According to Andrew Ng (@AndrewYNg), while the influx of capital into AI infrastructure, including OpenAI's $1.4 trillion plan and Nvidia's $5 trillion market cap, has raised concerns about an AI investment bubble, the reality is nuanced across different industry segments. Ng emphasizes that the AI application layer is currently underinvested, with substantial untapped business potential, citing a hesitancy among venture capitalists to back AI applications due to perceived difficulty in selecting winners (source: deeplearning.ai/the-batch/issue-329). In contrast, AI infrastructure for inference requires further investment to meet surging demand for token generation, particularly as agentic coding tools like Claude Code, OpenAI Codex, and Gemini 3 drive new use cases. Although supply constraints exist, overbuilding could result in low returns but benefit application builders. The riskiest segment is model training infrastructure, where rapid algorithm and hardware improvements—as well as the rise of open-source models—may erode competitive moats and threaten returns on massive investments. Ng warns that overinvestment in any one segment, especially training infrastructure, could trigger negative market sentiment, impacting the broader AI sector despite strong long-term fundamentals. He concludes that while short-term valuation swings are driven by sentiment, the long-term outlook for AI remains robust, with significant opportunities for business innovation and infrastructure scaling (source: deeplearning.ai/the-batch/issue-329).