Sinolink Securities Co., Ltd. released a research report stating that AI computing infrastructure buildout continues to drive escalating demand for chip thermal management, with single-GPU power consumption rising from 400W to 1400W. TIM has become a critical material for ensuring the performance and reliability of high-power chips, and the carbon-based route is expected to open a window for domestic breakthroughs, with clear industrialization opportunities for graphene thermal pads. The report recommends focusing on companies that have actively positioned themselves in graphene thermal pads and hold first-mover advantages.

The main points of the Sinolink Securities Co., Ltd. report are as follows.

AI computing upgrades are driving increased chip thermal management demand, and the performance bottlenecks of traditional TIM are gradually becoming apparent. Microscopic undulations exist on the surfaces of chips, package lids, and heat sinks. TIM fills interfacial gaps to transfer heat from the chip to the heat sink or liquid cooling plate, and its performance directly affects chip temperature and long-term operational reliability. In terms of packaging structure, lidless packaging typically uses TIM1 to connect the chip to the package lid and TIM2 to connect the lid to the heat sink; lidless packaging employs TIM1.5 to directly connect the bare die to the heat sink or cold plate, shortening the heat transfer path but imposing higher requirements on material softness, interfacial conformity, and assembly stress. NVIDIA's single-GPU power consumption has increased from 400W for the A100 to a maximum of 1400W on the GB300 platform. Meanwhile, 2.5D/3D packaging integrates chips, HBM, and other components into larger-area packages, where differences in material thermal expansion and package warpage increase the difficulty of interfacial conformity. Therefore, TIM must not only rapidly dissipate heat but also adapt to interfacial deformation and maintain long-term stability. Traditional polymer-based TIM such as silicone grease and gel is limited by low thermal conductivity of the matrix, and adding thermally conductive fillers tends to increase material stiffness. Although materials such as liquid metal offer higher thermal conductivity, they still face challenges including electrical conductivity, leakage, corrosion, and assembly compatibility, driving continuous iteration of high-performance TIM.

Graphene TIM combines efficient thermal conduction, soft conformity, and long-term stability, with industrialization steadily advancing. The advantages of graphene TIM materials are reflected in the following aspects: (1) Strong thermal conductivity. Graphene has high intrinsic in-plane thermal conductivity, and through structural design, heat transfer pathways in the thickness direction are formed. Publicly disclosed high-performance graphene thermal pads have thermal conductivity of approximately 90-200W/(m·K). (2) Good interfacial conformity. After compositing graphene with a flexible matrix, the pad possesses compression-rebound capability and can adapt to warpage and assembly gaps in large-size packages. (3) Good long-term stability. The flexible solid-state structure helps reduce material migration and pump-out during thermal cycling, maintaining interfacial contact. The technology route has evolved from early dispersion of graphene as a filler in polymers to the construction of three-dimensional continuous thermal conduction networks and control of vertical graphene orientation, converting the material's in-plane thermal conduction advantage into the longitudinal thermal conduction capability required by chips. On the market side, AI infrastructure buildout is driving deployment of high-power chips, with growing demand for high-performance TIM in applications such as GPUs, ASICs, and server CPUs, extending to scenarios including optical modules and power semiconductors. According to QYResearch data, the global TIM market is expected to grow from approximately USD 2.012 billion in 2024 to approximately USD 4.148 billion in 2031, with a CAGR of approximately 10.7% from 2025 to 2031. Market expansion and increasing penetration of high-end materials are expected to jointly open up growth space for graphene TIM.

The carbon-based route opens a window for domestic breakthroughs, and first-mover graphene thermal pad companies have significant positioning advantages. The traditional TIM market is dominated by international giants such as DuPont, Dow, and Henkel, with a relatively entrenched competitive landscape. However, as an emerging niche direction, graphene thermal pads have an unsettled market landscape, and domestic manufacturers are expected to break through first. Graphene thermal pads possess triple barriers in material design, mass production processes, and customer certification, and currently few manufacturers have achieved large-scale batch supply. Among them, Hongfucheng is one of the very few companies in the industry capable of mass-producing graphene thermal pads and achieving scaled application. The company mass-produces 130W/(m·K) vertically oriented graphene thermal pads with thermal resistance as low as 0.04-0.06℃·cm2/W, has already supplied in batches to global leading AI chip customers, and has achieved a breakthrough in small-batch supply of TIM1 thermal interface materials. Benefiting from product ramp-up, the company's carbon-based thermal pad revenue increased from RMB 35.61 million in 2023 to RMB 252.266 million in 2025, with year-over-year growth exceeding 138% for two consecutive years. In 2025, gross margin was 65.6% and non-GAAP net margin was 37.4%, significantly leading peers. Meanwhile, domestic manufacturers such as Sinochem Technology, Siquan New Materials, and Feirongda are also accelerating their layout in vertically oriented graphene TIM, and the industrialization process is expected to accelerate. As AI computing thermal management demand continues to grow, graphene thermal pads are expected to become a key entry point for domestic manufacturers to accelerate their breakthrough into the high-end segment.

Risk warnings: Risk of AI chip demand falling short of expectations; risk of customer validation and batch introduction falling short of expectations; risk of mass production and cost control falling short of expectations; risk of substitute material competition and penetration falling short of expectations.