DeepSeek: the 'Sputnik moment' for global AI

If you've been keeping up with the latest news in artificial intelligence, you've probably come across the name DeepSeek. This young Chinese startup, founded in 2023 by Liang Wenfeng, has just made a splash with the launch of DeepSeek-R1, an advanced language model that, despite competing directly with Western giants, has managed to develop on a surprisingly smaller budget. So what’s the secret? A different understanding of AI and a much more intelligent use of computational resources.

The DeepSeek-R1 model has gained industry attention because of its reasoning ability and because it does not rely on large amounts of specialised hardware (prompted by China's restrictions on the import of microchips). Rather than requiring an arsenal of high-performance chips, this model has adopted an optimised training strategy that reduces the need for expensive infrastructure. In other words, DeepSeek's developers have refined the way the model is trained, making more use of reinforcement learning to allow the AI to progressively develop complex behaviours. The immediate consequence has been significant savings in development and deployment costs, news that has not gone unnoticed in the financial markets, where the shares of some processor companies (Nvidia for example) have suffered heavy losses following the announcement.

DeepSeek has opted for an open source philosophy, which opens up a whole new range of questions and reflections

And while this has been the most striking, suggesting a worsening of the sector's prospects, it may actually be the opposite, as this has brought back to the table Jevons' Paradox, the 19th century theory that reminds us that improving the efficiency of a resource often ends up increasing its consumption. In the AI world, this could mean that if these models become cheaper and easier to create and use, the demand for AI-based applications and services will skyrocket, driving up the use of computational resources. So rather than quenching our thirst for processing power, inventions like DeepSeek-R1 could actually stimulate it further, raising questions about the environmental impact of such energy expenditure.

In addition, DeepSeek has opted for an open source philosophy, which opens up a whole new range of questions and reflections. On the one hand, transparency is generally welcome in any technology project. On the other hand, we are talking about a company that operates in China and that, according to some reports, has ties to state infrastructures, which could imply tensions over intellectual property or data privacy. In the end, any business that wants to integrate this technology would have to carefully weigh risks and benefits, especially when information security is at stake.

For many analysts, DeepSeek's entry onto the AI circuit is a true 'Sputnik moment': that instant that marks a before and after that can accelerate competition and innovation at a frenetic pace. There could well be a repeat of the phenomenon we saw in previous technological revolutions, when governments, investors and companies turned to new research and partnerships in order not to be left behind. In this context, efficiency and ethics become the pillars on which AI will have to sustain its rapid growth, as it seems clear that it will play a crucial role in the future of the economy, security and, ultimately, society as a whole.