Samsung and LG scale liquid cooling for AI data centers

In Pune, India, Samsung’s FlaktGroup completed a new production line on Aug. 12, after construction and equipment installation began early this year. The line has reached mass-production readiness and, at full capacity, can produce up to 6,500 units annually for India and the wider Asia-Pacific region. It is one visible sign that cooling equipment is becoming part of the AI infrastructure buildout.

The pressure comes from the servers themselves. Nvidia’s B200 and other next-generation AI chips consume more than 1,000 watts each, pushing data centers toward the practical limits of air cooling. In a direct-to-chip system, cold plates draw heat from processors; a coolant distribution unit, or CDU, controls the coolant’s temperature and pressure, removes impurities and circulates the fluid to a chiller before sending it back through the system.

The market is moving quickly. TrendForce estimates that liquid cooling was used with 33 percent of AI chips in 2025, with the share projected to reach 53 percent this year and nearly 60 percent in 2027. LG’s data-center chiller orders tripled in 2025 from the previous year, and the company expects to reach 1 trillion won ($730 million) in annual chiller revenue next year. LG is also considering a first US chiller plant, with Virginia among several locations under review.

Samsung is building a second manufacturing foothold in Gwangju, South Korea. It is investing about 240 billion won in an HVAC production line at the third campus of its Gwangju plant, covering 21,800 square meters. Production is scheduled to begin in 2028, initially focusing on FlaktGroup CDUs; the site would become FlaktGroup’s 15th manufacturing base.

Concretely, the expansion gives hyperscalers, colocation operators and other AI infrastructure builders more potential suppliers and production closer to major markets. LG’s Nvidia-certified 600-kilowatt CDU also places a Korean-made component inside the supply chain for systems linking chillers to server cold plates. The immediate gain is not more computing by itself, but more equipment available to remove the heat that limits dense AI deployments.

The timetable is not uniform. LG’s US factory remains a consideration rather than an operating site, Samsung’s Gwangju line is still scheduled for 2028, and Kyobo Securities’ forecast that LG could secure 2.5 trillion won in new orders by October has not been confirmed by LG. The direction is clear in the production plans, while the final scale of demand remains a forecast.