Zhi Tong Cai Jing·2025-11-27 06:37Core Insights - The global AI server shipments are expected to grow by over 20% year-on-year in 2026, driven by increased capital expenditure from North American CSPs and the rise of sovereign cloud initiatives [1][2] - The competition in the AI chip market is intensifying, with companies like AMD and various Chinese firms enhancing their self-developed ASIC capabilities, challenging NVIDIA's dominance [1][2] Group 1: AI Chip and Cooling Technologies - The thermal design power (TDP) of AI chips is projected to rise from 700W for NVIDIA's H100 and H200 to over 1,000W for upcoming models, necessitating liquid cooling systems in server cabinets, with a forecasted penetration rate of 47% for liquid cooling in AI chips by 2026 [2] - Microsoft is introducing new microfluidic cooling technologies for next-generation chip packaging, while the market is expected to transition from liquid-to-air (L2A) to liquid-to-liquid (L2L) cooling designs [2] Group 2: Memory and Data Transfer Innovations - HBM and optical communication technologies are becoming critical for overcoming bandwidth limitations in AI computing, with HBM4 expected to enhance I/O bandwidth and local bandwidth for AI chips [3][4] - The introduction of 800G/1.6T pluggable optical modules is underway, with expectations for higher bandwidth SiPh/CPO platforms to be integrated into AI switches starting in 2026 [4] Group 3: NAND Flash and Storage Solutions - NAND Flash suppliers are accelerating the development of specialized solutions to address the performance gap in AI training and inference workloads, including storage-class memory (SCM) SSDs and Nearline QLC SSDs [5][6] - QLC technology is anticipated to achieve a 30% market penetration in enterprise SSDs by 2026, significantly reducing the cost of storing large AI datasets [6] Group 4: Energy Storage Systems - AI data centers are evolving towards large-scale clusters, with energy storage systems transitioning from emergency backup to core energy solutions, expected to grow from 15.7 GWh in 2024 to 216.8 GWh by 2030, reflecting a CAGR of 46.1% [7] - North America is projected to become the largest market for AI data center energy storage, driven by major cloud providers [7] Group 5: Power Infrastructure and Semiconductor Demand - Data centers are shifting to 800V HVDC architectures to enhance efficiency and reliability, with third-generation semiconductors (SiC/GaN) expected to penetrate 17% of data center power supply by 2026 [8] Group 6: Advanced Semiconductor Technologies - The transition to 2nm GAAFET technology is underway, emphasizing higher transistor density and heterogeneous integration to meet the demands of AI applications [9] Group 7: Humanoid Robots and Market Growth - The global shipment of humanoid robots is projected to increase by over 700% in 2026, focusing on AI adaptability and application-specific designs [10][11] Group 8: Display Technology Advancements - OLED technology is set to accelerate in laptops, with Apple expected to introduce OLED panels in MacBook Pro by 2026, leading to a significant increase in OLED penetration in the laptop market [12][13] Group 9: Autonomous Driving and Robotaxi Expansion - The penetration rate of L2 and above advanced driver-assistance systems is expected to exceed 40% by 2026, with Robotaxi services expanding globally beyond just China and the US [15]