OpenAI’s $10 Billion Bet on Cerebras: The Race to Make AI Feel ‘Real-Time’

OpenAI just signed a $10 billion pact to outsource its AI compute needs to Cerebras—a move that could redefine how fast the world’s most high-profile AI models respond.

The agreement, spanning 2024 to 2028, allocates 750 megawatts of compute power to OpenAI. Sachin Katti of OpenAI explained the rationale:

"OpenAI’s compute strategy is to build a resilient portfolio that matches the right systems to the right workloads."

Cerebras CEO Andrew Feldman positioned the deal as a milestone in AI evolution:

"Just as broadband transformed the internet, real-time inference will transform AI."

Cerebras claims its AI-dedicated chip architecture outperforms GPU-based systems like those from Nvidia. The company is currently in talks to raise $1 billion at a $22 billion valuation. OpenAI CEO Sam Altman, already an investor in Cerebras, had previously explored acquiring the chipmaker.

Cerebras has postponed its 2024 IPO multiple times but continues securing large funding rounds. OpenAI’s stated focus on "real-time inference" contrasts with Cerebras’ unproven claims of speed advantages over established GPU systems.