科技股反彈，AI 擔憂緩解

The AI disruption concerns center on several technical and business model factors:

• Large Language Model Capabilities: Anthropic's release of Claude-based tools designed to automate legal work, finance, sales, and marketing tasks demonstrated practical applications that could displace existing software solutions[3][4]

• Capital Intensity of GenAI Infrastructure: Building out Large Language Model and Generative AI infrastructure requires massive capital expenditure for compute resources, with 2026 estimates at $660 billion across the sector, up 24% year-over-year[1]

• Hyperscaler Spending Levels: Individual company capex commitments include Meta ($55 billion), Alphabet ($180 billion, doubled from prior guidance), and Amazon ($200 billion, 50% increase)[1]

• Free Cash Flow Deterioration: The 12-month forward free cash flow for hyperscalers has fallen below 2022 cycle lows, with Amazon's capex intensity expected to push the company into negative free cash flow territory in 2026[1]

• AI Self-Improvement Risk: Economist Ed Yardeni noted AI's unique characteristic of being able to write software code, including AI code itself, creating potential for rapid obsolescence cycles where new AI-generated code displaces older implementations[3]

The February 2026 volatility signals a fundamental repricing of AI's economic impact across multiple dimensions. Rather than a broad-based productivity boost benefiting most companies, markets are now pricing in significant disruption risk to software, professional services, and knowledge-work sectors[3][4]. The sustainability of hyperscaler capex spending remains uncertain—if free cash flow turns negative across the sector, it could trigger a 'red flag' that fundamentally challenges the investment thesis supporting current valuations[1]. The divergence between AI disruptors (infrastructure providers, AI startups) and disrupted companies (software, legal services, consulting) is likely to create sustained return dispersion within technology and across the broader economy. Additionally, the pace of AI capability advancement and potential for rapid code obsolescence may compress technology adoption cycles, creating winner-take-most dynamics rather than broad-based benefits[3].