Assessment and current situation 

DeepSeek, a Chinese startup specialized in artificial intelligence, has recently unveiled DeepSeek R1, an advanced reasoning model that rivals OpenAI's top solutions while being developed at a fraction of the cost. Its development is said to have cost only $6 million, compared to over $600 million for OpenAI's GPT-4, though this figure remains to be confirmed. Beyond its financial accessibility, its operational cost is also significantly lower, at less than $4 per million tokens, compared to over $100 for OpenAI. As an open-source model distributed under a permissive license, DeepSeek R1 stands out by fully disclosing its reasoning steps, enhancing transparency and improving the understanding of its decision-making process.

Significant repercussions on the AI industry 

The meteoric rise of DeepSeek R1 has sent shockwaves through the artificial intelligence industry, triggering major repercussions in financial markets. Within days, U.S. stocks lost $2 trillion in market capitalization, while NVIDIA saw over $500 billion wiped off its valuation. This financial upheaval threatens Silicon Valley’s long-standing dominance in AI, calling its leadership into question as Chinese players rapidly ascend. Now, global attention is shifting toward China, whose growing influence in artificial intelligence could reshape the balance of power in the tech world.

Our Expert's opinion 

DeepSeek R1 marks a major turning point in artificial intelligence, redefining established standards and challenging assumptions about the resources required to create high-performance models. "DeepSeek R1 represents a paradigm shift in AI, questioning the assumptions about the resources needed to build powerful models." Contrary to popular belief, these models are not entirely new. "We have been testing DeepSeek-Coder-V2 for two months and are impressed by its ability to integrate context, environment, and frameworks used by our developers, enabling intelligent autocompletion and relevant reasoning." These tests, conducted on-premise with a DeepSeek-Coder-V2-Lite-Instruct 16B model, highlight this technology’s potential to adapt to developers' specific environments, paving the way for advanced and optimized applications in generative AI.