What is DeepSeek, the Chinese AI company behind a $1 trillion Wall Street hit?

DeepSeek — a Chinese artificial intelligence start-up rivalling OpenAI's ChatGPT, Google's Gemini and Meta's Meta AI — has caused a brutal $US593 billion ($943 billion) hit to Wall Street.

The company last week unveiled its AI model called R1, which demonstrated most of the features of its competitors but for a fraction of the cost.

Global stocks consequently tumbled and caused Wall Street to suffer its biggest one-day loss in history.

But what exactly is DeepSeek, the company leaving tech giants worried?

READ MORE: Television host Dr Phil joins immigration raid

What is DeepSeek AI?

DeepSeek chief executive Liang Wenfeng founded the company in Hangzhou, China, in 2023.

Wenfeng had also co-founded the hedge fund and AI company High-Flyer, which amassed 10,000 of tech company Nvidia's A100 graphics processor chips that are used to build and run AI systems in 2022.

DeepSeek has released a variety of competitive AI models in just over a year of its start-up.

It announced its latest model, R1, late last year and officially launched the AI assistant last week.

After capturing global interest, DeepSeek's mobile app surged to the top of app charts above ChatGPT as the most downloaded free app.

So far, it has been downloaded almost two million times.

Marc Andreessen — one of the world's leading tech investors who advised US President Donald Trump on policy and warned that overregulation of AI in the US will enable China to get ahead — said "Deepseek R1 is AI's Sputnik moment".

"Deepseek R1 is one of the most amazing and impressive breakthroughs I've ever seen — and as open source, a profound gift to the world," he said in a post to X.

READ MORE: AI child abuse material surged in past year, AFP warns

What makes DeepSeek different from ChatGPT and other AI?

Most notably, the cost of powering the model.

According to DeepSeek, R1's API is 20 to 50 times cheaper to use than OpenAI's o1 model.

DeepSeek-R1's API costs just $0.88 per million input tokens and $3.49 per million output tokens, compared to OpenAI's API, which costs $23.92 and $95.70.

The company is also spending less on powering its AI, claiming it is costing it just $8.9 million.

Rival US companies have comparatively spent hundreds of millions, if not billions, on their AI technologies.

DeepSeek also describes its models as "open source", which means key components are free for anyone to access and modify.

It, however, has not shared the data it used for its training.

READ MORE: Police dog bails up alleged CCTV camera thief after chase

What are DeepSeek's main features?

DeepSeek R1 has a similar interface to ChatGPT and other AI models, with users able to type in their questions and receive an AI-generated response.

It differentiates itself from other AI models by articulating its reasoning and showing its train of thought when answering questions, which is then used for further training without the company having to feed the model new sources of information — effectively through self-learning capabilities.

It also learns and improves through feedback it receives.

Its multi-stage learning and cold-start data have boosted its performance to be "on par" with OpenAI's o1 model.

— With CNN and AP

DOWNLOAD THE 9NEWS APP: Stay across all the latest in breaking news, sport, politics and the weather via our news app and get notifications sent straight to your smartphone. Available on the Apple App Store and Google Play.