Unlock the editor’s summary for free
Roula Khalaf, editor of the FT, selects her favorite stories in this weekly newsletter.
The superinflated technological actions of the United States seemed a correction for months, but the trigger comes from an unexpected source. The last large language model of the China Depseek artificial intelligence company may not be a “Sputnik moment.” There are parallels, even so, with the early space race, when the ingenuity helped Soviet engineers maintain rhythm and sometimes overcome US rivals, despite their relative lack of computer power and avant -garde technologies. The United States won the moon race and established a sustainable advantage in space. But Deepseek’s advance increases the assumptions that have supported the American technological assessments, of an impregnable supremacy in AI that would extend to spending billions of dollars in chips and infrastructure.
VeteranThe achievement is to have developed a LLM that IA experts say that it achieves a similar performance to the US rivals of cost. Some of his statements have not yet been verified. However, if they are true, it represents a potentially formidable competitor.
First, as the goal of the United States, but unlike OpenAi or Gemini of Google, it is open source, ready to share the recipe for its secret sauce instead of keeping it blocked in the hope of extracting maximum financial gain. That makes it attractive to developing and building developers. Secondly, you can gather in a much smaller budget and with a lot Less computer power. This explains Monday Prices of the actions that sink Not only of the first -line technological companies, but of those that manufacture chip equipment and supply electrical hardware for data centers.
The big American companies focus on a particular model of scale by throwing large amounts of capital, data and computer energy to the problem. First MistralA new French AI company, and now Depseek has shown that companies with less resources can end with more intelligent and more efficient models. It is likely that the growing efficiency of these models also stimulates a greater demand for their use.
Deepseek also the highlight, that China is achieving technological leaps in AI despite the export controls introduced by the Biden administration aimed at depriving it of the most powerful chips and the advanced tools necessary to do them. The new Chinese AI companies have been forced to find inventive ways of extracting the greatest amount of juice from the chips they have. Far from suffocating Chinese innovation, Washington may have He stimulated him. And the success of Chinese engineers often trained in the country to increase efficiency and the search for solutions raises questions about whether the technological “pit” established by high -expenditure groups such as goal, Google, OpenAi and Anthrope is so wide and impregnable as they had thought.
Many advanced democracies will distrust a Chinese government seen in many ways as hostile to western interests that potentially acquire leadership in the most transformative technology of our time. However, some political leaders in other places, along with many consumers and developers, can welcome a market that is less dominated by a handful of US companies.
The open question is not necessarily who will develop the best AI models, but who can better apply them to real world’s tasks. Kai-Fu Lee, a Chinese IA pioneer, has long argued that China stands out in the front of the application, even if it can be delayed in the infrastructure. That was before the world of new Chinese companies was squeezed by political repression against technological entrepreneurs and the increase in investment in new US companies. Much more uniform.