AI雙軸競爭力：建構未來二十年生技創投的核心能力

Abstract

人工智慧（AI）技術的快速發展，正同時從兩個方向重塑生技創投產業的競爭格局：一方面，AI工具可望強化創投合夥人（GP）在案件篩選、技術盡職調查、估值與投資後管理等決策環節的效能；另一方面，越來越多生技新創公司將AI能力標誌為其核心競爭優勢來源。然而，現有文獻對這兩個面向多分別討論，鮮少從整合視角探討兩者之間的互動關係，特別是在台灣這類資源相對受限的生技創投生態中，這種互動是呈現正向協同，還是雙重落差的惡性循環，仍是一個尚待回答的研究問題。

本研究提出「AI雙軸競爭力」分析框架，將創投GP的AI評估決策能力定義為「軸一」，將被投公司的AI競爭護城河定義為「軸二」，並提出四個研究命題：RP1（兩軸正向協同）、RP2（台灣兩軸落差正在惡性循環）、RP3（雙軸能力為GP帶來競爭優勢，但台灣目前幾乎不存在具備此能力者）、RP4（健保資料庫商業化將同時強化兩軸）。本研究採質量混合研究設計，對14位受訪者（涵蓋台灣、美國、以色列的創投GP、AI生技創辦人與跨界專家）進行15份半結構式深度訪談，並輔以22份量化問卷，以三角驗證方式進行系統性分析。

研究發現顯示：（一）台灣創投的AI工具使用呈「通用工具普及、專業工具尚未普及」的M型結構，AI在「技術文獻掃描」（平均4.20/5）等資料整理型任務效益顯著，但在「退出時機判斷」（平均2.53/5）等前瞻判斷型任務效益最低；（二）軸一能力缺口由能力性、制度性（如保密協議限制AI分析）與資料結構性三類障礙疊加構成；（三）台灣AI生技護城河呈現演算法型、平台整合型、數據壁壘型三種類型，其中Market Access準備度與敘事轉譯能力是現有評估框架中最被低估、卻最具關鍵性的兩個維度；（四）量化問卷對RP3-b（台灣幾乎不存在同時具備雙軸能力的GP）獲得77%的最高共識，確認雙惡性循環現況仍在運作。

基於研究發現，本研究提出GP評估被投公司AI競爭力的五維框架（護城河類型識別、Market Access準備度、AI不可複製性評估、敘事轉譯能力、CEO AI第一手認知），並針對創投GP、AI生技新創及政策制定者三個層次提出具體建議。本研究的學術貢獻在於建構了一個整合軸一與軸二的分析框架，提出AI護城河的三類型分類，並識別出Market Access能力和敘事護城河兩個現有文獻鮮少觸及的競爭力維度，為台灣生技創投在AI時代的能力建構提供了系統性的診斷與操作工具。

The rapid advancement of artificial intelligence (AI) is reshaping the competitive landscape of biotech venture capital from two directions simultaneously. On one hand, AI tools hold the potential to strengthen General Partners' (GPs') decision-making effectiveness across deal sourcing, technical due diligence, valuation, and portfolio management. On the other hand, an increasing number of biotech startups are positioning AI capability as their core competitive advantage. However, existing literature largely treats these two dimensions separately, with limited research examining their interaction—particularly whether, in resource-constrained ecosystems such as Taiwan's biotech VC sector, this interaction produces positive synergy or a self-reinforcing dual deficit.

This study proposes an "AI Dual-Axis Competitiveness" framework, defining GPs' AI-enabled evaluation and decision-making capability as "Axis 1" and portfolio companies' AI-driven competitive moats as "Axis 2." Four research propositions are tested: RP1 (positive synergy between the two axes), RP2 (a self-reinforcing dual deficit currently exists in Taiwan), RP3 (GPs with dual-axis capability gain significant competitive advantage, yet such GPs are nearly absent in Taiwan), and RP4 (commercialization of Taiwan's National Health Insurance Research Database would simultaneously strengthen both axes). Employing a mixed-methods design, this study conducted 15 semi-structured in-depth interviews with 14 informants—spanning VC GPs, AI-native biotech founders, and cross-sector experts across Taiwan, the United States, and Israel—supplemented by 22 quantitative survey responses, analyzed through methodological triangulation.

Key findings include: (1) AI tool adoption among Taiwanese biotech VCs exhibits an "M-shaped" pattern—general-purpose tools (e.g., large language models) are nearly ubiquitous, while specialized tools remain underutilized; AI delivers significant value in data-aggregation tasks such as technical literature scanning (mean 4.20/5) but minimal value in forward-looking judgment tasks such as exit-timing decisions (mean 2.53/5). (2) The Axis 1 capability gap stems from three distinct types of barriers—capability-based, institutional (e.g., NDA restrictions on AI-assisted analysis), and data-structural. (3) AI-driven moats in Taiwanese biotech manifest in three types—algorithmic, platform-integration, and data-barrier—with Market Access readiness and narrative-translation capability identified as the two most underweighted yet critical evaluation dimensions in existing frameworks. (4) Survey results show RP3-b ("Taiwan currently has almost no GPs possessing dual-axis capability") received the highest consensus (77% agreement), confirming that the self-reinforcing dual deficit remains operative.

Based on these findings, this study proposes a five-dimensional framework for GPs to evaluate portfolio companies' AI competitiveness—encompassing moat-type identification, Market Access readiness, AI replicability risk assessment, narrative-translation capability, and CEO's first-hand AI literacy—along with actionable recommendations for VC GPs, AI-native biotech founders, and policymakers. The academic contribution of this study lies in constructing an integrated framework bridging Axis 1 and Axis 2, proposing a three-type taxonomy of AI moats, and identifying Market Access capability and narrative moats as two competitiveness dimensions rarely addressed in existing literature—offering a systematic diagnostic and operational toolkit for capability-building in Taiwan's biotech VC sector in the AI era.