledge workers following the story through google news should look beyond the order headline. The important shift is the connection between silicon roadmaps and mechanical infrastructure.

LG has assembled many required pieces: facility equipment, rack-side products, controls, testing capacity, partnerships, and NVIDIA validation. Its next task is turning those pieces into repeatable customer deployments.

Watch the higher-capacity certifications first, the rack-side share of new orders second, and operating evidence third. Together, those signals will reveal whether LG’s expansion has moved from product roadmap to infrastructure standard.

The opportunity is credible, but the outcome remains open. Will LG disclose enough deployment evidence to show that liquid cooling has become a substantial business, rather than an extension of its established chiller sales?