TrendForce Hosts Seminar on "AI Era and the Global Semiconductor Landscape: 2025 Industry Forecast"
TrendForce recently held a seminar titled "AI Era and the Global Semiconductor Landscape: 2025 Industry Forecast," covering a range of topics, including foundry, HBM, NAND Flash, AI PMIC, AI servers, panel-level packaging, liquid cooling, AI PCs, and more. The rapid growth of the artificial intelligence (AI) industry is poised to reshape the global technology landscape, exerting a broad and profound influence on numerous sectors. TrendForce’s seminar explored the impact of AI on the semiconductor industry and how it will lead the future of the sector.
TrendForce recently held a seminar titled "AI Era and the Global Semiconductor Landscape: 2025 Industry Forecast," covering a range of topics, including foundry, HBM, NAND Flash, AI PMIC, AI servers, panel-level packaging, liquid cooling, AI PCs, and more. The rapid growth of the artificial intelligence (AI) industry is poised to reshape the global technology landscape, exerting a broad and profound influence on numerous sectors. TrendForce’s seminar explored the impact of AI on the semiconductor industry and how it will lead the future of the sector.
Foundry
Predicting AI Industry Growth Through the Foundry Market by 2025
The rising demand for high-performance computing (HPC) chips, driven by AI applications, has become the primary force driving advancements in semiconductor processes and the overall foundry industry. Starting in 2025, AI chip vendors and cloud service providers (CSPs) will continue to develop custom chips, while memory suppliers are increasingly collaborating with advanced process foundries to meet the demand for high-performance computing. The foundry ecosystem, encompassing IP, design services, and back-end testing, has become a key resource in the AI arms race, extending beyond the focus on front-end process advancements.
Beyond advanced processes, a key focus is whether AI-driven demand for power management could revive interest in mature processes, and how the industry will evolve under the dual pressures of Cloud AI and Edge AI. TrendForce projects that by 2025, the global foundry market will grow by 20%, with TSMC maintaining its leading position. Other foundries are also expected to experience approximately 12% annual growth.
Mature Processes
AI PMIC Breathing New Life into Mature Processes
While global economic uncertainties persist, the primary growth driver for the foundry industry in 2024 will be AI server-related chips, with advanced process utilization rates remaining high through 2025. However, the recovery of 28nm and above mature processes will be slower, with an estimated capacity utilization increase of only 5-10%, reaching around 80%. Eight-inch wafer capacity utilization is forecast to remain around 75%, indicating the need for new growth drivers.
The continuous advancement of AI chips is pushing thermal design power (TDP) to new heights, requiring enhanced power management solutions. For example, NVIDIA's A100 GPU has a maximum TDP of 400W, with the H100 rising to 700W, and the next-generation Blackwell series expected to exceed 1,000W. The growing demand for Smart Power Stage (SPS) components to manage power transfer and reduce energy losses will drive a 2-3x increase in demand for related products between 2023 and 2025, becoming a crucial force in supporting mature process capacities.
Liquid Cooling
AI Propelling Liquid Cooling Technology to New Heights
Major U.S.-based cloud service providers such as Google, AWS, and Microsoft are actively building new data centers and accelerating the deployment of AI servers. As chip TDP continues to rise, driven by AI applications, liquid cooling is becoming essential to effectively manage heat dissipation. For instance, NVIDIA's newly introduced GB200 NVL72 cabinet solution has a TDP of around 140kW, which requires liquid-to-air (L2A) cooling. By 2025, as the GB200 solution scales up, the penetration rate of liquid cooling in AI chips is expected to rise from 11% in 2024 to 24% in 2025.
Among cloud providers, Google is leading the adoption of liquid cooling for high-end AI ASICs, while other providers still rely primarily on air cooling. Additionally, as global ESG awareness grows, liquid cooling adoption is set to increase, with power supply vendors, cooling system providers, and system integrators entering the market, fostering new industry dynamics.
AI Servers
Key Trends in the AI Server Market by 2025
Driven by strong demand from CSPs and enterprise customers for AI infrastructure, global AI server shipments (including those equipped with GPUs, FPGAs, ASICs, etc.) are expected to grow by 42% in 2024. By 2025, shipments could rise by another 28%, increasing AI servers' share of the overall server market to nearly 15%. In the longer term, as cloud-based AI training and inference services expand, AI servers are projected to account for nearly 19% of the server market by 2027.
NVIDIA is anticipated to maintain a near 90% market share in the GPU market by 2024. Its new Blackwell platform, expected to ramp up in 2025, will drive growth for both standalone GPUs (like the B200) and integrated solutions (such as GB200, featuring Grace CPUs). Competitors like AMD, Intel, and CSPs will also push forward with new AI chips, further boosting demand for CoWoS and HBM, which are expected to see shipment volumes double.
HBM
HBM: Challenges and Outlook for a Rapidly Growing Market
HBM remains a high-growth market, with AI server deployments demanding increased GPU performance and memory capacity. HBM plays a key role in supporting these advancements. For example, NVIDIA's Blackwell platform will use 192GB of HBM3e memory, while AMD’s MI325 will surpass 288GB. Despite the high complexity and yield challenges of HBM production, which result in average costs being 3-5 times that of DRAM, revenue contributions are expected to increase with HBM3e mass production.
NAND Flash
NAND Flash Market: Opportunities and Challenges Amid AI Growth
Following significant losses in 2023, NAND Flash suppliers have adopted a more conservative approach to capital expenditures. AI-driven demand for DRAM and HBM is expected to limit NAND Flash investment by 2025, easing the oversupply that has plagued the market.
The rise of AI has transformed the NAND Flash market, particularly with the growing demand for high-speed, high-capacity storage solutions, spurring growth in the Enterprise SSD (eSSD) segment.
Panel-Level Packaging (FOPLP)
Panel-Level Packaging: Potential and Challenges in AI Chip Integration
FOPLP applications are divided into three categories: PMIC and RF IC, consumer CPUs and GPUs, and AI GPUs. Each category is at a different stage of development, with AI GPU panel-level packaging expected to be adopted as early as 2027. However, challenges such as technological barriers and diverse panel sizes will likely slow progress.
AI PCs
AI PCs: Building a Foundation for Breakthrough Applications
AI capabilities are expected to become standard in laptops, with AI penetration in notebooks forecast to reach 21.7% by 2025 and nearly 80% by 2029. AI integration will improve user experience through enhanced voice recognition and natural language processing and provide more accurate business insights through data analysis. As the market evolves and AI applications mature, the future of AI PCs remains promising, offering new growth opportunities for a stable and mature market.
