Amazon Web Services CEO Adam Selipsky on the battle for the hyperscale data center market
I spoke exclusively today with a WSCEO, Adam Solipsky. After Amazon earnings came in with cloud growth, arguably the star of the show, I asked about what drove that beat. It’s driven by a lot of customers you know getting back to migrating those workloads, getting out of their data centers and then of course also AI vectoring in on top of that. We at this point have a a multi billion dollar annualized AI business. It’s growing very quickly and that’s really being built on on on the the capabilities and the the rapid innovation we’re delivering to customers in all those areas. I spend some more time there on Bedrock, if you will and on how the the process that customers go through and choosing from the menu of OK, here are all the, here are all the sort of models that I can use maybe what I need for not only my industry but a specific piece of my business that I want to start experimenting with or even migrating to AI models on. It’s exactly as nuanced as what you just said and and that’s why there won’t be one model through the mall and and being able to evaluate and try and figure out, hey, which model do I want to use for this very specific piece of drug research or for this educational application. And so just last week, one of the several new capabilities we released in Amazon bedrock is the ability to do model evaluation. And so you can choose your criteria, do you care most about accuracy or about speed or about cost and many other characteristics. And you can basically upload sample data and then you can easily review results. And this is going to be really big in terms of helping customers actually understand and select which model to use. And then in addition, we also released the ability to to bring your own custom models to AWS to Amazon Bedrock. And so a lot of people are taking open source models, be they Mistral type models, meta models for meta and basically tweaking them, customizing them, making them their own. Now you can bring those to Amazon bedrock and inherit all of the security and all of the characteristics, all of the ability to complete multi step tasks and to to to to practice responsible and safe AI and to have those guardrails that people really want inside of Bedrock. I also asked him in a space where multiple AI models are emerging that customers want to use what’s going to make one cloud provider a better AI platform than another. I think there are a few things John, which are disproportionately important and that we’re really focused on, one of which is it’s ease of use. You know we’re really focused on on ensuring that AWS and that bedrock and QR it collectively are the easiest place to get started and to get moving quickly with generative. AII think second is a choice that we talked about a little bit and given that you are going to need to use multiple models, having them all instantly and easily available and having them really look and feel the same to you irrespective of which model you’re using is going to be really important. Third, I think security is a big differentiator. You know not some people think all the clouds are the same. I’m here to tell you it is not the case that AWS has a stronger track record of both security and operational excellence than a lot of competitors, some of which have really been in the news lately with perceived concerns around security. And that’s a big deal. And both privacy and security when it comes to AI is front and center top of mind for our customers. And it’s something that we put, you know, enormous energy into ensuring we’re truly best in class in security. And then finally availability, just making sure that we got enough capacity to service our customers demands. And we’re expanding really rapidly both in the US as well as worldwide to ensure we can be where customers need us to be. Yes, that was a little Microsoft shade there if you didn’t catch it. Finally, investment in infrastructure, I asked how he’s planning the billions of dollars of spend to meet cloud and AI demand while trying not to overspend. I think our approach is to try and be really, really close to customers and also understand Amazon’s internal needs and to make sure that we’ve got you know, multiple options for expanding that capacity really, really quickly. So for example, you know we if you look at supply chain, we’ve got both GPUs as well as the AWS design chips. It’s really important to our customers that there is that supply chain diversity and that there are multiple avenues to bring new capacity to market. So I think the fact that we’ve invested for many years in building these multiple slide supply chains enable us to react, you know, perhaps more quickly than a lot of other other folks in the market.