Amazon’s cloud division is gearing up for its biggest event of the year, when customers and developers descend on Las Vegas for re:Invent.

At the sixth annual re:Invent next week, Amazon Web Services executives are expected to announce new tools and share the stage with representatives from big companies who will tout their AWS usage.

AWS is the clear market leader in cloud infrastructure, but Google and Microsoft are growing faster and picking up marquee clients. To stay on top, AWS has to not only offer competitive prices and continue to expand geographically but also introduce new services built and operated by top-tier engineers.

Unlike developer events from Apple, Google and Microsoft, re:Invent is known to feature some surprises. But we already have a good idea of some of the things on the agenda. Here’s what you should expect at re:Invent 2017.

CNBC reported on Wednesday that AWS and health IT company Cerner are planning to talk about how they’re working together more closely, initially around a Cerner product called HealtheIntent. The health-care industry is a huge potential market that’s so far been slow to adopt the cloud.

Like Google and Microsoft, AWS sees artificial intelligence as a growth opportunity. Earlier this week AWS gave customers an AI tool for recognizing tech in images — technology that Microsoft and Google both introduced previously. Amazon has also been working on technology that translates text from one language to another, and the AWS division might well introduce a cloud-managed machine translation service for use by anyone.

AWS talks a lot about AI services and may be readying a general-purpose system for a trendy type of AI called deep learning, which involves training computers on lots of data, such as pictures, to make predictions. While Amazon’s preferred open-source library for deep learning, MXNet, will likely play a prominent role in the service, AWS also appears interested in making it easier to use Google’s TensorFlow on its cloud. Beyond that, AWS could discuss a data warehouse service for storing data that’s intended for neural network training.

People typically use Amazon’s computing power by packaging up their code inside virtual machines, many of which can run on a single physical server in an AWS data center. Recently, some developers have also started to use an alternative technology called containers, which can provide performance and efficiency benefits. Google has an open-source product called Kubernetes that’s become popular for managing the deployment of containers across multiple servers. Amazon is expected to provide better support for Kubernetes.

IBM and Oracle, which lag behind Amazon, Google and Microsoft in the public cloud infrastructure market, have sought to differentiate themselves by providing bare metal servers. They’re unconstrained by the software that enables virtual machines and can result in better performance for some computing tasks. We’ve heard rumblings that AWS has been working on a competing bare metal service.

Andy Jassy, the CEO of AWS, indicated earlier this week that Amazon’s technology chief Werner Vogels will be up to something a little different this year. Rather than announcing a string of new services during his keynote, Vogels will focus on data and serverless technology, or software that’s programmed to perform tasks when certain events happen. AWS got people talking about serverless technology in 2014, when it introduced Lambda.

CNBC be on site at re:Invent next week. Stay tuned for coverage.