Odaily News DeepSeek's new open source AI inference model R1 triggered a sell-off in Nvidia's stock and propelled its consumer app to the top of Apple's App Store charts.
Last month, DeepSeek said it trained a model in just two months at a cost of about $5.5 million using a data center with about 2,000 Nvidia H800 GPUs. Last week, DeepSeek published a paper showing that the performance of its latest model matches the world’s most advanced inference models. These models are being trained in data centers that spend billions of dollars on Nvidia’s faster, very expensive AI chips.
The entire technology industry has reacted strongly to DeepSeek's high-performance, low-cost model. For example, former Intel CEO Pat Gelsinger posted on X: "Thank you DeepSeek team."
Gelsinger, who recently stepped down as Intel CEO and is currently chairman of his own IPO startup, Gloo, spent four years at Intel and left in December last year after trying to catch up with Nvidia with Intel's alternative AI GPU, the Gaudi 3 AI.
He wrote that DeepSeek should serve as a reminder of the tech industry’s three most important lessons: Lower costs mean wider adoption; creativity thrives under constraints; and open source wins. DeepSeek will help reset the increasingly closed world of work on foundational AI models.”
Gelsinger also revealed that the R1 was so impressive that Gloo has decided not to adopt and pay OpenAI anymore. Gloo is building an AI service called Kallm, which will provide chatbots and other services. (TechCrunch)