The years of feverish hype about the technology of artificial intelligence has convinced that this is the next speculative bubble in the Silicon Valley – and the question arises as to how long giants like Openaai still raises billions in your search for a really groundbreaking AI can burn from dollars.
Now a Chinese company has presented a state-of-the-art AI model, which according to its own information has developed in less than two months, whereby the cost of training in the final stage was less than $ 6 million. Numbers that significantly undercut the investments of US companies in bots such as chatt and the infrastructure for their operation. And the market may now change fundamentally.
Deepseek-the name of the laboratory and its model-was created as a side project by Liang Wenfeng, co-founder of the HEGGINK High-Flyer, which began importing Nvidia process orchips in 2021. (The manufacturer based in California is an important supplier of high-performance graphics processors (GPUS), which provide enormous computing power that is required for training and operation of AI.) In 2023, high flyers brought a version of Deepseek as internal Tool onto the market to help recognize and predict market trends to improve trading decisions.
Who is more powerful: Chatgpt or Deepseek?
But last week the company published a “AI assistant” bot, Deepseek-V3, a large language model that is now the most frequently downloaded free app on Apple devices (even before chatt from Openai), as well as an argumentation model, Deepseek- R1, which according to its own statements fulfills the same standards as the comparable model from Openai. The laboratory has also presented a multimodal model that can create images that seem to exceed the competitive models Dall-E 3 and Stable diffusion.
These different newcomers alone would have been sensitive to risk capital companies and large technology companies that have invested billions in AI, including Microsoft, Meta, Google, the parent company Alphabet, Amazon and Nvidia. What really shaken these investors on Monday (January 27th) was the efficiency advertised by Deepseek: reports that Deepseek uses a limited number of chips with a reduced capacity of NVIDIA, which in turn significantly the operating costs and the price of premium models for consumers lowers.
In contrast, the CEO of Openaai, Sam Altman, only admitted a few weeks ago that the company loses money even on pro-subscriptions that cost $ 200 per month, due to the astronomical costs for the computing power that their software needed. With an alleged price of around $ 5.5 million for the last development phase, Deepseek-V3 is also a relatively cheap alternative to models, the development of which cost several ten million.
In addition, the codes from Deepseek Open-Source Codes, which can be freely distributed and changed by the users or can be carried out on a private device, without disclosing personal data. Meanwhile, US companies tend to keep the inner processes of their KIs as secret as possible.
The result of all this was a sudden loss of trust in the industry leaders, including several, which is involved in a $ 500-billion project to expand the AI infrastructure under President Trump, which is known as a stargate initiative. The chip manufacturer Nvidia was most affected and lost $ 600 billion in market capitalization when his share price collapsed by 17 percent-the largest one-day decline for a US company in history.
This sale indicated that tens of thousands of high-performance graphics processors may not be necessary for the next wave of AI models, which the giants of the Silicon Valley have merged into computing super clusters in order to accelerate their AI innovations. The strongly sunken demand for chips and huge data centers, as Trump proposed in the context of Stargate (an announcement that the AI shares drove up a few days ago) could completely redesign this business sector. Until the end of the stock market on Monday, the technology-based Nasdaq-100 share index had lost an incredible $ 1 trillion of its value on Friday.
When does the AI bubble burst?
Of course, Deepseek’s great success made the company to the popular goal from one moment to the other, and the company restricted the registration on Monday, while it recorded so -called “large -scale malicious attacks” to its service (but without access for existing ones To limit users).
At the same time, some analysts argued that the market panic about the long-term competitiveness of the United States could be an overreaction in the AI race. After all, Deepseek could show the way for a higher efficiency in models produced in America, some investors will buy in this break -in, and as a Chinese company, Deepseek is faced with some of the same national security concerns, the Bytedance, the Chinese owner of TIIKTOK, is plagued have.
On the other hand, it is difficult to ignore the questions that Deepseek raises in relation to the gigantic sums that American technology giants and risk capitalists have stuck from the Silicon Valley in AI, in the expectation of revolutionary (and profitable) results.
For some observers, it will look like such expenses are not only unsustainable, but ultimately also wasteful when you consider how much a foreign start-up has achieved with far less. The bladder has not yet burst, but with so many investors who hurry down, to put out assets that suddenly appear much riskier, you can hear how to drain air.