AI 新創時代的估值邏輯轉變：從員工規模到快速迭代的驅動因素探討

Abstract

在生成式人工智慧與大型語言模型快速普及之下，AI 原生新創展現出前所未見的成長樣貌：員工規模相對精簡、營收形成速度極快、估值倍數遠高於傳統 SaaS 公司。Cursor 以約一百五十人之團隊，在十八個月內將年經常性收入自一億美元推升至二十億美元；DeepSeek 以數百人規模之研究團隊，用遠低於前沿實驗室之訓練成本，產出足以撼動全球資本市場之開源模型。「員工規模下降、營收上升、估值上升」之三角關係，打破了傳統企業成長敘事中以人力擴張為核心之線性假設，使長期以員工規模作為估值錨點之評估邏輯面臨結構性挑戰。本研究之核心問題即在於：當人力規模不再能解釋企業價值時，AI 新創之估值邏輯轉向何處？

為回答此一問題，本研究採個案探討與分析方法，選取 Cursor 與 DeepSeek 兩家 AI 原生新創為主案例，以 Salesforce 與 Tyler Technologies 兩家傳統 SaaS 上市公司為對照組，並延伸觀察 Midjourney 與 Lovable 兩家 AI 原生公司。分析上以價值主張圖（VPC）與商業模式九宮格（BMC）作為統一分析架構，輔以 Porter 五力與動態能力理論之文獻基礎，依產品定位、商業模式、規模與效率之脫鉤、營收品質與位移風險、投資人估值邏輯之次序，對各個案進行同一工具下之並列分析與跨案例比較，自各家差異中歸納研究發現；關鍵數據均經雙來源交叉查證，無法確認者標注為市場估計。

研究發現，AI 原生公司得以透過演算法、AI 工具與運算效率之槓桿，使營收成長與人員、資本之擴張脫鉤；企業價值之計價基礎，正從「員工人數規模」轉向「運算與能力密度之效率」。在此主軸下，本研究進一步提出三項觀察：其一，估值之計價單位已由人頭移向運算與 token 效率，以人為分母之傳統人均指標正逐漸失去訊號意義；其二，傳統企業之「組織債」構成 AI 時代之隱形估值負債——大公司裁員與 AI 新創刻意維持精簡編制，實為演算法槓桿普及之一體兩面；其三，人少換得的是搶先之時間窗口，公司要持續發展仍需算力與資本之基建支撐——動量是「借來的時間」，對開源生態系建構者而言並非估值之主要來源。

綜合上述發現，本研究對台灣創投業者提出後 ARR 時代之估值觀察維度，對新創創辦人提出能力密度優先於人力規模之組織設計原則，對傳統企業管理者提出組織債之盤點警訊，並對政策制定者指出以員工數為門檻之補助設計在 AI 時代可能產生之反淘汰效果，期能為台灣 AI 新創生態之發展與估值評估，提供兼具理論視角與實務可操作性之參考。

Amid the rapid diffusion of generative AI and large language models, AI-native startups exhibit a distinctive pattern: lean headcount, extremely fast revenue formation, and valuation multiples far beyond those of traditional SaaS companies. This poses a structural challenge to the conventional valuation logic anchored in employee scale. Adopting a case-study approach, this research examines two AI-native companies, Cursor and DeepSeek, against two traditional SaaS incumbents, Salesforce and Tyler Technologies, with two additional AI-native cases, Midjourney and Lovable, presented in the appendix. Using the Value Proposition Canvas and the Business Model Canvas as a unified analytical framework, supported by Porter's Five Forces and the dynamic capabilities literature, the study compares the cases across product positioning, business model structure, scale-efficiency decoupling, revenue quality, displacement risk, and valuation logic.

The findings show that AI-native companies leverage algorithms, AI tools, and computing efficiency to decouple revenue growth from the expansion of headcount and capital; the basis of corporate valuation is shifting from employee scale to the efficiency of computation and capability density. Three observations follow. First, the pricing unit of valuation has migrated from headcount to computation and token efficiency, rendering headcount-based heuristics increasingly obsolete. Second, organizational debt operates as a hidden valuation liability in the AI era: large-scale layoffs by incumbents and the deliberate leanness of AI-native startups are two sides of the same algorithmic-leverage phenomenon. Third, a lean team buys a head start rather than durability; momentum is borrowed time, and sustained growth still requires infrastructure in computing power and capital, as the DeepSeek case illustrates.

The study concludes with actionable implications for venture investors, startup founders, corporate managers, and policymakers in Taiwan, offering a practical reference for evaluating AI-native ventures in the post-headcount era.