A new player stealthily entered the AI race—and it seems to be sleeker, smarter, and most importantly, way cheaper than anything Silicon Valley has to offer.
The market’s AI-fueled rally began to crack this morning after Chinese startup DeepSeek unveiled some head-turning results from its new R1 reasoning model—its free, open source version of ChatGPT—that makes American models look like your dad’s clunky old Ford Taurus compared to China’s fuel-efficient new sports car.
DeepSeek’s v3 chatbot, which it first released in late December, can perform just as well as the US competition, and in some tests actually beat American generative AI models like ChatGPT or Meta’s Llama.
The firm said its model was developed in just 2 months. And that’s not even the wildest part: According to DeepSeek’s own research paper, it cost just $5.6 million to develop. Compare that to the $78 million it cost to train Open AI’s ChatGPT-4, or the $191 million it took to train Google’s Gemini Ultra—not to mention the $1 trillion VCs and other investors are on track to pour into AI startups, according to Goldman Sachs.
What makes this all even more astonishing is that the US has been blocking China from getting some of its most advanced AI tech, including Nvidia chips. That means that the startup is reportedly doing all of this with the less-advanced H800 Nvidia chips it collected before the 2022 sanctions.
And to add insult to injury, DeepSeek took the number one spot on Apple’s App store today, smacking ChatGPT down to #2. Ouch.
The AI stock fallout
“DeepSeek-R1 is AI’s Sputnik moment,” VC Marc Andreessen wrote on X last night.
Following the revelation that the tech companies leading the AI arms race may not actually need to spend all that money to build a chatbot, Nvidia plunged 16.86%, shedding $589 billion in market cap—the largest single-day loss in stock market history. Microsoft fell 2.14%, Palantir sank 4.43%, Super Micro Computer dropped 12.52%, and Broadcom plummeted 17.40%.
Making sense of market moves
Stay up to date on the latest market news with daily analysis of the investing landscape, served up Brew-style.
Big tech names weren’t the only ones feeling deep pain. Energy and infrastructure stocks that once soared in hopes of an AI spending spree tumbled: Constellation Energy fell 20.85%, Oklo dropped 25.61%, and Vistra declined 28.28%. And Vertiv Holdings, which makes cooling systems for data centers, plunged 29.83%.
“DeepSeek raises concerns that future LLMs will be developed with fewer AI GPUs from Nvidia, lower capital expenditure from hyperscalers, and perhaps less of a need for a massive energy buildout in the US,” explained Morningstar equity strategist Brian Colello.
Should you deep-six your AI portfolio?
First off, don’t panic. There’s still a lot that’s unclear about how this will shake out when the smoke clears.
Some analysts are skeptical about DeepSeek’s reported spending. And there’s still a lot unknown about the elusive startup itself.
In fact, many analysts are pointing out that this could be a buying opportunity to get in on the AI trade at a more reasonable price. After all, fierce international competition could be a boon for AI hyperscalers such as Nvidia, explained TD Cowen analyst Joshua Buchalter in a note today.
“From Fed moments, rumors of Nvidia Blackwell delays, Tokyo Black Monday...and today is no different,” explained Wedbush analyst Dan Ives in a note. “These are just the opportunities to own the Nvidia, Microsoft, Alphabet, Palantir, Salesforce, Amazon and broader tech ecosystem that is under heavy pressure today.”—LB