Explaining-what is Deepseek and why does it interfere with the AI ​​sector?

On By Littum

By Eduardo Baptista

Beijing (Reuters) – Chinese Start -up Deepseeks Launch of its latest AI models, as it says, is on a pair or better than industry -leading models in the United States at a fraction of the cost, threatens to interfere with the world order.

The company has attracted attention in global AI circles after writing in a paper last month that the training of Deepseek-V3 demanded less than $ 6 million worth of Computer Power from NVIDIA H800 chips.

Deepseeks AI assistant, run by Deepseek-V3, has overtaken Rival Chatgpt to become the best-evaluated free application available in the Apple App Store in the US.

This has raised doubts about the rationale behind some US tech companies’ decision to pledge billions of dollars in AI investments and shares of several major tech players, including Nvidia, have been hit.

Below are some facts that the company is shaking the AI ​​sector worldwide.

Why Deepseek causes a voting?

The release of Openais Chatgpt at the end of 2022 caused an encryption among Chinese tech companies that rushed to create their own chatbots driven by artificial intelligence.

But after the release of the first Chinese chatgpt -equivalent, made by the search engine Giant Baidu, there was widespread disappointment in China at the gap of AI capacities between US and Chinese companies.

The quality and cost -effectiveness of Deepseeks models has turned this tale upside down. The two models that have been flooded with praise by both Silicon Valley leaders and US tech business engineers, Deepseek-V3 and Deepseek-R1 are on par with Openai and Meta’s most advanced models, the Chinese start-up said.

They are also cheaper to use. Deepseek-R1, released last week, is 20 to 50 times cheaper to use than the Openai O1 model, depending on the task, according to a post on Deepseeks official WeChat account.

But some have publicly expressed skepticism about Deepseek’s success story.

Scale Ai CEO Alexandr Wang said during an interview with CNBC on Thursday without presenting evidence that Deepseek has 50,000 NVIDIA H100 chips that he claimed would not be revealed because it would offend Washington’s export control, to ban such advanced AI chips from that be sold to Chinese companies. Deepseek did not immediately respond to a request for comment on the claim.

Bernstein analysts highlighted in a research note on Monday that Deepseek’s total training costs for his V3 model were unknown, but was much higher than the $ 5.58 million that Startup said was used for computing power. The analysts also said that the training costs for the equally recognized R1 model were not revealed.