Læs hele analysen fra Danske Bank her
AI’s Sputnik moment ?
A short Q&A on DeepSeek
What happened?
On 20 January the Chinese start-up DeepSeek released its AI model DeepSeek-R1.
The model quickly became the most downloaded app in the US beating OpenAI’s ChatGPT. The model stands out for the following reasons:
• Caught up in performance: DeepSeek-R1 is a reasoning model that performs at par or better with the most recent versions from major US competitors such as OpenAI, Meta, and Google. Reasoning models are needed for tasks where simple pattern recognition is not enough, not least useful for researchers and developers (see comparison of model performance with OpenAI models on page 2).
• Much lower cost: The model has been trained at a cost supposedly of USD6.5 million which compares to much more higher costs among peers. For example OpenAI’s annual budget is USD 5bn per year.
• Using less computing power: The model is developed with far less computing power and supposedly without using the most advanced chips from Nvidia, which are sanctioned by the US. A DeepSeek research paper says it is developed using Nvidia H800 chips which were developed by Nvidia specifically for the Chinese market to comply with US sanctions. Some claim, though, that DeepSeeks parent, hedge fund High-Flyer, has been able to buy the more advanced H100 AI chips despite sanctions.
• Offered at low cost: The app is offered at no cost in the app store while the model’s API is offered at very low cost of $0.14 for one million input tokens, compared to OpenAI’s rate $7.5 rate for the o1 model. • Open-source model: As Meta’s Llama model, the DeepSeek model is an opensource model making it transparent and available for developers to work on top of it. What are AI experts saying?
• US venture capitalist Marc Andreesen has called the model launch “AI’s Sputnik moment” adding that it “is one of the most amazing and impressive breakthroughs I’ve ever seen”. • In an interview at World Economic Forum, MicroSoft CEO Satya Sadella said “We should take the developments out of China very, very seriously”. In a post on X he added that “as AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of.”
• OpenAI CEO Sam Altman said on X that “r1 is an impressive model, particularly around what they’re able to deliver for the price.
Læs hele analysen
