Chinese AI startup DeepSeek announced on Monday, January 27, that it is temporarily restricting new registrations to ensure the smooth operation of its services. According to the incident report, the decision was made "due to large-scale malicious attacks on DeepSeek services."

  • It noted that users who are already registered can log in as usual.
  • The startup's website also experienced an outage on the same day. The company said the issues with the API and access to user accounts had been resolved.
  • The disruptions coincided with the company's rapid growth in popularity. Over the weekend of January 25-26, the AI assistant gained immense popularity, topping the free app rankings in the App Store in the United States, United Kingdom, United Arab Emirates, Japan, South Korea, and China.

About Deepseek

Last week, DeepSeek unveiled a free assistant, the Deepseek R-1, which the company claims uses less data and costs significantly less than competing models. The release could be a game-changer in driving down the cost of AI development.

The assistant is powered by the DeepSeek-V3 model, which the company calls "a leading open-source model that can compete with the world's most advanced closed-source models." According to the developers, the DeepSeek-V3 model was trained on less powerful Nvidia H800 chips at a cost of less than $6 million.

DeepSeek's achievement has caused a stir in Silicon Valley, calling into question U.S. dominance in AI and the effectiveness of U.S. sanctions on advanced chips for China.

DeepSeek is a small startup based in Hangzhou, founded in 2023. Since then, dozens of Chinese tech companies, large and small, have released their own AI models, but DeepSeek is the first to be praised by the US tech industry for matching or even surpassing the performance of leading US models.

Like other Chinese chatbots, DeepSeek has its limitations. For example, it avoids answering questions and redirects users to other topics when asked about Chinese leader Xi Jinping's policies.

UPD 29.01: On January 29, OpenAI accused DeekSeek of using of its models for training purpose, raising intelectual properties concerns.