DeepSeek, an AI startup, made waves in the global market with the release of its model in January this year. Overseas investors expressed concerns that cost-effective Chinese AI models might threaten the dominance of tech giants like Nvidia, prompting a sell-off of tech stocks and resulting in the evaporation of billions in market value during that week.
In response to the situation, DeepSeek announced it would temporarily halt and restrict access to its API services due to limited server capacity. On Thursday, the company disabled the top-up button on its platform to prevent further service disruptions. According to the South China Morning Post, the DeepSeek platform previously allowed users to interact via API, which served as a bridge between developers, applications, and the company's AI model.
On its official website, DeepSeek posted a notice stating that the ability for customers to recharge API credits had been paused, though existing balances could still be used without issue.
Since launching its chatbot, the Chinese AI startup has received strong market reactions. The company claims to have developed its AI chatbot at a significantly lower cost than competitors, leading to a surge in service demand since its January debut.
In a technical report, DeepSeek asserted the superiority of its model, noting it ranked higher than competitors such as OpenAI and Stability AI in text-to-image generation tasks.
Bloomberg reported that DeepSeek had restricted new user registrations to those holding mainland Chinese phone numbers.
DeepSeek's new model sent shockwaves through U.S. companies, triggering a $1 trillion decline in U.S. and European tech stock values in January. This event also prompted the U.S. to expedite efforts to address gaps in chip export restrictions for AI applications.
The U.S. government launched an investigation into DeepSeek’s AI model regarding data protection and security issues. Meanwhile, Ireland's Data Protection Commission requested additional details about the startup's operations.
Specifically, U.S. officials are investigating whether DeepSeek obtained Nvidia semiconductors through a third party in Singapore to circumvent restrictions. Additionally, regulatory agencies in South Korea and France have begun inquiries into how DeepSeek handles user data stored on servers in China.
DeepSeek revealed in its post that its low pricing strategy has startled both the market and some competitors. The company announced that discounts for model access will end on February 8. Currently, the price is set at 2 yuan per million input tokens and 8 yuan per million output tokens. Bloomberg reported that when the inference model goes live, DeepSeek will adjust its rates to 4 yuan per million input tokens and 16 yuan per million output tokens.
