What is DeepSeek and How is it Transforming the AI Industry?
The recent launch of DeepSeek’s advanced AI models has the potential to shake up the global technology landscape, as the Chinese startup claims their performance rivals or exceeds that of leading US models while costing significantly less.
DeepSeek gained traction in the international AI community after revealing in a paper last month that developing DeepSeek-V3 required under $6 million in computing resources utilizing Nvidia H800 chips.
Moreover, DeepSeek’s AI Assistant, which operates on the DeepSeek-V3 platform, has surpassed ChatGPT to claim the top spot among free applications on the US Apple App Store.
This development has sparked questions regarding the motivations behind certain US tech firms committing billions to AI initiatives, leading to declines in the stock prices of several major tech companies, including Nvidia.
What’s stirring the interest around DeepSeek?
The launch of OpenAI’s ChatGPT in late 2022 prompted a flurry of activity among Chinese tech companies eager to develop their own AI-driven chatbots.
However, following Baidu’s release of the first Chinese equivalent to ChatGPT, many in China expressed disappointment over the apparent disparity in AI capabilities between US and Chinese firms.
DeepSeek’s cost-effective and high-quality models have completely shifted this narrative.
DeepSeek-V3 and DeepSeek-R1, which have received accolades from leaders in Silicon Valley and engineers at US tech firms alike, are claimed by the startup to be competitive with the most advanced models from OpenAI and Meta.
Additionally, using DeepSeek’s models is significantly cheaper. The recently launched DeepSeek-R1 is reported to be 20 to 50 times less expensive than OpenAI’s o1 model, depending on the specific task, according to a statement on DeepSeek’s official WeChat account.
Nonetheless, some industry figures have voiced skepticism regarding DeepSeek’s achievements.
During a CNBC interview on Thursday, Scale AI CEO Alexandr Wang alleged, without presenting evidence, that DeepSeek possesses 50,000 Nvidia H100 chips, which he suggested could not be divulged due to Washington’s export restrictions against selling such advanced AI chips to Chinese companies.
DeepSeek did not provide a comment in response to this allegation.
In a research note released yesterday, Bernstein analysts indicated that while DeepSeek claimed to have spent $5.58 million on computing power for its V3 model, the actual training costs remain uncertain but are likely much higher.
The analysts also noted that training costs for the well-regarded R1 model were not disclosed.
Who is behind DeepSeek?
DeepSeek is a startup based in Hangzhou, with its controlling shareholder being Liang Wenfeng, co-founder of the quantitative hedge fund High-Flyer, according to Chinese corporate documents.
In March 2023, Liang’s fund stated on its official WeChat account that it was “starting anew,” shifting its focus beyond trading to establish a “new and independent research group to explore the essence of AGI” (Artificial General Intelligence), with DeepSeek emerging later that year.
OpenAI defines AGI as autonomous systems that excel beyond humans in most economically valuable tasks.
The extent of High-Flyer’s investment in DeepSeek remains unclear. Records reveal that High-Flyer shares office space with DeepSeek and holds patents related to the chip clusters utilized for training AI models.
High-Flyer’s AI division announced via its official WeChat account in July 2022 that it manages a cluster of 10,000 A100 chips.
How does Beijing perceive DeepSeek?
DeepSeek’s achievements have caught the attention of top political leaders in China. On January 20, the date DeepSeek-R1 was publicly launched, Liang was present at a closed-door symposium hosted by Chinese Premier Li Qiang for business leaders and experts, as reported by state news agency Xinhua.
Liang’s attendance at this event could symbolize that DeepSeek’s success aligns with Beijing’s objectives to navigate around Washington’s export restrictions and strive for self-sufficiency in critical sectors such as AI.
A similar forum last year was attended by Baidu’s CEO, Robin Li.