Amazon Web Services is entering a multi-year partnership with AI chip startup Cerebras and will introduce Cerebras chips in its data centres, the Wall Street Journal reported on Thursday.

AWS plans to deploy Cerebras' Wafer-Scale Engine chips in its data centres alongside its Trainium chips to provide AI inference services. Inference is the process by which an AI model produces answers to user questions.

The company says Cerebras chips can run the decode stage, a key step in inference, up to 25 times faster than Nvidia GPUs. AWS and Cerebras plan to offer a fastest-speed inference service at premium prices through the partnership.

Cerebras drew attention in January after signing a contract worth more than $10 billion with OpenAI. OpenAI plans to build up to 750 megawatts of computing power using Cerebras chips. Cerebras raised an additional $1 billion in February, lifting its valuation to $23 billion.

Nvidia signed a $20 billion licensing deal with chip startup Groq in December and plans to unveil an inference-focused processing system using Groq technology next week. Cerebras Chief Executive Andrew Feldman (앤드루 펠드먼) said, "More people are using AI more often and for harder problems," adding, "We have put the Cerebras-Trainium solution on the world's largest cloud. It is an opportunity to meet many customers."