The new model of reasoning to the Open Source of Deepseek, R1, has sparked a Sell-off of the Nvidia actions and has made its consumption app rise to the top of the App Stores.
Last month Deepseek declared that he had formed a model using a Data Center of about 2,000 NVIDIA H800 GPUs in about two months at a cost of approximately $ 5.5 million. Last week, he published a document showing that the performance of his latest model corresponded to the most advanced reasoning models in the world. These models are trained on the data center that spend billions in chips to the fastest and most expensive Nvidia.
The reaction in the technological industry to the high performance of Deepseek, the low -cost model was wild. Pat Gelsinger, for example, led to X with Glee, publishing, “Thanks to the Deepseek team”.
Gelsinger is, of course, the recent former CEO of Intel, a hardware engineer and the current president of his startup linked to Ipo, Gloo, a messaging platform and involvement for churches. He left Intel in December after four years and an attempt to chase Nvidia with the GPUs to the alternatives of Intel, The Gaudi 3 Ai.
Gelsinger wrote that Deepseek should remind the technological industry of his three most important lessons: lower costs mean a more widespread adoption; Ingenuity thrives under constraints; And “Open wins. Deepseek will help to restore the increasingly closed world of the work model of ai fundamental, “he wrote. Openai and anthropic are both closed.
Gelsinger told Techcrunch that R1 is so impressive, Gloo has already decided not to adopt and pay for Openi. Gloo is building a service to the called Kallm, who will offer a chatbot and other services.
“My glue engineers are performing R1 today,” he said. “They could have performed O1 – well, they can only access O1, through bees.”
Instead, in two weeks, Gloo expects to have reconstructed Kallm from scratch “with our basic model that is all open source,” he said. “It’s exciting.”
He said he thinks Deepseek will make the IA so convenient, the IA will not only be everywhere. Well Ai will be everywhere. “I want to the best in my Orra ring. I want to the best in my acoustic device. I want more artificial intelligence in my phone. I want a better artificial intelligence in my incorporated devices, such as vocal recognition in my EV “, he says.
The happy reaction of Gelsinger was perhaps in contrast to the others who were less electrified by the fact that the basic models of the reasoning now have a more performing and much more convenient challenger. Artificial intelligence has become more expensive, no less.
Others reacted by implying that Deepseek must have confused his number in some way and training must have been more expensive. Some thought that it could not say that he used high -end chips due to the export restrictions of US chip to China. Others were hitting the holes in its performance, finding points where other models did better. Still others believe that the next model of Openai, O3, will thus be overcome R1 when it will be released that the status quo will be repaired.
Gelsinger shakes up everything. “You will never have full transparency, since most of the work was done in China,” he said. “But anyway, all the tests are that they are 10-50x cheaper in their training than O1.”
Deepseek shows that artificial intelligence can be advanced “by creativity engineering, not launching hardware power and calculates resources on the problem. So he’s electrifying, “he said.
As for this being a Chinese developer with everything he implies, like concerns for privacy and censorship, Gelsinger metaphorically shakes his head.
“Having the Chinese who remind us of the power of open ecosystems is perhaps an embarrassing touch for our community, for the western world,” he said.
Techcrunch has a newsletter focused on artificial intelligence! Sign up here to take it to your mailbox every Wednesday.
