Market Update: Recent Selloff in U.S. Artificial Intelligence Stocks
- Many artificial intelligence (AI)-related stocks sold off on Monday on news that DeepSeek, a Chinese AI model, has matched existing AI model performance at a fraction of the cost.
- This drove fears that the competitive advantage of deep-pocketed U.S. mega-cap technology companies and demand for advanced chips and related infrastructure, could be at risk.
- Overall, a lower cost to incorporating AI can broaden adoption and use cases, ultimately benefiting the broader economy.
- We believe market fundamentals remain sound, and we expect a broadening in earnings growth outside of the Magnificent 7, as market participation becomes less concentrated relative to recent years.
What happened:
China-based startup DeepSeek launched its latest AI model, DeepSeek-R1, an open-source reasoning model that claims to yield similar performance to more established models at a fraction of the cost. While the figures have yet to be verified, DeepSeek claims to have spent only about $6 million to develop R1. That said, preliminary assessments suggest DeepSeek’s claims might be misleading and fail to account for the full training costs.
DeepSeek's R1 announcement caused market volatility, particularly surrounding AI spending by U.S. technology hyperscalers. In the near term, we expect hyperscaler capex to be scrutinized, with a focus on optimizing rather than increasing spending. We plan to provide more detail after their earnings reports are released later this week. If DeepSeek's cost reduction of 95% or more per unit of computing is embraced by competitors, it could drive adoption and greater total demand for technology infrastructure. While a decline in capex might initially cause negative market reactions, reallocating budgets and growing free cash flows with cheaper AI could benefit the global economy in the long run.
DeepSeek’s success also exposes the potential limitations of U.S. export controls aimed at limiting advancement in Chinese AI capabilities. Uncertainty remains regarding which Nvidia chips were utilized in training DeepSeek. Additionally, the DeepSeek interface has many similarities to ChatGPT-4. If this is in fact the case, we expect a policy response from President Trump aimed toward asserting U.S. AI leadership.
Market views:
Ahead of Monday, equity positioning and valuations were elevated, particularly in the technology sector, leaving room for a pullback. Monday’s price action pointed to a “sell first, think later” event, in which some of the indiscriminate selling of semiconductor and AI-related stocks was likely overblown. Market fundamentals remain sound, in our view, and despite a 1.5% decrease in the S&P 500, 70% of the S&P 500 stocks traded higher. Notably, six of the 11 S&P 500 sectors (Exhibit 1) and the equal-weighted S&P 500 were positive on the day. As of Tuesday morning, many of the Magnificent 7 stocks began to stabilize, with Nvidia, Apple, Amazon, and Meta rebounding as investors started to digest the implications of cheaper AI models and increased AI adoption.
Exhibit 1: S&P 500 Intraday Sector Returns, Monday, January 27
Key takeaway: The market sell-off was mostly confined to the technology sector.
Exhibit 1: S&P 500 Intraday Sector Returns, Monday, January 27
Key Takeaway: The market sell-off was mostly confined to the technology sector.
As of January 27, 2025. Past performance is no guarantee of future results.
Source: Bloomberg
Source: Bloomberg
While we did not predict the magnitude of yesterday’s move, we have been of the view that equity strength and earnings will broaden this year. Earnings growth for the Magnificent 7 is set to come down, while strength outside of the group is set to broaden. We are also seeing potential for robust earnings growth in areas of the market where earnings growth has contracted, for example small cap stocks. We discuss this further in our Quarterly Investment Perspective, 2025: A Strong Foundation for a Year of Transition.
Bessemer positioning:
Although Bessemer portfolios remain slightly overweight in technology, this overweight was reduced from 200 basis points at the start of 2024 to 90 basis points currently. Given our expectation for earnings to broaden, Bessemer portfolios are currently overweight five sectors, up from four at the beginning of 2024, with less concentration in specific sectors.
We note that innovation’s role in driving down training and inference costs is not a new phenomenon and thus far has not damaged the business models for Nvidia and Broadcom. In the near term, more resources may be dedicated to software to catch up to DeepSeek, but innovation should ultimately drive demand. We believe Bessemer portfolios remain well positioned to capture future innovation.
Furthermore, exposure to risks and opportunities is not uniform across the hyperscalers. Meta is positioned to further benefit from open-source innovation. Likewise, software vendors who have distribution and data advantages, such as Microsoft, can capitalize on cheaper inferencing and can enhance their existing products while reducing costs to drive adoption.
Volatility is a feature of equity markets, and we will likely experience additional episodes throughout the year, given elevated equity valuations and an evolving policy landscape. Bessemer’s portfolio managers will seek to capitalize on such periods, even if at the margin. At times like this, it is important to stay the course through market turbulence to achieve long-term financial goals.
