What to know about DeepSeek
Add Axios as your preferred source to
see more of our stories on Google.
Photo Illustration by Justin Sullivan/Getty Images
Chinese AI startup DeepSeek has rattled Silicon Valley, potentially disrupting the nascent industry with an inexpensive and accessible alternative to the technology U.S. giants have poured hundreds of billions of dollars into already.
Why it matters: Nvidia's stock closed almost 17% lower Monday, wiping out more than $600 billion in market capitalization, as investors worldwide grapple with the prospect that the market-sustaining AI spending boom might have been overdone.
- On the flip side, DeepSeek may provide the general population with AI that's potentially a) much more accessible and b) more environmentally-friendly.
What is DeepSeek?
DeepSeek seems to do more with less.
- DeepSeek was founded in 2023 by Liang Wenfeng, the co-founder of the hedge fund High-Flyer, which develops open-source AI models, meaning that outside developers can inspect and improve the software.
- Liang bought a "stockpile of Nvidia A100 chips," according to the MIT Technology Review, which were used to develop DeepSeek. The U.S. has since sanctioned those chips from China, but withholding technology might just have helped DeepSeek create a leaner, meaner machine.
- The DeepSeek mobile app soared to the top of the U.S. iPhone download charts after its release this month.
- DeepSeek claims that its R1 release performs on par with OpenAI, and several third-party tests have found that DeepSeek actually outperforms OpenAI's latest model.
How is it affecting the stock market?
Just last week it seemed that the U.S. had a considerable lead in the AI race, with the main concern being a need for general contractors to build all of the required data centers, energy to power them and chips to run them. DeepSeek's rise may pull the rug out from under those investors.
- This could be catastrophic for venture capital firms that went all-in on foundational model companies. Axios' Dan Primack reported on Monday that investors were "concerned," and that some deals in process could stall.
- The company says it only cost $6 million to train the DeepSeek V3 model released in December. In comparison, Meta announced Friday that it plans $60 billion to $65 billion in capital investment this year as it scales up its own AI projects.
- In a statement Monday, Nvidia said that DeepSeek is an "excellent AI advancement" that is "fully export control compliant." The company emphasized the role of their own chips in DeepSeek, adding that "inference requires significant numbers of NVIDIA GPUs and high-performance networking."
Is DeepSeek better for the environment?
- AI requires a lot of energy to train and run its models and water to cool its data centers, and it generates both waste and noise pollution. A single ChatGPT question requires many times the energy of a simple Google search. The exact footprint of DeepSeek isn't clear yet, but it likely uses a lot less.
- "The fact that this technology is supposed to take less energy and is more cost-effective than U.S.-based models has U.S. technology investors very concerned," Jay Woods, chief global strategist at Freedom Capital Markets, told Bloomberg.
Go deeper:
