Deepseek breaks beliefs about the cost of AI, leaving the technological giants of us staggering

Melbourne:

Almost $ 1 billion (US $ 600 billion) was eliminated from the value of the Nvidia artificial intelligence microchip manufacturer during the night on Monday, when a little -known Chinese startup, Deepseek, threatened to fly to the US technological market.

While Nvidia suffered the greatest loss of one day in the history of Sharemket, other technological giants: Microsoft, Alphabet and Amazon, which are strongly investing in competitors tools, including Chatgpt and Gemini, were also beaten.

The defeat was caused by the shock of investors for the claimed performance of the New Chatbot R1 of Deepseek. It was reported that China was more advanced than its competitors and less expensive to develop.

Deepseek R1 He has shot, becoming the best free downloaded application in Apple’s App Store, since American technology and related prices of actions fell dramatically.

Why technological actions were submerged deeply

The market was surprised by Deepseek, providing what is equivalent to cheaper technology but comparable performance.

This has drastically changed the expectations of the computer power market, showing more it can be done for less. It has also compromised the competitiveness of the existing products and developments of the US technological companies.

Shares prices are promoted by market expectations. Deepseek R1’s claimed performance caused an important review of expectations about what was technologically possible and how cheap could develop and operate.

Investors have quickly incorporated the news of an low -cost Chinese competitor in shares prices, anticipating that this new participant could interrupt the market and erode the competitive advantage of existing leaders.

Who is Elsek and what is R1?

Depseek was founded in 2023 by the Chinese Coverage Fund Steering wheelthat he had been using exclusively in trade since 2021.

Deepseek develops Large language models (LLM) That can support chatbots and other tools based on AI. R1 is the last iteration of the Depseek chatbot model and the underlying model. It is based on previous versions of generative models developed by Depseek, and considerable amounts of data, but it is a surprising leap forward in performance and cost.

Technology investors believe that R1 coincides or exceeds competitors, including Chatgpt 4.o1 of OpenAI in Numerous reference points.

However, there are some key differences:

1. The underlying model R1 works in a much less intensive way. It is much cheaper to develop and execute, which requires fewer data and computer power.

2. The training of the model was possible despite the fact that the export prohibition of the United States prevented Chinese companies, such as Deepseek, accessing chips from US companies such as Nvidia. The Biden administration had laws introduced Restricting the sale of certain chips and computer machinery to China, in a movement destined to prevent its rival from accessing some of the most advanced technologies in the world.

3. Training data and data loaded for R1 are found in servers in China. Given the concerns about the privacy of data and intellectual property, they have already raised on companies based in the United States, having data under jurisdiction of the Chinese Communist Party (CCP) is possibly even more worrying.

4. The Chatbot program code is free to download, read and modify, unlike Chatgpt. However, this is something a false transparency: what matters the most is the underlying model, not the chatbot code.

5. R1 is known for censor your answers In line with the values ​​of the Chinese Communist Party.

The future of AI and Technology actions

It is unknown if this blockade in the price of technological actions is an irrational panic that will be reversed or if it simply reflects the right prices. The future costs and benefits of AI are still uncertain.

This is a technological and economic question.

In technological terms, it has not yet been seen if R1 really requires less computer power and less data to train and use.

Economically, there are potential winners and losers. IA users can win with cheaper access to AI, and LLMS in particular, which leads to more associated adoption and productivity gains. Existing producers such as NVIDIA can lose in what was a market with few real competitors.

In more general terms, society can benefit from the less computational intensive AI and, therefore, more efficient in energy. However, the geopolitical risk of a single country that captures the market, together with concerns about data privacy, intellectual property and censorship can exceed benefits.

(Author: Michael J Davern, Professor of Accounting and Commercial Information Systems, University of Melbourne; Matt Pinnuck, Financial Accounting Professor, University of Melbourne)

(Declaration of dissemination: Michael J Davern has previously received funds from CPA Australia for industry research on artificial intelligence; Matt Pinnuck does not work, consult, owns actions or receives funds from any company or organization that benefits from this article, and Ha did not reveal relevant affiliations beyond his academic appointment).

(This article is published again The conversation Under a Creative Commons license. Read the Original article.