DeepSeek V4 launches with 1.6 trillion parameters and Huawei backing: the AI war shifts

The release of DeepSeek V4 marks a turning point in the global artificial intelligence race. This open-source model, with 1.6 trillion parameters, free of charge and optimized for the most widely used development tools, rivals closed American models, lagging behind by an estimated three to six months only. Backed by an alliance with Huawei, it is presented as proof of a Chinese technological ecosystem able to operate autonomously.

The event is unanimously described as major: after the 'Sputnik moment' of 2025, several actors see it as a new milestone. According to work from Stanford University, the performance gap between Chinese and American models has effectively closed. The central question then shifts: it is no longer only about who holds the best AI, but about who distributes it most widely. Companies from Lagos to Singapore can now access a model comparable to market benchmarks without paying any license fee.

Its interpretation, however, divides deeply. Some actors see innovation and technological emancipation, while Washington denounces a systematic extraction of intellectual property, with an official memo released on the very day of the launch. The open-source nature fuels the same ambivalence: a commercial opportunity for some, a vector of technological extraction for others.

Uncertainties remain. The alliance between DeepSeek and Huawei raises questions, still little addressed, about its consequences for competing semiconductor industries. The security of a widely open model and the share of Western-origin training data also remain debated points whose true scale has not been settled.