Topline
Here’s everything to know about Chinese AI company called DeepSeek, which topped the app charts and rattled global tech stocks Monday after it notched high performance ratings on par with its top U.S. rivals.
Key Facts
The DeepSeek startup is less than two years old—it was founded in 2023 by 40-year-old Chinese entrepreneur Liang Wenfeng—and released its open-source models for download in the United States in early January, where it has since surged to the top of the iPhone download charts, surpassing the app for OpenAI’s ChatGPT.
DeepSeek’s latest product, an advanced reasoning model called R1, has been compared favorably to the best products of OpenAI and Meta while appearing to be more efficient, with lower costs to train and develop models and having possibly been made without relying on the most powerful AI accelerators that are harder to buy in China because of U.S. export controls.
The company’s R1 and V3 models are both ranked in the top 10 on Chatbot Arena, a performance platform hosted by University of California, Berkeley, and the company says it is scoring nearly as well or outpacing rival models in mathematical tasks, general knowledge and question-and-answer performance benchmarks.
DeepSeek said training one of its latest models cost $5.6 million, which would be much less than the $100 million to $1 billion one AI chief executive estimated it costs to build a model last year—though Bernstein analyst Stacy Rasgon later called DeepSeek’s figures highly misleading.
Loading...
Despite the questions remaining about the true cost and process to build DeepSeek’s products, they still sent the stock market into a panic: Microsoft (down 3.7% as of 11:30 a.m. ET), Tesla (1.3%), Nvidia (15%) and Broadcom (16%) all slipped Monday following a selloff spurred by DeepSeek’s success, and the tech-heavy Nasdaq was down 3.5% on the way to its third-worst day of the last two years.
Crucial Quote
“Deepseek R1 is AI’s Sputnik moment,” billionaire investor Marc Andressen said Sunday.
Big Number
1.6 million. That’s how many times the DeepSeek mobile app had been downloaded as of Saturday, Bloomberg reported, the No. 1 app in iPhone stores in Australia, Canada, China, Singapore, the US and the U.K.
How Do I Use Deepseek?
DeepSeek is free to use on web, app and API but does require users to create an account.
How Does Deepseek Compare To Openai And Chatgpt?
DeepSeek-R1 is most similar to OpenAI’s o1 model, which costs users $200 per month. Both are large language models with advanced reasoning capabilities, different from shortform question-and-answer chatbots like OpenAI’s ChatGTP. R1 and o1 specialize in breaking down requests into a chain of logical “thoughts” and examining each one individually. R1 has achieved performance on par with o1 in several benchmarks and reportedly exceeded its performance in the MATH-500 test. Chatbot Arena currently ranks R1 as tied for the third-best AI model in existence, with o1 coming in fourth.
Is Deepseek-R1 Open Source?
Yes. DeepSeek-R1 is available for anyone to access, use, study, modify and share, and is not restricted by proprietary licenses.
Who Owns Deepseek?
DeepSeek operates independently but is solely funded by High-Flyer, an $8 billion hedge fund also founded by Wenfeng. The company released its first product in November 2023, a model designed for coding tasks, and its subsequent releases, all notable for their low costs, forced other Chinese tech giants to lower their AI model prices to remain competitive. In an interview last year, Wenfeng said the company doesn’t aim to make excessive profit and prices its products only slightly above their costs.
Is Deepseek Really That Cheap?
Not everyone is buying the claims that DeepSeek made R1 on a shoestring budget and without the help of American-made AI chips. Scale AI CEO Alexandr Wang told CNBC on Thursday (without evidence) DeepSeek built its product using roughly 50,000 Nvidia H100 chips it can’t mention because it would violate U.S. export controls that ban the sale of such chips to Chinese companies. Billionaire Elon Musk supported the theory and said it was “obviously” true in a post on X. Rasgon expressed similar thoughts in a note Monday, writing that DeepSeek did not quantify the resources it used to develop the R1 model in its reports and that while “the models look fantastic… we don’t think they are miracles.” He also said the $5 million cost estimate may accurately represent what DeepSeek paid to rent certain infrastructure for training its models, but excludes the prior research, experiments, algorithms, data and costs associated with building out its products.
What To Watch For
How big of a hit Nvidia, the maker of highly sought-after artificial intelligence chips, takes Monday. The company is tracking toward an 11%, or $400 billion, loss, which would be the biggest single-day value loss ever for any company. That record is already held by Nvidia, which dropped almost 10% in September to lose $280 billion in market value.
Key Background
China and the U.S. are the major players in the artificial intelligence arms race that seemed to be led primarily by American firms OpenAI (backed by Microsoft), Meta and Alphabet. Last week, President Donald Trump backed OpenAI’s $500 billion Stargate infrastructure plan to outpace its peers and, in announcing his support, specifically spoke to the importance of U.S. dominance over China in the AI space. Artificial intelligence is largely powered by high-tech and high-dollar semiconductor chips that provide the processing power needed to perform complex calculations and handle large amounts of data efficiently. And while not all of the biggest semiconductor chip makers are American, many—including Nvidia, Intel and Broadcom—are designed in the United States. In 2022, the U.S. began limiting semiconductor exports to China in an attempt to stymie the country’s ability to advance AI for military applications or other national security threats.
Loading...