President Donald Trump was barely a week when a new application of Chinese artificial intelligence (AI) called Deepseek shot the Silicon Valley.
Over the next day, Deepseek-R1 fired Apple graphics as the most downloaded free application in the United States.
The company said at the time that its new chatbot competed competed. Not only that. They said it had cost a simple fraction to develop.
These complaints – and the sudden increase in the popularity of the application – suffered $ 600 billion (446 billion pounds sterling) or 17% of the market value of the Nvidia flea giant, marking the largest loss of a day for a single stock in the history of the American stock market.
Several other technological actions with exposure to AI have also been taken during the descending current.
Deepseek also questioned the American domination of AI. Until then, China had been considered to have fallen behind the United States. Now it seemed that China had catapulted itself in the foreground.
The venture capital Marc Andreessen described the arrival of Deepseek-R1 as a “sputnik moment of AI”, a reference to the Soviet satellite which had launched the space race between the United States and the USSR more than half a century earlier.
Always relevant
Deepseek has been amazing the world for six months now.
Today, the revolutionary application of China has largely abandoned the headlines. This is no longer the burning subject of the Happy Hour here in San Francisco. But Deepseek has not disappeared.
Deepseek challenged certain key hypotheses on the AI that had been defended by American leaders like Sam Altman, CEO of Chatgpt-Maker Openai.
“We were on a path where greater was considered better,” according to Sid Sheth, CEO of the startup of Puce Ai D-Matrix.
Perhaps the maximum of data centers, servers, fleas and electricity to execute it was not the way to follow after all.
Although Deepseek does not have access to the most powerful technology available at the time, Sheth told the BBC that it showed that “with smarter engineering, you can actually build a capable model”.
The interest in Deepseek settled in a weekend at the end of January, before business staff could move to prevent employees from flocking to it.
When the organizations took the following Monday, many rushed to prohibit workers from using the application as an concern to know if user data was potentially shared with the People’s Republic of China, where Deepseek is based.
But although the exact figures are not available, many Americans still use Deepseek today.
Some Silicon Valley start-ups have chosen to stick to Deepseek instead of more expensive AI models of American companies in order to reduce costs.
An investor told me for companies short of money, the funds saved by continuing to use Deepseek help to pay critical needs such as additional workforce.
However, they are cautious.
In online forums, users explain how Run DEEPSEEK-R1 on their own devices Rather than online use of Deepseek servers in China – a bypass solution which, according to them, can protect their data against sharing surreptitiously.
“It is a good way to use the model without worrying about what it exfiltrates” for China, said Christopher Caen, CEO of Mill Pond Research.
American-Chinese rivalry
The arrival of Deepseek also marked a turning point in the American-Chinese AI rivalry, according to some experts.
“China has been considered a catch -up in models of great language to this point, with competitive models but still dragging the best Westerners,” the BBC analyst Wendy Chang told Mercator Institute for China.
A large language model (LLM) is a reasoning system formed to predict the following word in a given sentence or sentence.
Deepseek changed perceptions when he claimed to have obtained a leading model for a fraction of common calculation resources and costs among its American counterparts.
Openai had spent $ 5 billion (3.7 billion pounds sterling) only in 2024 only. On the other hand, the deep researchers said they had developed Deepseek -R1 – which came in addition to the O1 model of Openai through several landmarks – for only $ 5.6 million (4.2 million pounds sterling).
“Deepseek revealed the competitiveness of the Chinese AI landscape in the world,” said Chang.
The American developers of AI managed to capitalize on this change.
AI transactions and other announcements deceived by the Trump administration and large American technological companies are often considered essential to stay ahead of China.
The TSAR of Trump AI, David Sacks, noted that technology would have “deep ramifications for economics and national security” when the administration unveiled its AI action plan last month.
“It is simply very important that America continues to be the dominant power of AI,” said Sacks.
Deepseek has never managed to suppress concerns about the security implications of its Chinese origins.
The US government has evaluated the company’s bonds in Beijing, as indicated for the first time by Reuters in June.
A senior US State Department official told the BBC that he understood that “Deepseek had readily provided and would probably continue to provide, support for military and intelligence operations of China”.
Deepseek did not respond to the request for comments from the BBC, but the confidentiality policy of the company stipulates that its servers are located in the People’s Republic of China.
“When you access our services, your personal data can be processed and stored in our servers in the People’s Republic of China”, politics said. “This can be a direct provision of your personal data for us or a transfer that we or a third party make.”
A new approach?
Earlier this week, Openai rekindled Talk on Deepseek after publishing a pair of AI models.
These are the first free and open versions – which means that they can be downloaded and modified – published by the American IA Giant in five years, well before Chatgpt was inaugurated in the era of consumers.
“You can draw a straight line from Deepseek to what Optai announced this week,” said Sheth de D-Matrix.
“Deepseek has proven that smaller and more effective models could always offer impressive performance – and that has changed the state of mind of industry,” Sheth told the BBC. “What we see now is the next wave of this thought: a change to right -sized models that are faster, cheaper and ready to deploy on a large scale.”
But for others, for the main American players of AI, the old approach seems alive.
Only a few days after the release of free models, Openai unveiled GPT-5. From the perspective, the company said It has considerably increased its IT capacity and IA infrastructure.
A multitude of ads About the new data center The clusters necessary for the AI have come while American technological companies compete for the talents of high level AI.
The meta-PDG Mark Zuckerberg has plowed billions of dollars to make its AI ambitions, and I tried to attract the staff of its competitors with compensation packages of $ 100 million.
The fortune of the technology giants seemed more attached than ever to their commitment to the expenses of the AI, as evidenced by the series of results of the eruption revealed this last season of technological gains.
Meanwhile, the actions of Nvidia, which plunged just after the arrival of Deepseek, rebounded – affecting new heights which made it the most precious business in the world of history.
“The initial story has proven to be a little red herring,” said Caen de Mill Pond Research.
We are back to a future in which AI will consider ostensibly for more data centers, more chips and more power.
In other words, the upheaval of Deepseek of the status quo has not lasted.
And what about Deepseek himself?
“Deepseek is now confronted with challenges to support her momentum,” said Marina Zhang, associate professor at the University of Technology in Sydney.
This is due in part to operational setbacks but also in intense competition from businesses in the United States and China, she said.
Zhang notes that the next product of the company, Deepseek-R2, would have been delayed. A reason? A shortage of high -end fleas.
