AI Compute Demand Ignites the 'Liquid Cooling 2.0' Era: Why the Underlying Logic of Chip Investing is Shifting from Process Scaling to Cooling and Energy Efficiency Revolution

In midsummer 2026, the global semiconductor industry is undergoing an unprecedented paradigm shift. For decades, Moore's Law has driven the chip industry forward, and investors have habitually focused on increasing transistor density, shrinking process nodes, and doubling transistor counts. However, as the parameter scale of AI large models breaks through the tens-of-trillions level, the power consumption of AI compute clusters is soaring at a rate that makes traditional thermodynamic engineers take notice. Today, what determines whether an advanced AI chip can be successfully shipped and run stably is no longer just its compute metrics, but whether its Thermal Design Power (TDP) can be effectively controlled. In this compute arms race, cooling technology is leaping from a "marginal supporting facility" in data centers to the "core infrastructure" that determines the life and death of chips. The underlying logic of why we invest in chips is irreversibly shifting from process scaling to a cooling and energy efficiency revolution.

The End of the Air Cooling Era and the Full-Scale Outbreak of "Liquid Cooling 2.0"

Before exploring the current investment logic of the chip industry, we must understand a harsh physical reality: the single-card TDP of top-tier AI accelerators has generally breached the 1000-watt mark, and some customized versions for ultra-large training clusters are approaching 1500 watts. At such high heat density, traditional air cooling technology is not only inefficient, but the energy consumption and noise of the fans themselves have reached the limits of engineering feasibility. When heat cannot be dissipated in time, the transistors inside the chip will downclock or even be damaged due to thermal runaway, meaning expensive compute resources are wasted.

Therefore, 2026 has become the turning point for the comprehensive popularization of liquid cooling technology. Unlike the previous generation of "Liquid Cooling 1.0," which merely attached cold plates to CPUs or GPUs, the "Liquid Cooling 2.0" era is characterized by system-level reconstruction and the rise of immersion cooling. Although cold plate liquid cooling currently dominates as a transitional solution, it still harbors hidden risks such as the aging of Thermal Interface Materials (TIM) and the clogging of micro-channels in the cold head. Meanwhile, more aggressive immersion cooling technology—completely submerging the entire server motherboard in specially designed fluorinated fluids or synthetic oils—is accelerating its penetration in major hyperscale data centers. This technology completely eliminates the thermal resistance of air as a heat transfer medium, allowing the PUE (Power Usage Effectiveness) value to easily drop below 1.05. For chip design companies, this means they can push chip frequencies to higher limits without sacrificing lifespan.

Thermal Management Reconstructs the Semiconductor Industry Chain: From Auxiliary Parts to Core Assets

In the traditional valuation model of the semiconductor industry chain, cooling modules are often viewed as low-value-added machined or assembled parts with extremely low gross margins, rarely entering the视野 of core semiconductor investors. However, when cooling becomes a compute bottleneck, the value distribution of the entire industry chain is undergoing a drastic reconstruction. This is not just a victory for materials science, but a contest of system-level engineering capabilities.

1. Materials: Accelerated Domestic Substitution of Thermal Interface Materials and Coolants

In the Liquid Cooling 2.0 era, Thermal Interface Materials (TIM) and coolants have become the new "silicon wafers." In the past, the high-end fluorinated fluid market was almost monopolized by international giants like 3M. However, as 3M gradually exits certain PFAS (Per- and Polyfluoroalkyl Substances) product lines due to environmental concerns, a massive gap has appeared in the global supply chain. For material manufacturers in the Asia-Pacific region, this is a once-in-a-lifetime substitution opportunity. Enterprises with the R&D capability for coolants featuring low viscosity, high specific heat capacity, environmental friendliness, and insulation are enjoying high valuation premiums. Meanwhile, materials used to fill the tiny gaps between chips and cold plates, such as liquid metal thermal paste and high thermal conductivity silicone pads, have technical barriers no less than those of photoresists. Investors need to closely monitor chemical and new material enterprises with deep patent reserves in the field of thermal management materials.

2. Equipment: System-Level Integration from Chip Design to Cabinet Design

Currently, advanced chip design cannot proceed in isolation from the cooling system. Before tape-out, chip design companies must engage in deep joint design (Co-design) with thermal management suppliers. From the micro-bump layout inside the chip, to the thermal layer planning of the package substrate, and to the micro-channel water routing design of the external cold plate, thermal management has become a core parameter in the chip definition phase. This leads to a direct result: ODM/OEM manufacturers with system-level cooling solution capabilities have significantly increased their bargaining power in the data center supply chain. They are no longer just foundries, but integrators of compute systems. In the precision manufacturing and technology sector listed on the Singapore Exchange (SGX), tech companies deeply engaged in data center thermal management structural parts, Quick Disconnects (QD), and liquid cooling cabinets have order visibility extending to 2028.

3. Operations: Energy Efficiency is Compute, Cooling Determines TCO

For cloud service providers and large AI model developers, cooling energy consumption accounts for a very high proportion of the Total Cost of Ownership (TCO) of a data center. Under the "Dual Carbon" goals and the pressure of continuously rising electricity prices, whoever can dissipate more heat with lower energy consumption will gain an advantage in the bidding for AI inference and training. Therefore, the advancement of cooling technology directly determines the compute output efficiency and Return on Investment (ROI) of a data center. This is why top cloud vendors are sparing no effort to develop their own liquid cooling architectures, even exporting them as core competencies.

Why the Underlying Logic of Investing in Chips is Being Rewritten

For a long time, the answer to "why invest in chips" usually revolved around process technology, compute barriers, ecosystem moats, and domestic substitution. These logics still hold true today, but they are no longer sufficient to explain all industrial changes. In an era of compute inflation, if the compute growth of a chip is not accompanied by a cooling system, it is merely laboratory data that cannot be converted into commercial value. Therefore, the underlying logic of investing in chips must incorporate a "thermodynamic dimension."

First, cooling technology is redefining the available compute of a chip. A chip with a theoretical compute of 1000 TOPS, if limited by cooling to only run stably at 700 TOPS, sees its actual commercial value significantly discounted. Advanced thermal management technology can unlock the hidden compute of a chip, which is equivalent to creating additional compute supply without increasing silicon area and manufacturing costs. This path of "trading cooling for compute" has a marginal cost far lower than the cost of simply upgrading process technology.

Second, thermal management is reshaping the competitive landscape of the chip industry. In the past, the competition among chip giants was about process nodes; now, this competition has extended to system-level thermal design capabilities. Manufacturers capable of providing an overall solution of "chip + package + cooling" will have stronger pricing power. For example, certain top-tier AI chip companies not only design chips but also design matching dedicated liquid cooling cabinets and interconnect topologies. This deep binding means that once customers adopt their cooling architecture, they incur extremely high switching costs, thereby forming a new ecosystem moat.

Finally, from the perspective of selecting investment targets, the traditional framework of solely staring at wafer foundries and chip design has become thin. The hidden champions along the thermal management chain—those mastering core material formulas, precision processing techniques, and system-level simulation technologies—are becoming rising stars in the semiconductor investment map. These enterprises often possess strong cross-disciplinary capabilities, understanding materials science, fluid mechanics, and electronic engineering. Once their technical barriers are established, they are difficult to replicate easily.

Opportunities and Challenges in the SGX Tech Sector: Seeking Alpha Under Thermodynamic Limits

For investors focusing on the technology and semiconductor sector of the Singapore Exchange (SGX), this structural change triggered by cooling is both an opportunity and a challenge. As an important global hub for semiconductor packaging, testing, and precision manufacturing, Singapore hosts numerous hidden champions deeply engaged in thermal management, precision parts, and electromechanical integration.

Under the wave of "Liquid Cooling 2.0," the investment logic of the SGX tech sector can be unfolded from three dimensions:

• Precision Manufacturing and Quick Disconnect Track: One of the core pain points of liquid cooling systems is the risk of leakage, which requires quick disconnects to have extremely high sealing and pressure resistance. If precision manufacturing companies listed on SGX can enter the liquid cooling quick disconnect supply chain of global top-tier AI server manufacturers, they will see significant performance increments.
• Cooling Modules and Vapor Chamber (VC) Technology: Inside chip packages, vapor chambers and heat pipes remain indispensable near-end cooling components. Enterprises with mass production capabilities for ultra-thin vapor chambers and able to adapt to the thermal stress challenges of high-density packaging possess long-term investment value.
• Testing and Reliability Verification Services: The introduction of new cooling materials and architectures poses new tests for the long-term reliability of chips. Third-party verification service institutions with thermal-mechanical-electrical multi-physics coupling testing capabilities will play an increasingly important role in the industry chain.

However, investing in the thermal management track is not without risks. Rapid technological iteration means that today's advanced solutions may be replaced in a year or two by more aggressive immersion or even phase-change cooling technologies. Furthermore, the cooling industry is essentially still constrained by cost-sensitive data center procurement budgets, and the risk of price wars cannot be ignored. Therefore, when making investments, investors should prioritize leading enterprises with underlying material innovation capabilities, the ability to conduct joint R&D with chip giants, and a diversified customer structure.

In summary, the semiconductor industry in 2026 has crossed the era of simply competing on transistor counts. Amidst the endless expansion of AI compute demand, the laws of thermodynamics have become the Sword of Damocles hanging over the heads of all chip designers. Understanding the bottleneck of cooling and energy efficiency means understanding the true driving force of the chip industry's evolution over the next five years. Why invest in chips? Because at the edge of this physical limit, only through system-level innovation and a thermal management revolution can the afterglow of Moore's Law be transformed into sustained commercial value. This is not just a long march of technology, but a brand-new starting point for the value reassessment of the global semiconductor industry chain.