Huayuan Securities: The wave of high-power AI computing power and thermal interface materials ushered in domestic replacement opportunities

The Zhitong Finance App learned that Huayuan Securities released a research report saying that the global thermal interface materials industry has strong growth, and iterative AI computing power forms the core driving factor of the industry. Currently, the high-end TIM market is still dominated by overseas material giants. Relying on the development dividends of the local computing power industry chain, domestic manufacturers can achieve technological breakthroughs in the field of new thermal interface materials such as vertically oriented graphene and metal composites, or will accelerate the domestic substitution process. Industrial chain standards with independent material research and development capabilities and the collaborative layout of domestic and foreign markets may benefit. As a result, the communications industry maintained its “optimistic” rating.

The main views of Huayuan Securities are as follows:

AI computing power density continues to increase, and thermal management solutions are upgraded from server-end cooling to rack-level and near-chip collaboration

AI infrastructure has evolved from stand-alone servers to high-density racks and large-scale computing power clusters, and thermal management targets have also expanded from single chips and single servers to the rack and even data center level. As chip power consumption, cabinet power density, and cluster size increase simultaneously, thermal management needs to simultaneously enhance thermal conductivity, heat diffusion, and system-level cooling capabilities at the proximal end of the chip, and cooling standards continue to be upgraded. Looking at the heat transfer link, the chip heat is sequentially transferred through the package's internal conduction, interface heat transfer, heat diffusion, and finally discharged outward from the air cooling or liquid cooling system. TIM1, TIM1.5, and TIM2 are at different thermal interface positions, and each puts forward differentiated requirements for BLT, interface thermal resistance, warpage adaptation, and long-term reliability. The high-performance TIM evaluation system has moved from a single thermal conductivity index to comprehensive considerations of interface thermal resistance, Z-direction heat transfer capacity, mechanical adaptation, and thermal cycle stability.

System liquid cooling accelerates penetration, and low thermal resistance near the chip promotes continuous upgrading of high-performance TIM

The theoretical thermal conductivity of graphene can reach 2000-3000 W/ (m·k), but the native sheet structure only has high thermal conductivity within the surface; a continuous Z-guide heat channel can be constructed through vertical orientation and vertical stacking structural design, which fully unleashes the potential for heat transfer in the thickness direction. Compared with traditional polymer-based TIM, which can only reach a single digit to more than ten W/ (m·k) thermal conductivity levels, vertically oriented graphene TIM can achieve 680 W/ (m·k) level thermal conductivity. The highest thermal conductivity coefficient related to domestic manufacturer Hongfucheng products Up to 130 W/ (m·K), thermal resistance 0.04- 0.06 ℃ · cm2/W, also has excellent resilience, compression stress and thermal resistance stability, and can adapt to microscopic surface irregularities and chip warpage. Compared with traditional polymer TIM, it has outstanding performance advantages of high Z conduction heat and low interface thermal resistance.

AI computing power drives cooling demand to accelerate growth

Leading manufacturers such as Nvidia and Google continue to increase AI data center construction and iterate high-power computing power platforms, driving overall growth in cooling demand for entire cabinets and driving the continuous expansion of the global thermal management market. Leading overseas cloud vendors, including Microsoft, Google, Amazon, Meta, and Oracle, invested a total of 401.3 billion US dollars in fiscal year 2025. Google's 26FYQ2 earnings report further sharply raised the annual capital expenditure forecast, from the original $180-1900 billion to $1950-205 billion. AI computing power iterations have driven a sharp rise in the power consumption of AI chips from leading manufacturers. The Nvidia GPU single chip power consumption has surpassed from 400W of the A100 to the peak power consumption of 2300W on the Rubin platform, and Google's TPU has also iterated from 100 watts to kilowatts. The rise in chip power consumption is driving the power of the entire cabinet to continue to rise. Platforms such as the GB200, GB300, and Rubin Ultra have successively switched to rack-level all-liquid cooling architectures, and traditional air cooling has gradually reached the physical limit. Higher power in AI data centers has become a trend. It will continue to release demand for thermal management products such as cold plates and high-end thermal interface materials, and the full-chain cooling upgrade of the chip-interface-cabinet will open up room for growth in the industry.

The global thermal management space is vast, and domestic cooling suppliers are breaking through at an accelerated pace

According to MarketsAndMarkets data, the global thermal management market is about 10.7 billion US dollars in 2022, and is expected to reach 19.3 billion US dollars in 2028, corresponding to a CAGR of 9.7%; in terms of segmentation, QYResearch estimates that global sales of thermal interface materials in 2024 were US$2,012 billion, and it is expected to grow to US$4.148 billion in 2031. Leading overseas manufacturers still dominate the global TIM market. In 2024, DuPont, Dow Chemical, and Henkel had market shares of 11.5%, 6.48%, and 6.05% respectively. Companies such as Henkel, Shin-Etsu, Laird, and 3M continued to occupy the main market share of high-end TIM. Domestic manufacturers have accelerated their breakthroughs in the domestic market. According to Frost & Sullivan data, in 2025, Feirongda ranked first in the domestic TIM market with a market share of 13.1%. Local companies such as Zhongshi Technology, Siquan New Materials, and Hongfucheng have all achieved corresponding shares and are gradually participating in market competition.

Risk Alerts

The risk that global AI computing power capital expenditure falls short of expectations, the risk of price fluctuations in commodities and key raw materials, the risk of high-end customer certification and implementation falling short of expectations, and the risk of industry technology route iteration.