Beware of paying for growth, for growth is risky. That is one of the investing principles in chapter 1 and a theme throughout the book. Indeed, much of the book is oriented to dealing with the problem of buying risky growth.
An episode on Monday, January 27, 2025, starkly reminds us of the point.
Companies were engaged in a race to develop artificial intelligence (AI) engines requiring expensive computer chips and an enormous amount of electricity. In January 2025, firms were on a pace to spend a combined $1 trillion on the development, including investment in data centers and the required power plants. On that January day, DeepSeek, a company in Hangzhou, China, revealed that they had developed an artificial intelligence (AI) engine that performs nearly as well as those in the U.S. at a reported $5.6 million production cost and with cheaper, less sophisticated computer chips. That was a shock to Silicon Valley engineers but also to the stock prices of U.S. tech firms with their high market valuations based on growth expectations.
NVIDIA, the maker of high-powered chips deemed necessary for AI had reported $68 billion in profits over the prior 12 months and was trading at a P/E of 52 and a market capitalization of $3.5 trillion on growth prospects. It lost 17% of its value on that day although it recovered somewhat the next day. Power company stocks dropped 21% to 28%. The tech-heavy NASDAQ index dropped 3.6%. Overall, $1 trillion of value was lost on the stock market, $600 billion for NVIDIA. Ouch! There is presumably much value in AI and this might just be a bump in the road, but buying growth is risky.
For the fundamental investor, the episode was also a reminder of the forces of competition. The U.S. and China were in fierce competition, with the U.S. government subsidizing chip production in the U.S. and restricting the export of AI chips to China, the Chinese government likewise. But the forces of competition break through. On almost the same day, Alibaba released its Owen2.5-Max AI model. Tencent had multiple versions of its Hunyuan model that performed as well as Meta Platform’s Liama 3.1using a tenth of the power than Meta’s training model. Baidu’s Ernie Bot had 430 million users. Other upstarts beside DeepSeek, StepFun, Moonshot, MiniMax, and Zhipu were challenging (according to press reports). Buying growth is risky for it can be competed away.
