The release of a new generative AI app called DeepSeek has created uncertainty across global stock markets.
The Chinese app was significantly cheaper to develop than other AI systems.
In the days since its release, DeepSeek has topped Apple’s global App Store charts and been hit by a cyberattack.
Here’s what you need to know.
DeepSeek
DeepSeek was founded by Chinese hedge fund manager Liang Wenfeng in July 2023.
Recently, it has released two AI models: Deepseek V3 and R1 (more on this later).
V3 powers DeepSeek’s AI chatbot, which was released to app stores under the name DeepSeek earlier this month.
According to the company, V3 cost $US5.8 million ($AU9.3m) to develop. In comparison, Meta is expected to spend up to $US65 billion on AI systems this year.
Gen AI
Like OpenAI’s, the DeepSeek app is a ‘chatbot’ based on a ‘large language model’. These bots use a large database of online text and images to generate new material.
The DeepSeek app can summarise blocks of text, generate writing based on prompts, and answer user questions based on its interpretation of web search results.
Both DeepSeek and ChatGPT use computer chips produced by U.S-based tech giant Nvidia.
Nvidia chips
In June 2024, Nvidia became the most valuable company in the world, with a market value of $US3.2 trillion ($AU5.1t).
Its boom in value was due to Nvidia’s range of chips, which are able to handle the complex processing required for AI systems to function.
Your contribution ensures The Daily Aus can continue doing the work you love.
I want to give $10 monthly to help TDA keep doing the news.
Your details:
Liang (DeepSeek’s founder) has reportedly stockpiled thousands of one particularly powerful Nvidia chip. Since then, new U.S. restrictions have blocked the chip from being exported to China.
Last month, DeepSeek announced it had trained one of its new models, V3, using a less powerful Nvidia chip that’s not subject to U.S. restrictions.
It released the V3-powered app DeepSeek to the public this month.
The news that a powerful AI model could run on a cheaper chip contributed to the stock market drop this week.
Nvidia saw the biggest-ever drop in any company’s value on a single day.
The other factor is DeepSeek’s second new product, R1.
R1
As well as releasing the V3-powered DeepSeek app this month, the company released an AI model called R1.
This is a ‘reasoning’ model, which means it’s capable of more complex ‘thinking’ to solve problems, and of showing its ‘thought’ patterns.
OpenAI also has a reasoning model, called OpenAI-o1.
DeepSeek says R1 can outperform its OpenAI competitor and costs less to run. It’s made the model ‘open-source’, which means anyone can use or adapt its code.
Reaction
The release of V3 and R1 caused global stock markets to drop this week, as investors sold their tech stocks.
DeepSeek said it was subject to “large-scale malicious attacks” in the past 24 hours. It responded by placing a temporary limit on new registrations.
In a, OpenAI CEO Sam Altman said: “R1 is an impressive model, particularly around what they’re able to deliver for the price… It’s legit invigorating to have a new competitor!”
