This article originally appeared on Business Insider.
Venture investors have been pouring billions of dollars into large AI model providers such as OpenAI and Anthropic to create an impenetrable moat around the building blocks of generative artificial intelligence. Those assumptions are being tested by DeepSeek, a Chinese startup that says it developed powerful AI models more cheaply than US rivals.
It’s a scary prospect for a VC industry that suffered busts like Web3 not long ago. After all, AI has been largely responsible for propping up valuations for the entire tech industry.
Related: What Is DeepSeek? China’s ‘Cheap’ to Make AI Chatbot Climbs to the Top of Apple, Google U.S. App Stores
Deepseek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen — and as open source, a profound gift to the world. ??— Marc Andreessen ?? (@pmarca) January 24, 2025
Much about DeepSeek remains unknown, but VCs who have bet the farm on expensive LLM startups are taking notice.
“DeepSeek is threatening because they open-sourced a model that’s near state of the art that is priced far below anything the US labs had planned,” said Deedy Das, a principal at Menlo Ventures, which has backed the OpenAI rival Anthropic. “That’s a competitive threat to the business model even in the face of the declining costs of LLMs.”
DeepSeek was started as a side project at High-Flyer, a Chinese quantitative hedge fund, which remains its sole investor. Das estimates that if it were a US startup, DeepSeek would already be valued at as much as $10 billion.
Shares of Nvidia tumbled by as much as 17% on Monday amid a broader sell-off in US technology stocks over fears that DeepSeek could signal that companies don’t have to spend nearly as much on GPUs to stay ahead in the AI race. Bernstein Research has said that comparable models from OpenAI cost 20 to 40 times what some DeepSeek models cost to use.
Iris Sun, an investor at 500 Global, said the app’s rapid progress demonstrated the threat of inexpensive open-source development.
“DeepSeek not only levels the playing field in terms of performance and cost but also could bring about structural changes to the entire AI value chain,” Sun said. “It raises questions about the viability of the capital expenditures behind top companies hoarding computing power.”
A silver lining?
Ever the optimists, some VCs see a silver lining in DeepSeek.
Nah this is an exponential event for vertical SaaS
More startups than ever are going from zero to $10M per year in recurring revenue with less than 10 people
The next years will be IPO class companies getting to $100M to $1B/yr.
A thousand flowers will bloom https://t.co/V0i5ytvcPa— Garry Tan (@garrytan) January 27, 2025
Hadley Harris, a cofounder and general partner at Eniac Ventures, said that for VC firms priced out of hot funding rounds for the likes of OpenAI and Anthropic, DeepSeek could be a positive sign.
“Many of the largest VCs have poured money into the big, closed-source model providers, typically as a hedge in case those models become the primary source of value in AI,” Harris wrote on X.
“The prospect of this layer being democratized by cheap, open-source models is bad news for them. Meanwhile, smaller VCs (like @EniacVC), who are too small for those types of investments, have focused on application-layer companies that stand to benefit greatly from not having to pay a hefty tax to OpenAI and Anthropic.”
Jon Turow, a partner at Madrona Ventures, said DeepSeek’s apparent progress should embolden rather than frighten the leaders of top LLM companies.
“I wouldn’t frame DeepSeek as a direct threat to OpenAI, Anthropic, or xAI’s market position,” Turow said. “It reinforces that their competitive advantage lies in constantly pushing the frontier. If they ever slow down or rest on their laurels, they’re in trouble — that was always true, and DeepSeek makes it more apparent than ever.”
