The discussion posits that a technology can be world-changing without being a good investment. Historical examples like the airline and biotech industries are used to argue that high capital costs, long payoff delays, and intense competition can prevent revolutionary technologies from generating profits for investors.
Unlike traditional software with near-zero marginal costs, AI scaling requires massive physical infrastructure, including data centers, power, and skilled trades like electricians and plumbers. This creates significant real-world bottlenecks related to construction, supply chains, and labor shortages.
The construction of data centers is facing significant public and political opposition, even in areas that benefit from the tax revenue, like Loudoun County, VA. The discussion also touches on environmental justice concerns, with some projects being placed in less politically powerful communities.
The conversation expresses strong skepticism about the current valuations in the AI sector, citing examples like NVIDIA's market cap exceeding the value of Australia's arable land. The speaker suggests that while the technology is real, the market is exhibiting classic bubble characteristics driven by hype and speculative investment.
The technological lead of proprietary AI models is eroding quickly, with open-weights labs from China now only 4 to 8 months behind the leading edge. This rapid catch-up by open-source alternatives threatens the long-term pricing power and defensibility of companies that have invested billions in their models.
Keep pulling the thread on Scott McCandless.