Andrew Ng argues that the most significant constraints on AI progress in the US and the West are no longer just algorithms or data, but the physical infrastructure of electricity and semiconductors. While demand for AI inference is insatiable, the ability to build and power new data centers is hampered by permitting and energy shortages, a problem China is aggressively solving.
The discussion frames AI development as a central theater of geopolitical competition. Ng posits that US chip export controls have inadvertently spurred China to accelerate its domestic semiconductor industry. Concurrently, China's proliferation of high-quality, open-weight models is a potent form of soft power, while Europe is falling behind due to a focus on regulation over investment.
A significant portion of venture capital invested in AI application startups is immediately passed through to foundation model providers (like OpenAI, Anthropic) and ultimately to hardware companies (NVIDIA). This creates a subsidized market, similar to the early days of food delivery, with poor margins for application companies. The long-term viability of this model is questionable without a significant drop in token costs.
AI is already transforming high-skill professions, particularly software engineering. The most productive developers are now experienced engineers who are expert users of AI coding assistants. Ng disagrees with predictions that useful AI agents are a decade away, citing current, valuable agentic workflows in areas like tariff compliance, and emphasizes that learning to code remains a critical skill.
Keep pulling the thread on Andrew Ng.