LG Electronics Chiller Expansion Puts Cooling Supply at the Center of the AI Race

LG Electronics is committing KRW 150 billion to add one American chiller factory and expand two Korean production sites. The LG Electronics chiller expansion turns cooling capacity into a direct bet on continued AI infrastructure growth.

The plan covers a new plant in Windsor, Virginia, plus upgraded production in Pyeongtaek and Changwon, South Korea. The Virginia factory is scheduled to start making air-cooled chillers during the first half of 2027.

This is more than an appliance company adding another industrial product line. LG is trying to supply the entire thermal chain between high-density AI chips and the equipment that removes their heat.

That strategy puts LG against established infrastructure specialists such as Schneider Electric and Vertiv. It also tests whether demand for AI computing will translate into dependable orders for cooling equipment.

LG Electronics Chiller Expansion Reaches Three Factories

LG is moving chiller production closer to North American data center customers while increasing capacity at home.

The company announced the investment on October 1, 2026. Its production expansion covers factories in the United States and South Korea.

The Virginia facility will manufacture air-cooled chillers, which reject heat into the outdoor air without relying on a separate cooling tower. That design can simplify construction where water availability or operating complexity limits traditional water-cooled systems.

LG says the Windsor project will occupy about 30,000 square meters of floor space on a 170,000-square-meter site. A Korean version of the announcement lists the building at 32,000 square meters, creating a minor disclosure difference.

The facility is scheduled to begin production during the first half of 2027. That timing places it inside the current wave of North American AI data center construction.

Virginia offers a logical base because it contains one of the world’s largest concentrations of data center infrastructure. Local manufacturing can shorten delivery routes and improve access to replacement equipment, parts, and technical support.

Those advantages matter because chillers are not interchangeable boxes installed after a data center is completed. Operators must coordinate them with electrical capacity, water systems, rack density, controls, and redundancy requirements.

A delayed cooling system can therefore delay the computing capacity it supports. Local production cannot remove every construction bottleneck, but it gives LG more control over a critical component’s delivery.

The company is also adding air-cooled chiller lines at Pyeongtaek and Changwon. LG plans to install conveyor-based production, moving products through a standardized sequence suited to higher output.

This manufacturing change signals more than a routine equipment upgrade. Large industrial chillers have traditionally involved extensive project-specific work, testing, and assembly.

A conveyor system suggests that LG expects enough repeatable demand to justify a more standardized production model. It also supports faster output if customers begin ordering comparable systems across multiple campuses.

The investment does not guarantee that every planned AI facility will proceed. Grid connections, permitting, financing, and server availability can still postpone projects after cooling equipment has been specified.

However, factories require longer commitments than marketing campaigns. LG is putting capital, labor, and manufacturing space behind its forecast for sustained cooling demand.

The announcement follows LG’s September launch of an air-cooled centrifugal chiller designed for AI facilities. The company says that model can restore full cooling capacity within three minutes after electricity returns following an interruption.

LG also says the design uses an oil-free magnetic-bearing compressor. Magnetic bearings suspend rotating components without conventional lubricated contact, reducing friction inside the compressor.

Its refrigerant free-cooling system can use favorable outdoor conditions to reduce compressor work. LG based its efficiency comparisons on internal simulations, so operators will still need field data from different climates.

The production expansion gives that product strategy a physical supply base. It connects engineering claims with the less visible work of sourcing components, assembling equipment, and serving installed systems.

For buyers, that connection matters. A technically capable chiller has limited value if production slots, commissioning teams, or replacement components remain unavailable.

Why AI Data Centers Are Rewriting the Cooling Market

AI servers are increasing heat density, but the larger commercial opportunity comes from building complete, dependable thermal systems around them.

Traditional data centers already need continuous temperature and humidity control. AI clusters intensify that requirement because accelerators concentrate more computing and electrical power inside each rack.

Heat must move through several stages before it leaves the site. A cold plate can collect heat from a processor, while a coolant distribution unit manages fluid flow between equipment loops.

A chiller then removes heat from the facility’s water system. Pumps, heat exchangers, controls, and outdoor heat-rejection equipment complete the chain.

This explains why the LG Electronics chiller expansion includes both air-cooled products and a wider liquid-cooling strategy. Neither approach eliminates the need for the other across every facility.

Direct-to-chip cooling places a liquid-cooled plate directly against a processor or accelerator. It can handle dense heat loads more efficiently than moving large amounts of conditioned air through server rooms.

Yet that captured heat still needs somewhere to go. Chillers and related heat-rejection systems remain part of the broader design, even when liquid replaces air near the chip.

LG calls its approach “Chip-to-Chiller.” The phrase describes a portfolio spanning cold plates, coolant distribution units, chillers, controls, and associated infrastructure.

The company’s air-cooled products also address an important resource constraint. Water-cooled systems can deliver strong efficiency, but cooling towers can consume significant water through evaporation.

Air-cooled chillers avoid that particular water requirement. Their performance can still vary with outdoor temperatures, local climate, equipment configuration, and operating conditions.

This creates a design tradeoff rather than a universal winner. Operators must balance electricity use, water availability, climate, capital requirements, maintenance, and computing density.

The energy stakes are large. The International Energy Agency estimates that cooling represents roughly 7 percent of electricity use in efficient hyperscale facilities.

That share can exceed 30 percent in less-efficient enterprise data centers, according to the agency’s energy demand analysis. Better thermal management can therefore affect both operating costs and available computing capacity.

The agency estimated global data center electricity use at about 415 terawatt-hours in 2024. Its base case projects consumption near 945 terawatt-hours by 2030.

Those figures do not mean every proposed campus will be built. They show why manufacturers now treat data center cooling as a strategic infrastructure category.

LG cites Omdia projections that global data center capacity will rise from 226 gigawatts in 2026 to 420 gigawatts in 2030. LG says 60 percent of the additional capacity will appear in North America.

Those projections come through LG’s announcement rather than a publicly accessible Omdia report. They should be treated as a forecast supporting the company’s investment case, not a guaranteed outcome.

The more immediate evidence is physical congestion across the industry. Power equipment, grid access, advanced processors, skilled labor, and cooling systems all face competing demand.

The IEA reported that data center electricity demand increased 17 percent during 2025. It also warned that transformers, gas turbines, approvals, and grid connections had become constraints.

Cooling capacity now sits inside that same infrastructure contest. A building with servers but inadequate thermal equipment cannot safely operate those servers at their intended load.

LG is responding by manufacturing closer to demand and standardizing more of its production. The strategy aims to make cooling less likely to become the component that holds up a completed computing hall.

The Main Contest Is Complete Cooling Versus Component Sales

LG’s central challenge is proving that an integrated thermal portfolio creates more value than supplying isolated pieces of equipment.

The phrase “Chip-to-Chiller” is designed to reposition LG as an infrastructure partner. It moves the company’s pitch beyond selling large HVAC machines at the edge of a facility.

Under that model, LG can participate near the processor, inside the coolant loop, and at the plant level. Software and cooling-management controls connect those layers.

This approach can simplify accountability for customers. When several vendors supply cold plates, distribution units, controls, and chillers, integration failures can create unclear responsibility.

A wider portfolio gives one supplier more influence over performance across the thermal chain. It can also help coordinate product schedules and service agreements.

However, integration is not unique to LG. Schneider Electric expanded its liquid-cooling position through its acquisition of Motivair and now markets an end-to-end portfolio.

The company’s cooling portfolio includes cold plates, coolant distribution units, rear-door heat exchangers, chillers, and technology cooling loops. It combines those products with power, automation, and data center services.

Vertiv has followed a similar direction, joining power distribution, thermal management, monitoring, and deployment services. Other established suppliers also bring long customer relationships and large service networks.

LG must therefore compete on execution rather than portfolio breadth alone. Customers will examine efficiency, uptime, certification, commissioning speed, maintainability, and total system performance.

The company gained a useful credential in July 2026 when Nvidia validated LG’s 600-kilowatt coolant distribution unit. LG says the evaluation covered more than 100 criteria connected to Nvidia’s AI infrastructure standards.

A coolant distribution unit separates and controls facility water and the cleaner coolant loop serving computing equipment. Its performance affects pressure, temperature, flow, reliability, and leak management.

The 600-kilowatt validation reduces one procurement uncertainty for customers building around Nvidia systems. It does not certify every LG component or guarantee performance inside every data center design.

LG also has a real project through its work with Sinar Mas Group in Jakarta. The company is supplying cooling technology for a hyperscale AI data center scheduled to begin operating during the second half of 2026.

That project uses cooperation across LG companies. LG Electronics provides thermal equipment, while LG CNS contributes data center design, construction, and operational capabilities.

The Jakarta deployment gives LG a reference for its “One LG” strategy. It shows how cooling, batteries, infrastructure engineering, and operations can be packaged around one facility.

LG previously said it wanted annual chiller sales to reach the equivalent of USD 720 million. It also targeted more than tripling data center cooling orders over two years.

Those are company targets, not confirmed future results. The new factories will increase the importance of converting project discussions into contracted and delivered orders.

This is where the primary competitive tension becomes clear. Component sales can generate revenue without requiring LG to control the complete thermal architecture.

An integrated solution can capture more value, but it carries greater coordination risk. Customers will expect the combined system to perform better than a collection of individually capable products.

LG must also support equipment for years after installation. Chillers and liquid loops require monitoring, maintenance, spare parts, and trained technicians across different operating regions.

Local Virginia production can support that obligation in North America. However, manufacturing proximity must be matched by engineering and field-service capacity.

The company has spent more than a decade supplying cooling systems to data centers. Its previous chiller experience in power plants, factories, and commercial buildings also provides an industrial foundation.

AI facilities still introduce a different operating profile. Rapid computing-load changes, dense server racks, high uptime requirements, and evolving processor designs place new demands on thermal controls.

The winner will not necessarily offer one superior component. It will deliver an efficient system that can be installed, controlled, repaired, and expanded without disrupting computing operations.

Air Cooling and Liquid Cooling Must Work Together

The production plan succeeds only if LG connects conventional chiller engineering with the liquid loops required by increasingly dense AI hardware.

Cooling discussions often frame air and liquid as competing technologies. That framing misses the different roles they can play inside one facility.

Air cooling regulates rooms and supports equipment that remains within manageable heat densities. Direct liquid cooling moves heat from the hottest processors with less dependence on room airflow.

The facility still needs equipment that transfers collected heat outdoors. Chillers, dry coolers, pumps, and heat exchangers remain central to that process.

LG’s new air-cooled centrifugal chiller sits at this facility level. Its compressor circulates refrigerant so heat can move from chilled water into the outside environment.

The company says each unit offers 1,750 kilowatts of cooling under specified data center conditions. Actual output will depend on water temperatures, outdoor temperature, configuration, and operating load.

LG’s internal simulation compared two free-cooling approaches in the Seoul climate. The company reported lower modeled annual energy use for refrigerant free cooling than for hydronic free cooling.

That result has not been independently validated across production sites and different climates. Buyers should treat it as a design claim requiring project-specific analysis.

The same caution applies to recovery performance after an outage. A three-minute return to full cooling is meaningful only when the surrounding pumps, controls, power systems, and thermal buffers also recover correctly.

A data center does not experience thermal failure as a collection of independent product specifications. It experiences the combined behavior of every connected system.

That systems view explains LG’s modular hydronic design. Hydronic systems move thermal energy through water or another liquid between cooling equipment and heat-producing loads.

LG plans to prefabricate pumps, a thermal storage buffer tank, and other components into standardized modules. Factory assembly can reduce work and testing at a construction site.

Modularization can also support phased capacity. Operators can add cooling blocks as new computing halls or server clusters become active.

However, standard modules still need adaptation to local weather, utility conditions, water quality, building layouts, and operating procedures. Standardization reduces some engineering work but cannot remove it.

The LG AI data center cooling strategy also depends on controls. Computing demand changes quickly, so thermal equipment must respond without wasting energy or allowing unsafe temperature swings.

LG’s Data Center Cooling Management system is intended to coordinate cooling assets. The company has not disclosed enough field performance data to compare it fully with established data center control platforms.

That gap matters because hardware efficiency does not automatically produce site efficiency. Poor control sequences can make efficient components operate at unfavorable temperatures or loads.

Operators commonly evaluate power usage effectiveness, which compares total facility energy with energy consumed by computing equipment. Cooling improvements can reduce the non-computing share.

Yet a single efficiency ratio cannot describe every tradeoff. Water use, redundancy, local temperature, utilization, maintenance, and workload patterns also affect performance.

The IEA notes that cooling’s share of facility electricity varies widely by data center type. That variation leaves room for improvement, but it also warns against universal savings claims.

LG’s chip-to-chiller cooling portfolio offers a credible architecture for addressing those variables. The unresolved question is whether deployments will produce consistent results outside controlled tests.

Customers should watch measured energy use, water consumption, fault recovery, maintenance hours, and delivered computing uptime. Those outcomes matter more than isolated laboratory specifications.

They should also examine how LG divides responsibility among affiliates and outside contractors. Integrated branding helps only if operational accountability remains clear during commissioning and failures.

The three-factory expansion strengthens the supply side of this strategy. It does not, by itself, prove that the complete thermal system will outperform rival designs.

The Investment Still Carries Demand and Execution Risk

LG is building capacity before the AI data center market has resolved its largest questions about power, project timing, and cooling architecture.

The strongest case for expansion begins with sustained computing investment. Hyperscalers and other operators continue planning facilities for training, inference, cloud services, and enterprise AI.

The IEA said five large technology companies spent more than USD 400 billion in capital during 2025. It expected that figure to rise another 75 percent in 2026.

Those totals cover more than data centers, and spending plans can change. They still illustrate the scale of infrastructure investment supporting equipment demand.

The risk is that announced data center pipelines exceed the number of projects that receive power, financing, permits, and customers. Cooling suppliers can face delays even when long-term demand remains strong.

Grid access represents a particularly serious constraint. Large facilities may wait years for transmission upgrades, substations, transformers, or generation resources.

The IEA expects global electricity consumption from data centers to double by 2030. Its analysis also emphasizes wide uncertainty around hardware efficiency, AI adoption, and infrastructure bottlenecks.

That uncertainty makes the LG Electronics chiller expansion a timing bet. The Virginia plant must reach productive utilization as customer projects move from plans into construction.

Production scheduled for the first half of 2027 gives LG limited time to staff, qualify, and stabilize the facility. Industrial equipment customers will expect consistent quality from its earliest shipments.

Component availability creates another risk. Chillers depend on compressors, motors, drives, heat exchangers, controls, refrigerants, and electrical equipment.

A local assembly plant does not make every upstream component local. Supply disruptions can still affect lead times if important parts come from concentrated sources.

The technology mix could shift as well. Some projects will use air-cooled chillers, while others may favor water-cooled systems, dry coolers, or hybrid arrangements.

Higher rack density will increase direct liquid cooling. Warmer coolant temperatures might also allow some facilities to reject heat without conventional mechanical chilling during favorable conditions.

That does not make chillers obsolete. It means demand will depend on geography, architecture, operating temperature, and redundancy choices.

Air-cooled chillers solve one water challenge but can face efficiency pressure during hot weather. Water-cooled alternatives may perform better in some climates while requiring cooling towers and water treatment.

LG will need a portfolio broad enough to avoid betting on one configuration. Its current mix of air-cooled, water-cooled, and liquid-loop products supports that flexibility.

Commercial execution remains uncertain because LG has not disclosed the planned output of its expanded network. The company also has not provided customer commitments tied directly to the Virginia factory.

Without those details, investors and buyers cannot calculate expected utilization. The KRW 150 billion commitment shows intent but not secured demand.

Competitive responses will add pressure. Schneider Electric can combine cooling with electrical distribution and automation, while Vertiv offers a broad infrastructure and service portfolio.

Specialized liquid-cooling companies may also move faster around cold plates, distribution units, and novel heat-rejection methods. Hyperscalers can influence designs through internal engineering and direct supplier relationships.

LG’s consumer brand will not decide these industrial purchases. Procurement teams will focus on technical qualification, delivery schedules, service coverage, and lifecycle performance.

The company’s Nvidia validation helps at the coolant-distribution level. Its Jakarta project can provide operational evidence for integrated deployments.

Even so, one certification and one reference project cannot settle a global competition. LG needs repeated installations across climates, customer types, and computing architectures.

The company must also avoid overstating certainty around AI demand. Rising model use supports infrastructure growth, but efficiency gains can reduce computing needs for individual tasks.

Lower costs can also increase total usage, offsetting those savings. The relationship between model efficiency and infrastructure demand remains difficult to forecast.

This is why production utilization matters more than the factory announcement alone. Consistent orders and on-time deliveries will show whether LG read the market correctly.

Three Signals Will Show Whether LG’s Cooling Bet Is Working

The next evidence should come from factory execution, customer adoption, and verified operating performance.

The first signal is the Windsor plant’s progress toward first-half 2027 production. LG must complete the facility, install equipment, qualify products, and build a local workforce.

Any delay would weaken the advantage of manufacturing close to North American customers. On-time production would support LG’s claim that it can respond more quickly to regional demand.

The quality of those first shipments will matter as much as their timing. Data center operators will expect stable performance, documentation, and service readiness from the new line.

The second signal is disclosed order growth for LG AI data center cooling. Investors should look for contracts rather than broad references to discussions with hyperscalers.

LG has relationships with major technology companies, but its October announcement did not identify new customers for the expanded capacity. Named projects would make the demand case easier to assess.

Order composition will also reveal whether the integrated strategy is working. Chiller-only deals would validate manufacturing demand without proving customer adoption of chip-to-chiller cooling.

Contracts combining chillers, coolant distribution units, controls, and service would provide stronger evidence. They would show that buyers see value in LG’s wider thermal architecture.

The third signal is operating data from real deployments. The Jakarta facility and early installations of LG’s new air-cooled centrifugal chiller can provide important evidence.

Useful measures include delivered cooling capacity, annual energy use, water consumption, recovery after faults, maintenance demands, and computing uptime. Independent customer data would carry more weight than internal simulations.

These signals should also be compared with rival progress. Schneider Electric’s Motivair integration will test how quickly an established infrastructure provider can scale liquid cooling worldwide.

Vertiv and other suppliers will continue expanding thermal portfolios and manufacturing. Their delivery times, reference projects, and technical partnerships will shape LG’s competitive position.

Broader infrastructure indicators remain relevant. The electricity outlook shows strong data center demand alongside tighter physical constraints.

If grid delays repeatedly postpone projects, cooling orders may shift even when long-term AI demand remains intact. If power connections accelerate, suppliers with available capacity can benefit sooner.

The LG Electronics chiller expansion therefore deserves attention as an industrial commitment, not just a product announcement. It shows that the AI race is moving into factories that make physical infrastructure.

The next question is whether LG can turn three production sites into an integrated cooling business with repeatable performance. Watch the Windsor schedule, contracted system orders, and verified field results.

Those three measures will separate manufacturing ambition from operational success. They will also show whether cooling capacity becomes a durable advantage in the next phase of AI infrastructure.