By Yannick Laurent, Research Analyst, YLF
The recent unveiling of DeepSeek rattled markets and raised questions about the current and future dynamics of the AI industry.
The situation as I understand it
The paradigm until now has been that western tech firms had cornered the market for cutting edge AI. This, in turn, justified their large investments into hardware. With DeepSeek replicating the best-in-class AI models for under $6 million, the economic logic underpinning big tech’s large capital expenditures becomes uncertain. With fewer expected investments in computing power, it is no surprise that Nvidia, TSMC, and others, have taken a hit to their stock prices.
The tech innovation lens
A distinct pattern appears when examining past technological revolutions. At the beginning, new inventions differentiate themselves based on features. The companies that are at the cutting edge push the envelope forward by expanding their products’ capabilities and growing their use cases. At some point, the new technology matures, having built out every possible feature and fully percolating throughout the economy. At this point, firms pivot, focusing on being the lowest cost producer to stay ahead. To learn more, I encourage you to take a look at these articles.
AI market evolution
So, what should we make of the initial decimation of AI stocks’ prices? On the surface, the market’s reaction seems to imply that investors believe the industry has reached the inflection point that was mentioned earlier. However, the narrative until now has been that the AI revolution is in its infancy, with many more features and integrations on the horizon. There is no reason that DeepSeek’s emergence as a low-cost competitor at the current functionality level would necessarily change this outlook. In fact, the availability of cheaper AI should expand, not contract, the industry overall.
To understand what’s going on, we must examine the industry in more detail. Some customers – those who require the most advanced AI capabilities and/or get the most added value from using AI – are willing to pay up for the technology. At the other end of the spectrum are users for whom the benefits have not been worth the cost, until now. Cheaper AI expands the market to encompass these new users. Whether or not DeepSeek will eat the market share of incumbents depends on two questions:
How fast can DeepSeek catch up to western firms in terms of new functionality (so far, their innovations have mostly been in more efficient chip usage)?
And how fast can western firms catch up to DeepSeek in terms of cost?
I can see several scenarios playing out from here:
Convergence: DeepSeek can consistently match the functionality of western firms while western firms figure out how to reduce costs to match DeepSeek. In this case, American and Chinese companies compete fiercely on both cost and features, with both experiencing slower but consistent growth. For chip makers, it will be a question of whether the increased demand for lower-end chips will offset the decreased sales from their most powerful offerings. Firms that specialize in top-of-the-line chips will be hurt the most.
Made in China: The second scenario is one where DeepSeek keeps up with western firms in terms of functionality and maintains its cost advantage. This is the worst case for the American AI industry, as all but the most price-inelastic customers will choose to wait a couple of months for the cheaper option. The potential winners (other than DeepSeek) will be firms that focus on implementing third party AI models into their customers’ workflows. Chip producers will face the same calculus as in the Convergence scenario, where changes in their price-volume mixes will yield uncertain results.
Split market: While scenarios 1 and 2 are the most likely long-term outcomes, the AI sector could experience a third dynamic before settling there. In this transitional scenario, western firms take the lead in creating cutting-edge features while competing with DeepSeek at the less advanced and lower cost end of the market. Given that DeepSeek is only two to three months behind OpenAI, this scenario would require American firms to dramatically increase both the rate at which they develop new functionality, and the lead times required for rivals to catch up to them. Eventually, AI will become feature-complete, at which point firms will have to focus on costs, transitioning the industry to scenarios 1 or 2. This would be similar to the dynamic that we see in the semiconductor sector. Speaking of which, scenario 3 is the best for chip makers, as they will see strong demand for top-end processors while simultaneously experiencing growth in their low-to-medium end offerings, as long as the race for better AI features persists.
The stock market’s initial broad selloff appeared to indicate that investors expected (or at least substantially feared) that the industry is headed straight for scenario 2. As of the time of writing, shares of firms such as C3.ai, Palantir, and Microsoft have mostly recovered their losses while those of Nvidia, TSMC, and ASML have not. This would indicate that the market is now expecting some version of scenario 1. While the AI industry is still on track for growth, it is too early to know what that growth will look like with any certainty. The only way to navigate these changes is by understanding the nuances of the AI landscape.
About the Contributor
Yannick Laurent conducts original investment research, economic analysis, and discussions of topics in asset management, which he publishes on his website, YannickLaurentFinance.com. He has a background in wealth management and has mentored students who are interested in pursuing a career in financial services. Yannick is a member of the CFA Society of San Francisco and received his bachelor’s degree in economics from Macalester College.
Learn more about CAIA Association and how to become part of a professional network that is shaping the future of investing, by visiting https://caia.org/