This article is more than 1 year old
Meta won't migrate future acquisitions out of AWS
The Artist Formerly Known as Facebook and the cloud colossus become best cloudy buds – especially around PyTorch
Mark Zuckerberg's recently rebadged Meta and Amazon Web Services have announced they're going to be cloud BFFs, with the emphasis on the second F.
A joint announcement styles AWS as Meta's "Key, Long-Term Strategic Cloud Provider". Details of just what that means have not been offered, but a few specific initiatives were revealed.
One is the plan for Meta to "use the cloud to support acquisitions of companies that are already powered by AWS". That's notable, because when the firm (under its former nomenclature) acquired messaging service WhatsApp in 2014, it migrated the service from AWS to its own infrastructure. The new deal appears to mean Meta won't bother doing that again – perhaps because it's had a rotten time managing a MySQL migration?
The joint blurb also reveals that Meta already uses AWS "to complement its existing on-premises infrastructure". More of that is coming: Zuck's umbrella brand plans to "broaden its use of AWS compute, storage, databases, and security services to provide privacy, reliability, and scale in the cloud".
- UK competition regulator to Meta's Facebook: Sell Giphy, we will not approve the purchase
- 'We are not people to Mark Zuckerberg, we are the product' rages Ohio's Attorney General in Facebook lawsuit
- Facebook ditches its creepy, controversial robot – yes, its facial-recognition AI
The two companies are also both fired up about the PyTorch machine learning framework, which The Social Network™ spawned in 2016. The pair have promised to work together to optimise the tool so it works well with Amazon's EC2 and SageMaker services.
"To make it easier for developers to build large-scale deep learning models for natural language processing and computer vision, the companies are enabling PyTorch on AWS to orchestrate large-scale training jobs across a distributed system of AI accelerators," the announcement states, going on to promise "native tools to improve the performance, explainability, and cost of inference on PyTorch."
While that work will be open sourced, the plan is to drive PyTorch users to AWS.
Meta's veep of production engineering, Jason Kalich, said the special bond between them means "The global reach and reliability of AWS will help us continue to deliver innovative experiences for the billions of people around the world that use Meta products and services" – which reads like the sort of language it's appropriate to write off as marketing doggerel.
Or perhaps not: Meta's plan to build a metaverse – whatever that is – has been sketched out as requiring a lot of real-time multi-party video set in detailed virtual environments. That sort of thing is rather well suited to an elastic cloud that constantly upgrades its infrastructure with new instance types and GPUs, and perhaps less apt for a social network operator that essentially runs exceptionally large databases and analytics engines. ®