Pat Gelsinger, who recently stepped down as Intel’s CEO, has announced his startup Gloo is implementing DeepSeek’s AI technology instead of OpenAI solutions. This decision comes after DeepSeek’s latest R1 model demonstrated comparable capabilities to top-tier reasoning models at a fraction of the cost.
DeepSeek made waves in the tech sector by training their model using 2,000 Nvidia H800 GPUs in approximately two months, spending just $5.5 million. This achievement stands in stark contrast to competitors who invest billions in advanced AI chip infrastructure.
“My Gloo engineers are running R1 today,” Gelsinger stated, noting that his company’s AI service Kallm will be completely rebuilt using open-source foundation models within two weeks. This shift away from OpenAI’s closed-source solutions represents a significant vote of confidence in DeepSeek’s capabilities.
Gelsinger, who now chairs the messaging platform Gloo aimed at churches, emphasizes three critical insights from this development: reduced costs drive wider adoption, constraints fuel innovation, and open-source approaches prevail in AI development.
While some industry observers question DeepSeek’s claims about training costs and performance metrics, Gelsinger remains convinced of the technology’s potential. “All evidence suggests it’s 10-50x cheaper in their training than o1,” he noted, referring to competitive models.
The move signals a potential shift in the AI landscape, where cost-effective, open-source solutions might challenge the dominance of established players in the foundation model space.
