Black Monday at Wall Street for the main players related to the AI. To trigger sales is the case of the Chinese startup Deepseek, whose success is putting the silicon Valley giants as alert who in the last two years have thrived on the euphoria linked to the boom of artificial intelligence.
Deepseek quickly distinguished himself for his innovation, creating models that, with limited resources, compete with the US giants. The Chinese company has gained attention to its ability to obtain high performance with less expensive hardware, challenging the limitations imposed by technological sanctions and thus showing that the ingenuity can overcome certain barriers.
The new model proposed by the Chinese Deepseek is so much apprehenting the Big Tech uses as it requires lower investments in artificial intelligence infrastructures. The prospect of a more efficient technology from an energy point of view and capable of turning on less powerful processors than those necessary with US technology inevitably feeds the fears of a new status quo for the IA.
The decomposed reaction of the markets. Black swan for the whole US tech?
The Nasdaq Composite came to sell 3%, while the S&P 500 gives 1.5%. Among the individual securities, the 17% collapse of Nvidia stands out, thus losing the scepter of largest company in the world for market value. Broadcom also very badly with -13%. Considering the sum of the losses of Nasdaq and the European technological underground, over 1,000 billion market capitalization are going in smoke today.
Deepseek presented his latest R1 model which is capable of reaching comparable performance, if not better, compared to GPT-4 by Openii, Claudeai and Llama 3.1 of Meta, at decidedly lower costs. Deepseek is likely to be the black swan for the Wall Street Tech sector? Soon to say it. Certainly Deepseek is a bit of the Temu for Chatgpt, Claude and Copilot. “We believe that the main actors in the world of artificial intelligence (Openai, Deepmind, Google, Meta Ai, Anthropic, Stability Ai) will take inspiration from Deepseek to review their processes, using less hardware (i.e. GPUs) but optimizing with choices Made by the Chinese start-up ”, asserts Filippo Diodovich, Senior Market Strategist of IG Italia.
Nvidia is the giant that risks the most
What jumps to the eye is the efficiency in terms of costs. Deepseek trained the model using significantly less powerful hardware, in particular the NVIDIA H800 chips, and with a much lower overall expense than habitual standards. With about 55 million dollars, Deepseek has reached what many thought of hundreds of billions.
“Nvidia, key actor in the chip market for the AI, has long benefited from the insatiable demand for its high -performance GPUs. However, Deepseek’s ability to provide competitive models using less chips and less recent models has raised Concerns on a potential slowdown in the application for the premium products of Nvidia “, argues Giacomo Calef, Country Head Italy of NS Partners. Analysts provide that, if other startups replicated the approach oriented to the efficiency of Deepseek, the growth of Nvidia revenues could face significant obstacles.
In general, Deepseek’s ascent questioned the prevalent assumption that the advanced development of AI requires massive and hardware calculation resources at the cutting -edge. By demonstrating that efficiency and innovation can overcome the rough calculation power, the company has forced a revaluation of the trajectory of the AI sector.
A change of scenario that could push the US techs to rethink their strategies, potentially leading to a wave of cost reductions and innovations focused on efficiency. “In addition – adds Calef – the open -source model of Deepseek, which allows free commercial use and study, could accelerate the democratization of IA technology, further intensifying the competition”.
Openi and the Chinese threat
Investors are realizing only in these hours of the threat represented by Deepseek. The Deepseek-R1 and Deepseek-V3 models, according to the benchmark shown, exceed top models such as those of Openii in specific areas, such as mathematics and programming, using a fraction of resources. “This implies that Deepseek can offer performance similar to much lower operating costs, threatening the market position of Openai which is based on more expensive models to train and maintain (we talk about a saving of 50x)”, remarked Paolo D’Alfonso of The Lighthouse, Finint Private Bank Finint Office.
Deepseek has also made its open-source models, which allows developers and companies to use, modify, and improve the model without the cost of licenses. This could attract a global community of developers, reducing dependence on proper solutions such as Openii.
“With the open -source approach – adds D’Alfonso – Deepseek offers a transparency that contrasts with the closed models of Openai, making the process of reasoning and development more accessible, which can undermine trust in the operai models perceived as less transparent.