A Follow-On to the Thermal Inflection Thesis
GZC published initial research on the liquid cooling transition in March 2025. That piece established the thesis: as GPU rack densities exceeded the thermal limits of air cooling, the data center industry was forced to transition to liquid cooling systems on a timeline that the supply chain was not prepared to serve.
This piece is an economic follow-on — not a repeat of the supply chain analysis, but an examination of the economics: where does the value accrue in the liquid cooling supply chain, what do sustainable margins look like, and what is the duration of the investment opportunity?
The Thermal Economics Framework
Liquid cooling for data centers is not a single product category. It is a hierarchy of solutions, each serving a different thermal management requirement:
Rear-door heat exchangers: The least invasive form of liquid cooling. A liquid-cooled door attaches to an existing server rack and captures heat exhaust before it enters the data center airspace. This is an additive solution — it works with existing air-cooled servers without redesigning the server platform. The economics are favorable for data center operators because it requires minimal infrastructure modification.
Direct liquid cooling (DLC): Liquid is circulated through cold plates attached directly to heat-generating components. DLC is more thermally efficient than rear-door approaches and is increasingly required for GPU rack densities above 40 to 60 kilowatts per rack. Server manufacturers must design DLC compatibility into the server platform, creating a longer upgrade cycle but also deeper integration.
Immersion cooling: The highest-performance thermal management approach — servers are submerged in dielectric fluid that absorbs heat directly. Immersion cooling enables the highest rack densities and is increasingly considered for the highest-wattage GPU configurations, but it requires complete data center redesign and currently carries higher operational complexity.
Where Value Accrues
The economics of the liquid cooling supply chain vary significantly across solution types. The most important observation is that the thermal management supply chain has historically been a low-margin, commoditized components business. Air cooling — fans, heat sinks, airflow management — was not a high-value-add category. Liquid cooling is different.
System integration is where the margin lives: The critical differentiator in liquid cooling is not the commodity components — the pipes, the coolant, the pumps. It is the system engineering that ensures reliable operation in a 24/7 mission-critical environment. The companies that own the system integration relationship with major data center operators are the ones that capture durable margin, because the switching cost of changing cooling infrastructure systems in an operating data center is extremely high.
Service and maintenance creates recurring revenue: Air-cooled data centers require relatively simple maintenance. Liquid cooling systems require more sophisticated monitoring, maintenance contracts, and fluid management. This service layer represents a recurring revenue opportunity for companies that build it into their customer relationships from the beginning.
Specification is a moat: The hyperscalers are developing their own liquid cooling specifications — defining the performance, reliability, and integration requirements that cooling systems must meet. Companies that achieve certification for these specifications gain a structural competitive advantage: alternatives must go through the same qualification process.
Duration of the Opportunity
The liquid cooling market is in its early stages. The transition from primarily air-cooled to primarily liquid-cooled data center infrastructure is a 5-to-10 year process — not because the technology is immature, but because data center infrastructure replaces on a long cycle. Legacy data centers built with air cooling will serve their useful life before being replaced or retrofitted.
The forced-spend component of the liquid cooling thesis is in new construction and high-density retrofit. Every new large AI data center being built today is being designed with liquid cooling as a core infrastructure requirement rather than an optional upgrade. This represents a permanent demand floor for liquid cooling systems in new AI infrastructure that does not depend on retrofit decisions.
BTT Assessment
Applying the BTT framework to liquid cooling: the demand is non-discretionary for new AI data center construction at high GPU densities. The supply is concentrated among a small number of companies with the engineering expertise, customer relationships, and qualification status to serve hyperscale procurement. The margin opportunity is real but requires differentiation at the system integration and service layer, not the commodity component layer.
Our coverage focuses on companies in the system integration and thermal management positions, not commodity component suppliers. The duration of the opportunity, combined with the sticky customer relationships that thermal management contracts create, makes this one of the longer-duration forced-spend themes in our Technology pool.
