According to OpenAI, the processor delivered nearly double the work per watt of its competitors while significantly reducing the time it takes to return data to a user.
These performance gains matter because they address the traditional technical trade-off between throughput and latency, the delay experienced during data processing.
By achieving higher efficiency and lower latency simultaneously, OpenAI says it can power more responsive digital agents and ensure more stable service even as global demand for AI grows.
This move signals a shift in the competitive AI infrastructure market, as OpenAI joins other major tech firms in developing internal hardware to potentially lower the costs of operating massive data centers.
OpenAI plans to begin deploying the new chips in limited quantities by the end of this year, with a larger production ramp-up scheduled for 2027.
The hardware is specifically optimized to run large-scale models, including GPT-OSS 120B and DeepSeek R1, more reliably.
As production increases, the company expects the specialized silicon to help it manage the heavy computing requirements of its expanding ecosystem of AI products and services.