OpenAI said it will integrate AI hardware from Cerebras Systems into its computing infrastructure to speed up real-time responses from its artificial intelligence models, according to a company statement.
Cerebras develops purpose-built AI systems designed to accelerate long-form outputs by combining compute, memory and bandwidth on a single, wafer-scale chip, a design intended to eliminate data-transfer bottlenecks that can slow inference on conventional hardware.
See also: OpenAI Teams Up With SB Energy on U.S. Data Centre and Power Developments
OpenAI said the integration is aimed at reducing latency during inference, the stage when models generate responses to user requests. Faster response times are expected to support more interactive applications, including code generation, image creation and the operation of AI agents, while encouraging users to spend more time on higher-value workloads.
The company plans to roll out the Cerebras-based capacity in phases, gradually expanding its use across different workloads within its inference stack. The deployment is expected to take place in multiple tranches through 2028.
See also: OpenAI Introduces Healthcare-focused AI Suite With HIPAA Compliance Support
OpenAI said the move aligns with its strategy of building a diversified and resilient compute portfolio, matching specialized systems to specific workloads to improve performance and scalability. Cerebras said the partnership will place OpenAI’s models on its high-speed AI processors, positioning real-time inference as a key driver of future AI applications.
The announcement highlights a broader trend among AI developers to seek alternative and specialized hardware as demand grows for faster, more natural interactions with AI systems.
Source: OpenAI