- Amazon Web Services announced the Inferentia chip that will be come available to customers late next year.
- Google and Alibaba have previously announced A.I. chips.
Amazon's cloud business is developing its own computer chips for artificial intelligence projects.
Amazon Web Services said on Wednesday at its AWS Re:Invent user conference in Las Vegas that its new Inferentia chips will provide A.I. researchers "high performance at low cost." It's the latest example of a giant provider of cloud services building next-generation processors.
Among providers of public cloud services, Amazon is following Google into the chip market. Google announced its first Tensor Processing Unit, or TPU, in 2016. Alibaba, a public cloud provider that's popular in China, has also announced an AI chip.
AWS is by far the leader in public cloud infrastructure, which companies can rely on to remotely run software and store data. Microsoft, Google, IBM are competing with AWS for business as companies move their workloads from traditional data centers to the cloud.
The Inferentia chips will become available in late 2019. Like with other AWS services, customers will be able to pay based on how much they use.
There are two common phases in AI — training models by feeding them lots of data, and then showing them new data that they can then use to run predictions. Since 2016 Google has introduced new TPU chips that compete with Nvidia for training AI models. Inferentia is focused only on inference for now.
Amazon said that some inference workloads require an entire graphics processing unit, which is expensive. "Solving this challenge at low cost requires a dedicated inference chip," the company said.
Earlier this week AWS announced ARM-based chips that represent an alternative to traditional computing processors from chipmakers like Intel. Those are more focused on low-cost, energy-efficient computing workloads. The new Inferentia silicon is specialized for AI.
AWS said customers will be able to use Inferentia with TensorFlow AI software (created by Google), as well as other AI frameworks like PyTorch and the ONNX format for converting models.