【台北訊】
功率半導體設計公司海馳科技（AzureMoto Technology, Inc.）受邀參加凱基證券
「2026 Q1 Taiwan Corporate Day」投資論壇，並由執行長林橙寬以
「綠色 AI 時代的推手：從矽到寬能隙的功率半導體演進」為題發表專題演講。
隨著生成式 AI、大型語言模型與高效能運算（HPC）需求快速成長，全球資料中心正面臨前所未有的電力需求與能源效率挑戰。功率半導體設計公司海馳科技（AzureMoto Technology, Inc.）表示，AI 產業的競爭核心正逐步從單純算力，轉向「能源效率與電力架構」的競爭，而功率半導體正是支撐 AI 時代的重要關鍵基礎設施。
根據產業研究，AI GPU 的功率需求正快速上升，從過去數百瓦提升至 700W–1200W，未來甚至可能突破 2kW 等級。這使得 AI 伺服器機櫃功率密度由傳統資料中心的 5–10kW，大幅提升至 100kW–300kW 等級。當資料中心功率密度持續提升時，電源效率、散熱能力與電力架構設計將成為限制 AI 算力成長的關鍵因素。
海馳科技執行長林橙寬（Stanley Lin）表示：
「AI 的競爭是算力競賽，而算力的本質其實是能源競賽。未來 AI 的發展深度，將取決於電力效率與功率半導體技術的突破。」
AI 電力架構進入高壓直流時代
為支撐 AI 算力需求快速成長，資料中心電力架構正從傳統 12V 電源系統，逐步升級至 48V 高電流母線，並進一步邁向 400V 至 800V HVDC（高壓直流）供電架構。
高壓直流電系統可顯著降低電流與銅損，並提升整體電源效率，使 AI 伺服器能在高功率密度環境下穩定運作。
在此電力架構演進過程中，功率半導體元件成為整體系統效率的核心，包括 MOSFET、SiC（碳化矽）與 GaN（氮化鎵）等先進材料元件。
寬能隙半導體加速導入 AI 電源
為滿足高效率與高功率密度需求，AI 電源系統正快速導入第三代半導體材料（Wide Bandgap Semiconductor）：
SiC（碳化矽）
主要應用於資料中心電源供應器（PSU）與高壓電源轉換模組，具備高耐壓、低開關損耗與高溫運作能力。
GaN（氮化鎵）
廣泛應用於 48V 高頻 DC-DC 轉換模組，可在更高開關頻率下運作，顯著提升功率密度並縮小電源模組尺寸。
Si MOSFET / DrMOS
在 GPU 核心供電等低電壓、高電流場景中仍具有優異的成本效益與可靠度。
因此，AI 電源架構正逐步形成「SiC + GaN + Si Hybrid」的混合技術路線，以同時兼顧效率、成本與系統可靠度。
AI 電源技術的三大關鍵突破
海馳科技分析，未來 AI 資料中心電源技術將聚焦三大方向：
1. 高壓電力架構
AI 機櫃供電正逐步邁向 800V HVDC 架構，以降低電流與傳輸損耗。
2. 高功率密度與散熱技術
隨著 GPU 功率突破 1kW，資料中心正從傳統空冷系統轉向液冷與浸沒式冷卻技術。
3. 系統級整合能力
未來競爭將不再只限於單一元件，而在於功率器件、控制 IC 與散熱材料的系統整合能力。
AI 時代的能源效率競賽
海馳科技指出，AI 資料中心的電力需求正快速成長，能源效率將成為未來科技產業的重要競爭指標。透過寬能隙半導體與高壓電力架構，AI 電源效率可由傳統約 92–94% 提升至 97% 以上，顯著降低資料中心能源消耗。
海馳科技將持續投入先進功率半導體研發，以高效率、高可靠度的電源解決方案，推動「綠色 AI 基礎設施」的發展。
關於海馳科技（AzureMoto Technology, Inc.）
海馳科技為專注於功率半導體設計的科技公司，產品涵蓋 MOSFET、SiC、GaN FET、Rectifier 與 ESD 保護元件，主要應用於 AI 伺服器、汽車電子、工業控制與高效率電源系統。
公司致力於透過創新功率半導體技術，打造下一代高功率密度與高效率的電力基礎設施。
[Taipei, Taiwan] – AzureMoto Technology, Inc. (“AzureMoto”), a leading power semiconductor design company, was invited to the KGI Securities "2026 Q1 Taiwan Corporate Day" investment forum. CEO Stanley Lin delivered a keynote speech titled "The Driving Force of the Green AI Era: The Evolution of Power Semiconductors from Silicon to Wide Bandgap (WBG)."
As demand for Generative AI, Large Language Models (LLM), and High-Performance Computing (HPC) surges, global data centers face unprecedented challenges in power consumption and energy efficiency. AzureMoto stated that the core of AI industry competition is shifting from raw computing power to "energy efficiency and power architecture," positioning power semiconductors as critical infrastructure for the AI era.
Industry research indicates that AI GPU power requirements are rising sharply—from hundreds of watts to 700W–1200W, with future projections exceeding 2kW. Consequently, AI server rack power density is jumping from the traditional 5–10kW to 100kW–300kW. As density increases, power efficiency, thermal management, and power architecture design become the ultimate bottlenecks for AI scaling.
"AI competition is a race for computing power, but the essence of computing power is an energy race," said Stanley Lin, CEO of AzureMoto. "The future depth of AI development will be determined by breakthroughs in power efficiency and semiconductor technology."
Transitioning to High-Voltage DC Power Architectures
To support rapid AI growth, data center architectures are upgrading from traditional 12V systems to 48V busbars, moving toward 400V to 800V HVDC (High-Voltage Direct Current) systems. HVDC significantly reduces current and copper loss while enhancing overall efficiency, ensuring stable operations in high-density environments. In this evolution, power semiconductor components—including MOSFETs, Silicon Carbide (SiC), and Gallium Nitride (GaN)—sit at the heart of system efficiency.
Wide Bandgap (WBG) Semiconductors Accelerating AI Power Adoption
To meet demands for high efficiency and power density, AI systems are rapidly adopting Third-Generation (WBG) materials:
- SiC (Silicon Carbide): Applied in Power Supply Units (PSU) and high-voltage conversion modules, offering high voltage resistance, low switching loss, and high-temperature stability.
- GaN (Gallium Nitride): Used in 48V high-frequency DC-DC modules to enable higher switching frequencies, significantly increasing power density while shrinking module size.
- Si MOSFET / DrMOS: Remains cost-effective and reliable for low-voltage, high-current scenarios like GPU core power delivery.
The industry is gravitating toward a "SiC + GaN + Si Hybrid" roadmap to balance efficiency, cost, and reliability.
Three Key Breakthroughs in AI Power Technology
AzureMoto identifies three focus areas for future AI data center power:
- High-Voltage Architectures: Moving toward 800V HVDC to minimize transmission losses.
- High Power Density & Thermal Management: As GPUs exceed 1kW, cooling is shifting from traditional air-cooling to liquid and immersion cooling.
- System-Level Integration: Competition is moving beyond individual components toward the integrated synergy of power devices, controller ICs, and thermal materials.
The Energy Efficiency Race
AzureMoto points out that as AI power demand grows, energy efficiency will become a top-tier competitive index. Through WBG semiconductors and high-voltage architectures, AI power efficiency can be boosted from 92–94% to over 97%, significantly reducing carbon footprints. AzureMoto remains committed to developing advanced power solutions to drive the development of "Green AI Infrastructure."
About AzureMoto Technology, Inc.
AzureMoto is a power semiconductor design company specializing in MOSFETs, SiC, GaN FETs, Rectifiers, and ESD protection components. Its products serve AI servers, automotive electronics, industrial control, and high-efficiency power systems. The company is dedicated to building next-generation high-density and high-efficiency power infrastructure through innovation.