In recent days, a name has been shaking up the world of artificial intelligence: DeepSeek. If you haven't heard of it yet, it's worth keeping an eye on. The AI model developed in China has not only overtaken ChatGPT as the most downloaded app in the US App Store, but is also challenging the entire business model of the big techs. And the most impressive thing? It did it with a fraction of the resources.
David versus Goliath: DeepSeek and the AI giants
While companies like OpenAI spend billions to train models like GPT-4, DeepSeek claims to have achieved similar performance with just $5.5 million. Efficiency. US restrictions on chip exports have forced the Chinese to innovate, using fewer resources more intelligently.
The numbers are shocking:
- US$ 6 million to develop the model (vs. US$ 100M+ for the GPT-4)
- 2,000 specialized chips (vs. 16,000 from competitors)
- API 50x cheaper
- Performance comparable to GPT-4 and Claude 3.5
- Open source model, available for anyone to test and use
What does this change in practice?
On a day-to-day basis, you may not notice much difference between DeepSeek and other AIs like ChatGPT or Claude. But the impact on the market is huge. The open source model allows developers to run it locally, without relying on expensive data centers. This could force AI giants to review their strategies.
And the impact is already being felt: the hype surrounding DeepSeek alone caused Nvidia's shares to fall by 12%. It also raises the question: if China has managed to get this far with few resources, what's next?
What lies ahead?
DeepSeek isn't just a new AI - it's a geopolitical message. The race between China and the US for supremacy in artificial intelligence has just entered a new, much more unpredictable phase. And in the meantime, we users are gaining more options and access to advanced technologies without having to pay a fortune.
Whether this revolution will change everything or whether it's just another passing hype, only time will tell. But one thing is certain: the game of AI will never be the same again.
