DeepSeek shakes the American AI market

Translated by Irene Cecchi

Despite the tag “artificial intelligence” is often used as a slogan in contexts that have nothing to do with it, the development of such technology is rapid and ongoing. The US appeared to be the uncontested leader of this new tech sector, dominated by OpenAI, Microsoft, Google and NVIDIA, but on January 27th 2025 the investors realised that the “AI race” is not only American but that China may have already overtaken the US.

DeepSeek is a Chinese startup of artificial intelligence that develops open source large language models (LLM) that was founded by Hangzhou only in May 2023. Despite the investors’ initial reluctance, doubting the possibility of putting on the market a valid product in the short-term, the first series of DeepSeel LLM was released in November of the same year. One year later, the series of models DeepSeek-V2 is out with two models and two chatbots, praised by the Financial Times for their price below the competitors’ average.

On January 20th 2025, the Chinese startup released new LLMs, DeepSeek-R1 and R1-Zero, based on the new architecture DeepSeek-V3. Based on these models, a chatbot app is now available for iOS and Android, an AI assistant that in just a few days overtook ChatGPT in terms of higher grading in the US App Store. What amazes about the Chinese chatbot is not only the rapidity to conquer the market but also the fact that DeepSeek-V3 required way less resources for development and functioning than its equal competitors. In particular, the main companies in the sector utilize up to 16.000 integrations in their computer to train their chatbots while DeepSeek claims using only 2.000 H800 specialized computer chips produced by NVIDIA. The Chinese chatbot training took less than two months and a cost that doesn’t go above five million dollars, about 5% of what OpenAI needed to develop the o1 model and even less if compared to the previous ones.

On January 27th, the US tech sector suffered a deep fall, with unprecedented market value losses. NVIDIA, the most affected company, lost 17% of its market value, about 600 billion dollars: it is by far the biggest impairment for a listed company ever (the previous record refers to META, when three years ago it lost 240 billion value). In just one day, NVIDIA passed from first to third place in the list of higher values, overtaken by Apple and Microsoft. Also META and Alphabet (Google group) suffered significant losses. The total losses registered by the American tech sector sums up to a thousand billion dollars.

The sector that seemed dominated by the US and with impossible entrance standards, reveals to be a “tech race” where China might be way ahead despite the American efforts to cut it out of the market. In October 2022, the Biden administration implemented new exports controls on merch directed to the People’s Republic of China, with the aim of obstructing the access to microprocessors’ market. Thea D. Rozman Kendler, Assistant Secretary of Commerce for Export Administration, commented on the amendment explaining that China invested “a lot of resources in supercomputing development aiming to become the world leader in artificial intelligence by 2030”. The US National Security Commission on Artificial Intelligence published a report in 2021 according to which, if China overtakes the US in the microchip sector, it would be able to get a strategic military advantage “in every war sector”. Regardless of all of this, even before the US implemented this restriction, Liang Wenfeng (founder of DeepSeek) already stored about 10.000 GPU A100 by NVIDIA.

Does this AI race only have two competitors right now? Yes, or at least only two can really aim to win. Despite the European Union's intensified investments in the sector, it is not able to get to the US or China’s level. The AI Act is not helping: aiming to create a jurisdictional framework for AI, risks slowing down innovation. India is another player that might emerge in the sector: in 2022 the Indian AI market reached 680 million dollars thanks to a blooming startup environment.

OpenAI and Microsoft already accused DeepSeek of stealing the data used by OpenAI to train ChatGPT: interesting if we consider that thousands of content creators, including the New York Times, accused OpenAI of stealing copyrighted materials to train its LLM. OpenAI never denied the charges but instead claimed that it is not copyright infringement since the content is allegedly included in the “legitimate usage” policy. The website 404 Media made a recap of the situation ironically called “OpenAI Furious: DeepSeek Might Have Stolen All the Data OpenAI Stole From Us”.

In a constantly developing context like this one, it’s difficult to think about what might happen in the future, what's sure is that DeepSeek’s impact on the market will not be remembered as one and only but as the symbol of the beginning of a tech race that will affect every other sector, from civil to industrial to militar, like never before. In this framework, the question of Taiwan will become more and more central.

Mondo Internazionale APS - Riproduzione Riservata ® 2025