Humans thrive on stories. Psychological experiments document their effectiveness in communicating information and changing behavior. Neurologists measure the actual changes in brain chemistry that they trigger. It is no surprise then that stories drive equity markets. A compelling narrative about a technological revolution or, alternatively, a dying industry triggers investors much more than a nuanced argument based on finance and economics. Narratives create risk when events challenge their premises, causing disillusioned investors to lose faith in previously held beliefs.
The massive outperformance of NVIDIA last year was driven by a compelling narrative that relied on a couple of widely accepted premises. First, that generative AI is the future of technology, and secondly, that this future depended on massive investments in NVIDIA chips, data centers and electricity generation. This second premise explains the massive gains in NVIDIA and a number peripheral stocks benefiting from this expected investment.
Chinese AI startup DeepSeek may have changed the narrative around AI this weekend with the release of their Large Language Model (LLM) that performs about as well as the generative AI products launched by Microsoft, Meta or Alphabet. No Western customer wants a Chinese-based AI platform that conforms to all the speech restrictions imposed by the Chinese Communist Party (for example, DeepSeek does not know who Xi Jinping is or what happened in Tiananmen Square in 1989), so why is this news causing a selloff of tech stocks, most notably NVIDIA?
DeepSeek developed this platform without using the advanced NVIDIA chips that US tech companies have invested tens of billions in. DeepSeek instead created better algorithms that could work with less expensive chips and require less computing power. They then made these algorithms freely available for others to use. By demonstrating they could create a world-class large language model without massive investments in NVIDIA chips and data centers, DeepSeek threatens to disrupt the entire tech industry and destroy the prevailing narrative about the massive capital investments required for corporate America to widely adopt AI.
While believers in the long-term promise of generative AI, we have been wary of the massive amounts of capital currently being invested in this space. History provides numerous examples, from railroads in the 1800s to the telecom infrastructure in the early 2000s, where the economic benefits of investments in new technologies accrued almost entirely to the consumers rather than the developers. The last glaring example was the fracking revolution in the early 2010s that allowed the US to re-emerge as a leading exporter of oil and gas and subsequently led to widespread bankruptcies among producers after they created an oversupply that took years to rebalance.
It will take time for others to vet and implement the advances in AI development that DeepSeek shared this weekend. However, regardless of the degree that DeepSeek’s innovations reduce the need for investments in data centers and high-end chips, the narrative has been disrupted. The often breathless predictions of chip demand and data center construction that have circulated through the investment world now warrant some healthy skepticism.
If today’s market movement reflects a new reality of a less capital-intensive AI future, it will continue a trend of the ‘next big thing’ turning into a loser investment. From the supposed death of the oil & gas industry in 2020, through the biotech bubble in 2021 and the inflation / natural resource scarcity narrative in 2022, popular market narratives have not aged well. Diversification and focus on company fundamentals remain the best approach to long-term investing.