News
Time to 'panic' or 'overblown'? Wall Street weighs how DeepSeek could shake up the AI trade
AI stocks fell across the board Monday as the investment community absorbed the announcement of a reportedly cheap Chinese AI model — and began to question the Western world’s gargantuan spending on artificial intelligence infrastructure.
Chinese AI startup DeepSeek released a new generative AI model called R1 last week positioned as a competitor to OpenAI. Analysts have cited DeepSeek’s models as much less costly than OpenAI's. DeepSeek said recently that it spent just $5.6 million to train another one of its latest models, V3, while OpenAI spent more than $100 million to train its GPT-4 model.
Nvidia ( NVDA ) fell nearly 17% Monday, shaving off hundreds of billions from the AI chip giant’s market cap and putting short sellers up more than $5 billion, per data from S3 Partners.
Meanwhile, rival Advanced Micro Devices ( AMD ) dropped around 6%, and Broadcom ( AVGO ) plunged more than 17%. The broader markets also suffered: The S&P 500 ( ^GSPC ) was down 1.5% Monday morning, while the tech-heavy Nasdaq ( ^IXIC ) dropped 3%.
"The performance of the DeepSeek models developed in China may have a very profound impact on the US AI companies," DA Davidson analyst Gil Luria told Yahoo Finance in an email. "The models from DeepSeek are performing as well as the most advanced models developed in the US at a small fraction of the price."
“That means US hyperscalers such as Microsoft ( MSFT ), Amazon ( AMZN ), Google ( GOOG ) and other[s] may be severely overinvesting in their data center buildout,” he added. “If companies can run AI models with very little compute, they do not need data centers with hundreds of thousands of NVDA GPUs.”
Raymond James analyst Srini Pajjuri echoed those sentiments in a late Sunday note.
“DeepSeek clearly doesn’t have access to as much compute as U.S. hyperscalers and somehow managed to develop a model that appears highly competitive,” Pajjuri wrote.
US President Donald Trump last week announced a venture called Stargate backed by OpenAI, SoftBank, and Oracle ( ORCL ), which would immediately invest $100 billion in US AI infrastructure (i.e., data centers) and spend an additional $400 billion over the next four years. Shortly after that announcement, Meta ( META ) said it would significantly ramp up its capital expenditures to as much as $65 billion in 2025.
“We think investors should take the innovation from China seriously, as it puts into question whether the current pace of capex spend/technology upgrades is necessary," CFRA analyst Angelo Zino said Monday.
“We think heightened risks may cause multiples for leading edge chipmakers to compress, with those most at risk being NVDA, MRVL, AVGO, AMD, and MU,” he added.
Damindu Jayaweera, an analyst with UK investing firm Peel Hunt, told Yahoo Finance that the DeepSeek news means Nvidia’s AI chip prices may come under pressure from “people getting more performance out of its chips." But Jayaweera added that shift could also create more demand for Nvidia chips “beyond the current Big Tech stronghold on supply.”
“I think while there will likely be a sentiment impact on Nvidia near-term where people think Big Tech is going to pause AI capex, medium to longer-term, we think there is potential for greater volumes, albeit at lower margins,” Jayaweera told Yahoo Finance in an email. Peel Hunt does not cover Nvidia and Jayaweera does not hold an official rating on the stock.
Some Wall Street analysts were more skeptical of the buzz around DeepSeek, however.
Bernstein’s Stacy Rasgon said DeepSeek likely spent more money training its V3 model than it reported. He said the spending cited by the firm is “misunderstood” and does not include costs associated with prior research.
“[It] seems categorically false that ‘China duplicated OpenAI for $5M,’” Rasgon wrote in a note to investors Monday, adding, “the resulting Twitterverse panic over the weekend seems overblown.”
"We have no way to know whether DeepSeek's claims about the type of GPUs, the number of GPUs, or the time to train its model are true," William Stein of Truist Securities added in a note to investors Monday.
Citi analyst Atif Malik said in his own note that he doesn’t see advanced AI chipmakers’ business coming under pressure: “We see the recent AI capex announcements like Stargate as a nod to the need for advanced chips.”
The DeepSeek buzz comes as Big Tech earnings season gets underway, with Microsoft, Meta, and Tesla ( TSLA ) all set to report earnings Wednesday.