Former Intel CEO Pat Gelsinger has announced that his new startup, Gloo, is switching from OpenAI to DeepSeek's open source AI model R1 for its upcoming AI service Kallm.
The decision comes after DeepSeek's recent breakthrough in developing a high-performance reasoning model at a fraction of the typical cost. DeepSeek reportedly trained their model using 2,000 Nvidia H800 GPUs for approximately $5.5 million, achieving performance comparable to leading models that required billions in investment.
"My Gloo engineers are running R1 today," Gelsinger told TechCrunch, noting that within two weeks, the company plans to rebuild their Kallm service using entirely open source foundational models.
Gelsinger, who now chairs the church-focused messaging platform Gloo, sees DeepSeek's achievement as a reminder of key technology principles: affordable computing leads to wider adoption, constraints drive innovation, and open systems ultimately prevail.
While some industry observers have questioned DeepSeek's claims about costs and performance, particularly given its development in China, Gelsinger remains enthusiastic. "All evidence is that it's 10-50x cheaper in their training than [OpenAI's model]," he stated.
The former Intel executive envisions DeepSeek's cost-effective approach enabling AI integration across a broader range of devices. "I want better AI in my Oura Ring, in my hearing aid, in my phone, in my embedded devices," he explained.
Gelsinger's embrace of DeepSeek represents a notable shift in the AI landscape, where closed-source models from companies like OpenAI and Anthropic have dominated. "Having the Chinese remind us of the power of open ecosystems is maybe a touch embarrassing for our community," Gelsinger reflected on the Western AI industry's current state.