Deepseek, Stargate and the new AI arms race

On the episode of today DecoderWe are talking about the only thing that the AI ​​industry – and almost everyone about technology – was able to talk about last week: it is, of course, Deepseek, and how the model of AA Open Source built by a Chinese startup has completely upset conventional wisdom around chatbots, which they can do and how much they should cost to develop.

Deepseek, for those who do not know, looks a lot like Chatgpt – there is a website and a mobile application, and you can enter a small text box and make it speak. What makes him special is how it was built. On January 20, the latest major version of the startup, a reasoning model called R1, dropped only a few weeks after the last V3 model of the company, which both started to show very impressive IA reference performance. It quickly became clear that the Deepseek models work at the same level, or in some cases even better, as competitors of Openai, Meta and Google. Also: they are completely free to use.

But here is the real problem: while the OPENAI GPT-4 said that the cost of training was high that $ 100 million, the DEEPSEEK R1 costs less than $ 6 million to train, at least according to the Complaints from the company. In a few days, Deepseek became viral, becoming application No. 1 in the United States, and on Monday morning, he hit a hole on the stock market.

Panicked investors have suffered more than $ 1 billion on technological actions in a frantic sale earlier this week. NVIDIA, in particular, underwent a record drop in the stock market market of almost $ 600 billion during its decrease by 17% on Monday.

For more than two years now, the leaders of the technology have told us that the unlocking of the full potential of the AI ​​was to throw GPUs to the problem. Since then, the scale has been king. And the scale was certainly in the lead less than two weeks ago, when Sam Altman went to the White House and announced a new company center of $ 500 billion called Stargate which would be supposed to capacity ‘Openai to train and deploy new models.

The consequences were a bloodbath, to say it lightly. The venture capital Marc Andreessen sounded the alarm, calling Deepseek “The Spoutnik moment of AI” – and this seems to be how the AI ​​industry and the global financial markets treat it.

In Deepseek and Stargate, we have a perfect encapsulation of the two competing visions for the future of AI. One is closed and expensive, and it is necessary to place a sum of money and an increasingly growing faith in the hands of Openai and its partners. The other is disjointed and open source, but with major questions around information censorship, data confidentiality practices and if it is really as low cost as we say.

What is clear is that we have entered a new phase of the AI ​​arms race, and Deepseek and Stargate represent more than two distinct paths towards superintelligence: they also represent a new growing front in the relation American-Chinese and the geopolitics of AI geopolitics. It becomes particularly difficult, because President Donald Trump continues to wreak havoc on foreign relations with a new threat of prices on foreign semiconductors.

Many things happen here – and the news cycle is evolving very quickly. So to break everything, I invited Edge The main journalist of AI Kylie Robison in the show to discuss all the events of the last two weeks and to determine where the AI ​​industry then heads.

If you want to know more about what we talked about in this episode, see the links below: