Last Monday, a small Low AI Laboratory with a low profile called Deepseek unveiled a set of models of artificial intelligence so effective that they are Silicon Vall Ey.
These models are fifty times more effective than even American offers. The announcement has launched American technology giants like Openai, Google and Meta in a full -fledged crisis, while their camera strategies suddenly stop cutting it. Now everyone is laughing at them.
The man behind this shock wave is Liang Wenfeng, a healing fund manager who has transformed a secondary project into one of the most disruptive progress in History. The DEEPSEEK R1 model is a self-learning system that can improve itself without human supervision.
And here is what makes Silicon Valley Sweat: Liang Nivelle the playground for anyone – whether in China or elsewhere – which rises in the development of AI.
Le Long Match de liang wenfeng
In 2021, while all the others on the Chinese AI stage were focused on Big Tech, Liang quietly bought thousands of NVIDIA graphics treatment units (GPU) and began to experiment with AI. The inhabitants of the industry thought he was only another billionaire who flowed a hobby.
According to a trading partner who spoke to the Financial Times: “When he told us that he wanted to build a cluster of 10,000 chips, we thought he was crazy. He didn’t even explain why, just said, “It will change everything.” »»
Quick advance until 2023 and Liang launched Deepseek, hiring the best engineers of the AI directly from its coverage fund. Using the benefits of the Haid Fund High-Flyer, he built a team that not only understood AI but also controlled the infrastructure behind.
By 2024, Deepseek had developed R1, a language model that the initiates of the industry describe as a direct challenge to each adult of AI in the United States, unlike its American competitors, Deepseek was not interested in marketing. He focused on research, Liang using his own money to finance the operation.
“Deepseek offices look like a university laboratory,” said Financial Times. Based in Beijing and Hangzhou, the laboratory uses some of the best minds of AI in China, offering wages that compete with the owner of Tiktok Bytedance.
According to the report, Liang’s singular objective was to prove that China could innovate at the same level as the United States, and now it has done so.
-Timing could not have been more deliberate either. Liang’s release from model R1 coincided with its appearance at a high -level meeting in Beijing led by Li Qiang, the second most powerful politician in China.
Liang was the only leader in the guest AI. Li told entrepreneurs present to focus on permission of key technologies – a clear wink to China’s ambition to overcome the United States in the development of AI. Something tells us that we have tooth Donald Trump is not too happy with that.
The Silicon Valley race to catch up
The giants of American technology, caught off guard, are now rushing to answer. OPENAI, the company behind Chatgpt, announced a joint venture of $ 100 billion with Japanese softbank, nicknamed Stargate, aimed at building new IA infrastructure in the United States
Meanwhile, the XAI of Elon Musk extends its colossus supercaluler, providing to deploy more than a million GPU to form its Grok AI models. Google, Meta and Anthropic also launch billions to improve their computer grapes with the new generation blackwell fleas from Nvidia.
But American societies have a major drawback: secrecy. For years, the Silicon Valley has worked on a closed -door model, keeping the Percées of the AI locked behind the proprietary systems. Deepseek’s decision to take the open source of R1 overturned this story.
This triggered a panic wave in the United States, where companies are now pressure to decide to follow the example of Deepseek. But there are also concerns about whether Deepseek can maintain its momentum. Despite its success, the company’s resources are certainly limited compared to American giants.
“They built one of the largest computer grapes in China,” said a familiar source with the company. “But compared to what Openai and Google build, it is not enough. They will have to evolve if they want to follow.”
Cryptopolitan Academy: How to write a CV 3 that picks up interviews – Free cheat sheet
Related News :