News

In the last 24 hours, NVIDIA's shares have dropped by 17%, and the company's market capitalization has decreased by $600 billion. The reason was the unexpected emergence of the Chinese startup DeepSeek, which introduced a revolutionary AI model capable of competing with OpenAI's solutions.

DeepSeek R1 – a model developed with a budget of only $6 million, yet it rivals OpenAI's multi-billion dollar projects.

Who is behind DeepSeek?

The company's founder, Lian Wenfeng, started by creating the quantum hedge fund HiFlyer, which managed assets worth over $15 billion. To analyze financial markets, the company accumulated NVIDIA graphics processors over the years and developed high-performance computing solutions.

When the US imposed sanctions on the supply of advanced chips in 2023, Chinese companies found themselves in a difficult position. However, DeepSeek already had 10,000 NVIDIA A100 chips at its disposal, and the team found a way to overcome technological limitations.

Innovative approach to AI training

Unlike traditional models trained on ready-made data, DeepSeek R1 uses the principle of reinforcement learning. This means the model independently searches for solutions, evaluates their effectiveness, and adjusts its approach.

Moreover, DeepSeek R1 is based on two key technologies:

• Multi-Head Latent Attention (MLA) – allows the model to analyze data more efficiently by simultaneously working with different aspects of information.
• Mixture of Experts (MoE) – an intelligent system that activates only the necessary parts of the neural network for specific tasks, reducing energy consumption.

DeepSeek transitions to Chinese chips

Another revolutionary news was the possible use of the Chinese processor Huawei Ascend 910B by DeepSeek instead of NVIDIA solutions. Despite a 5% performance loss, this allowed the cost of running the neural network to be reduced by 70%.

DeepSeek R1 requires only 0.1 of the computational power needed by Meta's Llama 3.1, while providing similar results.

AI for pennies: a threat to OpenAI?

DeepSeek's main blow to OpenAI was in terms of price. The cost of DeepSeek's API is 55 cents per million input tokens and $2.19 per million outputs. Meanwhile, OpenAI charges $15 and $60 respectively for similar services – a 27-fold difference.

Open source – a blow to monopolies

DeepSeek made its code available under the MIT license, allowing developers to modify the model, run it locally, and adapt it to their needs. This provides enormous opportunities for scientific research and independent development.

Impact on the tech market

Despite the hype around DeepSeek R1, it's important to understand that it is not an equivalent to ChatGPT 4.0, but a tool for solving complex mathematical and programming tasks. However, the very concept of AI accessibility has already alarmed Western tech giants.

In the long term, OpenAI and NVIDIA may adapt by implementing new optimization methods, but the fact remains – the Chinese model has shown that multi-billion dollar investments do not always guarantee leadership in AI.

The future of AI: West versus East?

Western companies have been developing AI for years with a focus on giant data centers and multi-billion dollar budgets. DeepSeek has shown that another direction is possible – optimization and efficiency.

Can OpenAI respond to this challenge? Is an era of cheap but powerful artificial intelligence coming? The coming months will show how drastically the technology market will change.

DeepSeek has proven: a breakthrough in AI is possible without multi-billion dollar investments. The future of artificial intelligence now looks completely different.