China-based artificial intelligence startup DeepSeek caused a major shake-up in global technology stocks with the low-cost artificial intelligence model it introduced to the market. Nasdaq futures and technology stocks in Japan fell sharply on Monday, shaking investors’ confidence in the profitability of artificial intelligence and the massive demand for high-tech chips.

While DeepSeek rose to the top, others crashed

Nasdaq 100 futures fell 2.6 percent, while S&P 500 futures fell 1.4 percent. While Nvidia’s shares lost approximately 7 percent of their value, shares of giant companies such as Tesla, Amazon and Meta decreased by over 2 percent in Europe. Nvidia supplier Japanese Advantest attracted attention by falling 8.5 percent and Tokyo Electron falling 5 percent.

While all this was happening, DeepSeek rose to first place in application stores and left behind derivatives such as ChatGPT.

On the other hand, little is known about the small Hangzhou startup behind DeepSeek. DeepSeek researchers wrote in an article published last month that they used Nvidia’s H800 chips for training the DeepSeek-V3 model, which was released on January 10, and spent only $ 6 million. H800 chips were originally produced to get around US chip sales restrictions to China, but were later banned entirely. For perspective, let’s note that more than 100 million dollars are required for the training of Meta and OpenAI’s new models.

Additionally, SoftBank Group, which stands out with its artificial intelligence investments, experienced its biggest daily decline since September 30, losing more than 8 percent. The company announced last week that it would invest in Stargate, a $19 billion data center joint venture with OpenAI. However, the rise of DeepSeek has raised questions about the sustainability of such large-scale investments.

Did the balances change overnight?

Companies such as Nvidia, which are the stars of the artificial intelligence industry, have gained 196 percent in value since the beginning of 2024. However, DeepSeek’s low-cost model signals that the balance of power in the industry may be reshaped. This shows that artificial intelligence can grow not only with expensive infrastructures, but also with innovative and cost-oriented solutions.

However, we should not blindly dive into the “hype culture” surrounding DeepSeek.

We should point out that the DeepSeek V3 model and the DeepSeek R1 model, OpenAI’s “thinking” artificial intelligence model O1 rival introduced last week, are definitely powerful models. DeepSeek-v3 has comparable or better performance to GPT-4, Claude-3.5-Sonnet, and LLlama-3.1, depending on the benchmark. The first tests of R1, published on January 20, show that the model is at the same level as O1 in certain tasks in chemistry, mathematics and coding. According to the company, the R1 model is built on the V3 model.

However, in the world of artificial intelligence, what appears and what actually happens are often different. It is best to wait for some time to come to a conclusion. In our news about DeepSeek V3, we told you that this artificial intelligence thinks it is ChatGPT. Moreover, it was making the same jokes of ChatGPT.

Since DeepSeek does not disclose training data for DeepSeek V3, it is difficult to determine why the model says it is ChatGPT. However, this may indicate that the model may have been trained on public data generated by GPT-4 via ChatGPT. If this is the case, DeepSeek V3 is probably memorizing and repeating some of ChatGPT’s output.

So is this a problem? Actually not. It is known that the huge data required to train artificial intelligence models has been exhausted as of last year. Therefore, using artificial intelligence data while training artificial intelligence is becoming an increasingly acceptable phenomenon. However, this can also lead to situations where quality may decrease (as in DeepSeek V3).

Ultimately, both DeepSeek V3 and R1 are seen as a powerful part of the AI ​​industry. Most of the “strong” discourse here comes from cost effectiveness. These models are not a “revolution” for the artificial intelligence world, but it is a fact that they have dealt a major blow to the US’s stance on blocking China’s artificial intelligence progress.