Core Insights - The AI industry faces significant challenges primarily related to power supply rather than chips or talent, highlighting the critical role of power management in AI servers [1] - The global AI server power market is projected to grow exponentially, from $7.4 billion in 2025 to $32.5 billion by 2027, with a compound annual growth rate (CAGR) of 110% [1] - Domestic manufacturers are increasingly entering the AI server power supply market, with current domestic replacement rates estimated at 10-20% [1] Market Trends - AI server power architecture is undergoing a transformation, with three key levels: power supply systems (UPS, HVDC, PDU), AC-DC conversion (PSU, PMC, BBU), and DC-DC conversion (PDB, VRM) [3] - The shift towards high-voltage direct current (HVDC) solutions is gaining momentum, improving system efficiency from 90-95% to 95-98% [3] - Centralized power architectures are being adopted, with PSU power increasing from 5.5kW to over 22kW, and output voltage rising from 48V to 400V [4] Technological Innovations - The use of wide bandgap semiconductor devices like SiC and GaN is becoming central to AI power design, offering higher switching frequencies and lower conduction losses compared to traditional silicon devices [4] - Companies like Jiehua and Chipone Micro are developing advanced power management solutions, including high-current multi-phase power supplies and digital power controllers [5][20] - The introduction of intelligent power stages (DrMOS) is addressing the increased power density requirements of AI servers, with products like Jiehua's 90A DrMOS achieving high efficiency and reliability [7] Company Developments - Jiehua is focusing on comprehensive analog chip solutions for AI servers, with products that have reached international advanced levels [5] - Chipone Micro has launched a series of power chips for AI servers, emphasizing efficiency and reliability in response to rising power density demands [20][22] - Jingfeng Mingyuan has entered the NVIDIA recommended supplier list and is expanding its high-performance computing power chip offerings, achieving significant revenue growth [24][26] Conclusion - The competition in AI is not only about algorithms and chips but also about efficiently and sustainably powering these computational giants, with the value of power management chips for AI servers being 3 to 5 times that of traditional servers [30] - The rise of intelligent power stages (DrMOS), multi-phase controllers, and hot-swappable eFuse as core components for AI servers is evident, with domestic manufacturers increasing their focus on these trends [31]