Listen and subscribe to Opening Bid on Apple Podcasts, Spotify, YouTube, or wherever you find your favorite podcasts.
DeepSeek has caused a permanent reset in how AI development will unfold in coming years.
At least that’s the take from Snowflake (SNOW) CEO Sridhar Ramaswamy, whose data storage company provides technology used by companies developing large language models like DeepSeek and ChatGPT.
“So what this [news] basically does is two things. One, it says there’s still a lot of innovation left and many companies can aspire to train these models. But, two, it also raises the really interesting question of do you need to spend billions of dollars in order to train cutting-edge, world-class models? I think the jury is still out there for things like that, but it’s basically upended our assumptions about where AI is going,” Ramaswamy told me on Yahoo Finance’s Opening Bid podcast (video above; listen in below).
This embedded content is not available in your region.
DeepSeek is a Chinese company that has burst onto the tech landscape seemingly out of nowhere.
Last week, it surprised more than just markets after unveiling RI, its AI model that gave a ChatGPT-esque performance at an arguably cheaper price tag. RI costs a reported $5.6 million to build a base model, compared to the hundreds of millions of dollars incurred at US-based companies such as OpenAI and Anthropic.
Watch: Why billionaire investor Ray Dalio is concerned about tech valuations
Fears mounted instantly that US companies are overspending on AI infrastructure, which includes Nvidia (NVDA) chips.
In response, stocks went haywire on Monday. Nvidia’s market value dropped by $588.8 billion, or 16.97%. It was the biggest ever single-day loss in a stock’s market cap in absolute terms, according to Deutsche Bank data. Monday’s decline for Nvidia was larger than the total market cap of Exxon or Mastercard.
The sell-off in Nvidia has persisted into Wednesday, with shares down 4.5%. The AI darling has seen its stock tank 11% so far this year.
“DeepSeek model suggests the cost of training GenAI models is set to decline substantially … which we see leading to faster gen-AI product innovation and availability,” Morgan Stanley tech analyst Brian Nowak said in a note.
Nowak stayed upbeat on Microsoft (MSFT) despite its investment exposure to OpenAI.
Interestingly, Nowak thinks Apple’s stock could be a winner as fears about DeepSeek’s impact weigh on the Magnificent Seven names.
“Apple’s AI ambitions are far more contained than the majority of the Mag 7 (or other frontier model developers), as Apple is currently focused on smaller, function-specific, on-device LLM’s, rather than large frontier models. As a result, Apple’s annual capex ($9.4B in FY24) is about 1/20th the scale of the US Tier 1 hyperscalers combined, and therefore if the market hyper-focuses on capex ROI, the threshold for Apple to deliver an attractive return is far lower (i.e. less risk) than the T1 hyperscalers,” explained Nowak.