The Startup Helping OpenAI Optimize Its AI For Cerebras Chips
Not a pilot. OpenAI is now optimizing models across multiple chip vendors—and startups are getting paid to make it work.
Why it matters
As NVIDIA scarcity forces AI labs into multi-chip strategies, a new service layer is emerging to handle chip-specific optimization. This reveals how the AI infrastructure stack is fragmenting beyond GPU monoculture—and creating new business opportunities in the plumbing.
The key facts
6 to knowOpenAI hired Gimlet Labs to optimize models for Cerebras chips
OpenAI running Codex-Spark (faster coding tool version) on Cerebras hardware
Cerebras expected to IPO this week (May 2026)
Major AI labs (OpenAI, Meta) diversifying compute sources due to NVIDIA chip scarcity
Gimlet Labs CEO: Zain Asgar
New market emerging for chip-specific model optimization services
Go to the source
The Informationtheinformation.com
Publisher excerpt: As access to Nvidia AI chips become more elusive, major developers such OpenAI and Meta Platforms are diversifying their sources of computing power, including by striking deals to use a variety of AI server chips. But making that approach work is not a cinch. The AI labs need to tailor the code…