- The debut of an AI tool from a Chinese startup has sparked panic among US tech investors.
- The DeepSeek AI app is shaking the faith in sky-high valuations of Magnificent Seven stocks.
- It’s also likely to prompt questions about huge spending on AI by the largest tech companies.
A new artificial intelligence tool from a startup in China has sent shockwaves across the US tech sector, with the cheaper AI model trained on older chips forcing investors to confront two big risks to the market.
DeepSeek joined the AI race in recent days, sparking a big sell-off to start the week as investors faced concerns about risks that had largely been glossed over while the bull rally continued for the last two years — namely, risks associated with stretched tech-stock valuations and enormous spending by some of the biggest tech firms as they pursue their AI ambitions.
The S&P 500 and tech-heavy Nasdaq 100 fell sharply Monday morning, down 1.8% and 3.2% around noon ET. Semiconductor stocks faced the steepest losses, while tech giants including Microsoft and Alphabet also dropped.
The debut of the DeepSeek AI tool blindsided investors who have been willing to accept the tech sector’s historically stretched valuations as justified since the bull market took off in 2022.
The new AI app is also likely forcing tough questions about AI spending by the cloud “hyperscalers” that have pursued ambitious AI programs in the last few years.
Overnight, it’s become much harder to shrug off two big risks to the stock market’s dominant narrative.
Overvaluation
Chatter about a possible correction has risen, but has still mostly been confined to the fringes of Wall Street’s consensus views. The market has predicted another strong year for stocks, even after the S&P 500 hit more than 50 records in 2024 and posted a second year of 25% gains.
With help from the AI investing frenzy, the Magnificent Seven stocks — Nvidia, Microsoft, Alphabet, Meta Platforms, Amazon, Apple, and Tesla — have surged, and account for over a third of the value of the S&P 500.
Writing in a note earlier this month, Bank of America analysts said that of the 20 valuation metrics they track, 19 were sitting at extreme levels. The S&P 500’s trailing price-to-earnings ratio of 25.3 times, for example, is 70% above its 125-year average of about 15 times.
Valuation metrics like that have been easy to accept while the market’s biggest stocks faced little outside competition and could woo investors with promises of the transformative power of new technology.
But DeepSeek is now laying bare the danger that the US market has been overvalued. As its AI tool undercuts OpenAI on pricing, outperforms leading US peers, and is being trained on older and cheaper Nvidia GPUs, tech valuations are in doubt.
That’s evident in the plunge in the price of Nvidia stock on Monday. The chip titan, which reclaimed the title of the world’s most valuable company from Apple last week, dropped as much as 18% and was headed for a loss that would mark the worst market-cap wipeout in history. Broadcom fell more than 17% and Taiwan Semiconductor Semiconductor Manufacturing Company was down more than 14% at midday.
“What makes Monday’s tech selloff so jarring is that the valuations of many of these AI and tech companies offer no margin of error,” David Bahnsen, chief investment officer at The Bahnsen Group, said in written commentary.
“Excessive valuation always becomes a problem, eventually, but fundamental news becomes a heightened problem when it is combined with excessive valuation.”
Potential overspending
The tech meltdown could also lead to a painful reckoning among investors over high spending among the “hyperscalers” — companies like Microsoft, Amazon, and Meta Platforms — that have poured billions of dollars into AI to further build out cloud computing capabilities.
Morgan Stanley estimated at the end of last year that Amazon, Google, Meta, and Microsoft would collectively see $300 billion in AI capital expenditures through 2025, with even bigger figures to come in 2026. Before that, Bernstein projected that Big Tech capex on data centers and real estate would exceed $1 trillion over five years.
Microsoft alone said it would spend $80 billion this year on AI data centers.
“These high and rising capex numbers again speak to the importance of continued disclosure about new/incremental adoption, engagement, and revenue opportunities each of the four companies are seeing and investing in,” Morgan Stanley analysts wrote last November.
Investors had been getting somewhat anxious to start seeing a return on this spending, but have largely still been willing to tolerate it as long as earnings growth continued. Bullish analysts had argued that heavy spending was required to fuel a profitable AI boom over the next decade.
However, DeepSeek is challenging this view.
For one, the AI tool is trained on cheaper Nvidia H800 chips, as the US government has restricted the sale of some of the company’s most advanced GPUs to China.
The fact that a cutting-edge AI app has been developed with last-gen tech may damage the thesis that companies need to spend more and more on high-powered GPUs to keep up. This has sparked some discussions about future demand for Nvidia’s more expensive hardware.
“While there is a contention that DeepSeek’s efficient training methods could reduce the demand for high-end Nvidia GPUs, potentially affecting sales, it is also plausible that the more cost-effective approach will result in more demand for hardware for those looking to train proprietary models,” wrote Mark Klein, CEO of SuRo Capital.
What’s to come
While the market’s tech panic has revived dormant issues concerning valuations and spending, Bernstein analysts said on Monday that the reaction may be overblown. Analysts at Jefferies, meanwhile, suggested reduced exposure to semiconductor supply chains while the impact becomes more clear.
Investors are also bracing for tech earnings to begin this week, and DeepSeek has raised the stakes for mega-cap results. While strong Big Tech earnings have been a catalyst for fresh gains, DeepSeek could spoil the party this time.
“We do not expect big tech earnings over the next two weeks to help justify the sector’s elevated valuations. These companies may very well report a wonderful quarter, and they may guide higher for intended orders and revenue expectations. Current valuations do not factor in the risks of competition, which seems inevitable, and risks of diminished order books in the future, and all the other reasons that lofty valuations eventually revert to the mean,” he wrote.
Importantly, the broader stock market’s fate in 2025 might hinge on the performance of the Magnificent Seven. At the end of last year, Bank of America said that the elite group must keep winning to avoid spurring a wider market correction.