What is DeepSeek: Chinese AI model overtakes ChatGPT in app store downloads, sparks big losses on Wall St

On Monday, Chinese artificial intelligence startup DeepSeek took over rival OpenAI’s coveted spot as the most-downloaded free app in the US on Apple’s App Store, dethroning ChatGPT for DeepSeek’s AI Assistant.

Global tech stocks sold off and were on pace to wipe out billions in market cap.

Later on Monday, DeepSeek said it would temporarily limit user registrations “due to large-scale malicious attacks” on its services, though existing users will be able to log in as usual.

Tech leaders, analysts, investors and developers say that the hype — and ensuing fear of falling behind in the ever-changing AI hype cycle — may be warranted. Especially in the era of the generative AI arms race, where tech giants and startups alike are racing to ensure they don’t fall behind in a market predicted to top $USD1 trillion in revenue within a decade.

What is DeepSeek?

DeepSeek was founded in 2023 by Liang Wenfeng, co-founder of High-Flyer, a quantitative hedge fund focused on AI.

The AI startup reportedly grew out of the hedge fund’s AI research unit in April 2023 to focus on large language models and reaching artificial general intelligence, or AGI — a branch of AI that equals or surpasses human intellect on a wide range of tasks, which OpenAI and its rivals say they’re fast pursuing.

DeepSeek is still wholly owned by and funded by High-Flyer, according to analysts at Jefferies.

The buzz around DeepSeek began picking up steam earlier this month, when the startup released R1, its reasoning model that rivals OpenAI’s o1.

It’s open-source, meaning that any AI developer can use it, and has rocketed to the top of app stores and industry leader boards, with users praising its performance and reasoning capabilities.

Like other Chinese chatbots, it has its limitations when asked about certain topics: When asked about some of Chinese leader Xi Jinping’s policies, for instance, DeepSeek reportedly steers the user away from similar lines of questioning.

Another key part of the discussion: DeepSeek’s R1 was built despite the US curbing chip exports to China three times in three years.

Estimates differ on exactly how much DeepSeek’s R1 costs, or how many GPUs went into it. Jefferies analysts estimated that a recent version had a “training cost of only $US5.6m (assuming $US2 hour rental cost). That is less than 10 per cent of the cost of Meta’s Llama.” But regardless of the specific numbers, reports agree that the model was developed at a fraction of the cost of rival models by OpenAI, Anthropic, Google and others.

As a result, the AI sector is awash with questions, including whether the industry’s increasing number of astronomical funding rounds and billion-dollar valuations is necessary — and whether a bubble is about to burst.