台灣電力市場的未來挑戰：夜尖峰供需缺口與備用容量率之預估

Abstract

在「算力即國力」趨勢驅動下，人工智慧（AI）與高效能資料中心對電力的全天候需求急遽攀升。這些負載多為全天候用電需求， 晚間運行時將與太陽光電發電高峰出現時序錯配，進一步壓縮台灣本就吃緊的夜尖峰供電餘裕。 研究方法：係整合歷年 《全國電力資源供需報告》 與 《長期電源開發計畫》，納入機組歲修、退役排程與招標延宕，並結合高科技產業擴產資訊，建立 10 年期夜尖峰供需模型，追蹤備用容量率變動。 需求面分析顯示， 半導體先進製程與 AI 雲端資料中心屬 24 小時高負載 ，其新增電力需求在 2030 年前占全國增量 82%，帶動夜尖峰負載自 2024 年的 3,725 萬瓩升至 2033 年的 4,766 萬瓩， 十年間成長 31.8% 。供給面挑戰則包括，離岸風電 Round 3-1（2,335 MW）統一展延至少一年，Round 3-2 最快 2028 年後才陸續併網，顯示官方容量時程偏樂觀；以及大林、通霄、協和等燃煤轉燃氣計畫接連流標或重議價，導致燃氣機組上線時程普遍落後。 系統現況方面 ，台電備轉容量率雖因新機組與儲能併網略升至 7–11%， 仍低於國際建議值 10–15% ； 歷次大停電事件暴露備援脆弱 ，任何大型機組故障都可能觸發限電。模型結果指出，在離岸風電延宕與燃氣機組遲滯的基準情境下，夜尖峰備用容量率將長期滯留 6–7% 警戒邊緣。

Propelled by the maxim that “computing power is national power,” round-the-clock electricity demand from artificial-intelligence (AI) applications and high-performance data centers is surging. Because these loads run 24/7, their night-time consumption is chronically out of phase with solar output, further squeezing Taiwan’s already-tight night peak reserve margin. We synthesize historical National Power Supply–Demand Reports and Long-Term Power Development Plans, incorporate maintenance and retirement schedules, tender delays, and high-tech industry expansion timelines, and construct a ten-year night-peak supply-demand model that tracks reserve-margin dynamics. On the demand side, advanced semiconductor fabs and AI cloud data centers are continuous high-load users. Their incremental power needs will account for 82 % of national load growth before 2030, driving night-peak demand from 37.25 GW in 2024 to 47.66 GW in 2033—a 31.8 % increase. On the supply side, the Round 3-1 offshore-wind (2,335 MW) has been uniformly deferred by at least one year, while Round 3-2 will not reach commercial operation until 2028 at the earliest, indicating that official capacity timelines are overly optimistic. Coal-to-gas conversions at Talin, Tunghsiao, and Xiehe have repeatedly failed tenders or required repricing, pushing gas-fired units behind schedule. Although Taipower’s operating reserve margin ratio has edged up to 7–11 % with new units and energy storage systems coming online, it remains below the 10–15 % internationally recommended range. Past blackouts reveal fragile backup capacity: a single large-unit failure can trigger load-shedding. Model results show that, under baseline assumptions of offshore-wind delays and gas-unit slippage, night-peak reserve margins will hover at a precarious 6–7 % throughout the decade.