Chinese AI Startup DeepSeek shocks world with low-cost open-source models

(ECNS) -- Leading Chinese AI lab DeepSeek has successively released two cost-effective open source models, DeepSeek-V3 and DeepSeek-R1, quickly becoming the talk of Silicon Valley.

DeepSeek has made a significant mark in the generative AI landscape with the groundbreaking release of its latest large-scale language model (LLM), according to previous report.

Alexander Wang, CEO of Scale AI, a company providing training data to key AI players including OpenAI, Google, and Meta, said at the World Economic Forum in Davos that DeepSeek-R1 can compete with OpenAI’s recently released o1 model.

Wang added that DeepSeek-V3 represents a bitter lesson in Chinese Tech. “They work while America rests and catch up cheaper, fast, and stronger,” he wrote in a post released last December on X.

Many western media outlets believe that  recent advancements in China's large models have sounded an alarm in Silicon Valley.

Financial Times released a report How small Chinese AI start-up DeepSeek shocked Silicon Valley on the same day, focusing on whether U.S. AI companies can defend their technological edge.

“U.S. export controls, which limit Chinese companies' access to the best AI computing chips, forced R1's developers to build smarter, more energy-efficient algorithms to compensate for their lack of computing power,” Life Science, one of the biggest popular science websites, said Saturday.

“ChatGPT reportedly needed 10,000 Nvidia GPUs to process its training data, while DeepSeek engineers say they achieved similar results with just 2,000,” it added.

Last week, U.S. president Donald Trump announced a joint venture with OpenAI, Oracle and SoftBank to invest billions of dollars in U.S. AI infrastructure. The project is called Stargate, with an initial investment of $100 billion and up to $500 billion over the next four years, while DeepSeek V3 is more cost-effective and priced at less than $6 million dollars.

A Meta staff member said in a post on Teamblind that DeepSeek V3 rendered Llama 4 behind in benchmarks. “Adding insult to injury was the ‘unknown Chinese company with 5.5 million training budget,’” the staff member said.

Geiger Capital, a renowned investor, said “Deepseek is just as good, if not better, than OpenAI and costs 3% of the price…It took them 2 months and less than $6 million to build, using reduced-capability chips, while US companies are pouring in hundreds of BILLIONS. So… what happens to the Nasdaq?” he said in a X post  on Saturday.

According to Liang Wenfeng, founder of DeepSeek, his research team is composed of young talents  eager to prove themselves, including PhD students from China’s top universities, such as Peking University and Tsinghua University. “Our core technical positions are mostly filled by people who graduated this year or in the past one or two years,” Liang told 36Kr in 2023.

"Recruiting the top 1 percent of geniuses to accomplish what 99 percent of Chinese companies cannot achieve," a graduate interviewed by DeepSeek said.