AI Beneficiaries Beyond Nvidia: The 'Liquid Cooling' Infrastructure Supercycle Created by Power and Heat Constraints
As power consumption and thermal issues in AI data centers reach critical limits, 'Liquid Cooling' technology has rapidly emerged as essential infrastructure. We analyze the structural growth drivers and investment perspectives of the server and cooling system value chain, including Hewlett Packard Enterprise (HPE).
Act II of the AI Rally: The Shift from 'Chips' to 'Infrastructure'
While the first act of the artificial intelligence (AI) rally was driven by semiconductor companies led by Nvidia, the market's focus in 2026 is rapidly shifting toward the 'infrastructure' that physically powers AI. As the computational power required for inference and training of Large Language Models (LLMs) increases exponentially, the physical limitations of data centers supporting them have emerged as a new bottleneck.
The Limits of Air Cooling and the 60kW Wall
Traditional data centers have adopted air cooling methods, circulating cold air to dissipate heat from servers. However, the power density of modern AI racks—packed with high-performance accelerators like Nvidia's Blackwell architecture or AMD's MI300X—now frequently exceeds 60kW. This is more than triple the 15-20kW capacity limit that conventional air cooling systems can handle. Failure to control heat generation leads to chip performance degradation (throttling), sharply reducing the efficiency of heavily funded AI systems.
Why 'Liquid Cooling' Now?
'Liquid Cooling' systems are emerging as the only viable alternative to simultaneously solve power shortages and thermal issues. Liquid cooling controls heat by directly circulating coolant, which has a thermal conductivity over 3,000 times higher than air, to the chips. This goes beyond simply lowering temperatures; it is a core technology that reshapes the structural economics of data centers.
Improving PUE and Reducing Operating Costs
Adopting liquid cooling systems can significantly lower a data center's Power Usage Effectiveness (PUE) to below 1.2. This means minimizing the surplus power consumed for cooling outside of running the servers. According to analyses by global consulting firms, data centers adopting Direct Liquid Cooling (DLC) can reduce cooling power by up to 70% compared to traditional methods, allowing the secured surplus power to be allocated to additional AI computing resources.
HPE's Differentiation Strategy as a Key Player in Enterprise AI
Among the companies drawing the most attention in the AI infrastructure supercycle is Hewlett Packard Enterprise (HPE). Their high-density thermal management technology, accumulated from past experience in building supercomputers (Cray), perfectly aligns with the current demand for AI data centers.
100% Fanless Direct Liquid Cooling Technology
Through its AI server lineup, the ProLiant Compute XD series, HPE recently introduced a 100% fanless Direct Liquid Cooling (DLC) system to the market. Unlike existing hybrid cooling methods, it features the complete removal of cooling fans inside the server, extremely reducing power consumption. Furthermore, their smart cooling software technology—which combines 'Digital Twin' technology and AI to dynamically adjust coolant flow according to real-time workloads—is receiving a highly positive response from enterprise clients.
Infrastructure Value Chain Outlook from an Investment Perspective
The AI infrastructure market is not a one-off theme but is in the early stages of a massive Capital Expenditure (CapEx) cycle that will last for years. Global Cloud Service Providers (CSPs) are rushing to redesign their data centers around liquid cooling, driving structural growth not only for server manufacturers (such as HPE, Dell Technologies, and Supermicro) but also across the entire thermal management solution component ecosystem.
Investors need to broaden their portfolio perspective beyond relying solely on the earnings of AI semiconductor chipset companies, moving toward the infrastructure, power, and cooling value chains that are indispensable for physically realizing computational power. The companies that overcome the limitations of infrastructure will capture the practical profits of the upcoming AI ecosystem.