DeepSeek is 'fantastic' but not a 'miracle', says Bernstein, doubts $5 million price tag

DeepSeek

DeepSeek looks "fantastic" but is not a "miracle" and not built with $5 million, Bernstein has said in a report about the little-know Chinese firm whose low-cost AI assistant has taken the tech world by storm and even roiled markets across the world.

The report said the claim of it being built in $5 million is misleading and doesn't show the full picture.

It stated that "we believe that DeepSeek DID NOT "build OpenAI for USD 5M"; the models look fantastic but we don't think they are miracles; and the resulting Twitter-verse panic over the weekend seems overblown".

The report said DeepSeek has developed two main families of AI models— DeepSeek-V3 and DeepSeek R1.

The V3 model is a large language model that uses a Mixture-of-Experts (MOE) architecture.

To train the V3 model, DeepSeek used a cluster of 2,048 NVIDIA H800 GPUs for about two months, totalling approximately 2.7 million GPU hours for pre-training and 2.8 million GPU hours including post-training, it said.

GPU hours is the total amount of time that a graphics processing unit is used for processing tasks.

Also Read | What is DeepSeek and how to download it

The report said the “$5 million” figure does not take into account the extensive research, experimentation and other costs involved in developing the model.

AI chip leader Nvidia  was hit the hardest by the unveiling of DeepSeek’s assistant and lost close to $593 billion in market value on January 27, a record one-day loss for any company on the fears of Chinese models stealing a march over the more expensive and richer American rivals.

The report said the R1 model was particularly impressive, though additional resources required to build this were not quantified in the company's research paper.

Bernstein concluded that while DeepSeek's achievements are fantastic, the “exaggerated claim” about building an OpenAI competitor for $5 million are “overblown”.