On 21st January 2025, the day after his second inauguration, Donald Trump announced a massive investment of 500 billion dollars over the next four years to support the “Stargate” project. The operation, involving Japanese investment company SoftBank, MGX (an investment fund backed by the United Arab Emirates), cloud solution provider Oracle, and OpenAI (developer of ChatGPT), aims initially to develop physical and digital infrastructures across the country to support the development of “the next generation of AI”. Up to twenty massive data centres will be built in Texas. Work has already begun.
The promise of Stargate is commensurate with its cost: it will be able to develop “general-purpose AI”, create tools to fight cancer (statement by Mr. Ellison, founder of Oracle), “create hundreds of thousands of American jobs, and generate massive economic benefit for the entire world” (OpenAI statement, link below).
But above all, it seems that this project serves a vision, that of the advent of the “golden age” of the United States in terms of AI, of the creation of the GAI but also ultimately of superintelligence (statement by Masayoshi Son).
All these statements support the idea that only the United States can produce the AI tools of tomorrow and that this future necessarily involves the massive exploitation of data and the abundant investment that only this country can provide. In short, the United States says that it holds the future of humanity in its hands.
And yet, on another continent, a small whale is preparing to create a major tidal wave. DeepSeek-R1, China’s new open-source LLM, offers performance similar to that of certain OpenAI models and would appear to contradict this well-worn narrative (but not the Chinese surveillance narrative, as researchers discovered that DeepSeek users’ data are stored by China Mobile, which has strong ties with the state).
The success of the DeepSeek application and the announcement of its performance have tarnished the golden future foreseen by the United States. The major US AI players are feeling the pinch. In particular, Nvidia lost $590 billion in market value in one day, while SoftBank also fell by 6%.
DeepSeek models are said to have been trained for around $5.5 million, compared with $100 million for ChatGPT-4. The Chinese secret sauce? Among other things, it favours reinforcement learning rather than supervised learning, uses 8-bit coded numbers that require less computing power than 16- or 32-bit coded numbers, and uses a “mixture of expert” model (also used by MistralAI). Bigger is not better. Not always, at least. This idea is not new and was demonstrated by researchers (article’s link below).
As some compare it to the ‘Sputnik of AI’ (Marc Andreessen, venture capitalist from Silicon Valley, on Platform X), DeepSeek AI should be the signal for European and worldwide AI actors. Maybe 500 billion dollars are not required to create competitive AI systems and generate innovation.
The USA’s narrative of exclusivity on high-performance AI models seems weakened, and President Donald Trump admitted that DeepSeek should be “a wake-up call” for US companies. Will it be?
This raises another question. Is artificial general intelligence really the next step of AI, or is it just another narrative? Companies don’t need GAIs. They need tailored, high-performance tools adapted to their needs. This goes for consumers in their private use of AI tools. We already use several applications or devices adapted to our needs or the task to be tackled.
This is for us to define what progress will be, and it seems it doesn’t have to come from the USA. States are now in a race to achieve new AI innovation, and the EU, which stands between a hammer and an anvil, has to seriously start running.
Sources
https://openai.com/index/announcing-the-stargate-project/
https://www.nature.com/articles/d41586-025-00259-0
