Meta is selecting Amazon Web Services as its long-term strategic cloud provider to complement its on-premises infrastructure and round out its integration and PyTorch strategy.
The partnership has broadened the existing relationship between Meta and AWS over the last five years. Meta uses AWS compute, storage, database and security services to complement its own on-premises infrastructure. However, when Meta (then Facebook) acquired Instagram in 2012, the social media giant moved Instagram from AWS to in-house infrastructure. Multiple tuck-in acquisitions by Meta have also leveraged AWS.
At AWS' re:Invent conference, Meta and AWS outlined the following:
- Meta will use AWS to complement its on-premises infrastructure.
- The two companies will collaborate to help enterprises use PyTorch on AWS and scale deep learning models.
- Meta will run third-party collaborations in AWS. Meta also is using AWS to integrate better companies it acquires that run on the cloud giant.
- Meta will use AWS compute services for its AI research and development for its Meta AI unit.
- Meta and AWS will aim to improve the performance for customers running PyTorch on AWS.
The companies' overall mission is to accelerate how developers build, train, deploy and operate AI and machine learning models.
If successful, Meta and AWS could optimize PyTorch performance on Amazon Elastic Compute Cloud (EC2) and Amazon SageMaker. The companies aim to create native tools to improve PyTorch's performance, explainability and cost of inference. In addition, Meta and AWS will continue to enhance TorchServe, the serving engine native to PyTorch.
PyTorch is already attracting a lot of enterprise attention as companies try to scale AI and machine learning models.