AWS is introducing a large number of higher level services. Amazon Web Services now spans the entire spectrum from IaaS (Infrastructure as Service, e.g. EC2 VMs and VPCs) to SaaS (Software as a Service, e.g. hosted machine learning). The main themes of this year's conference were serverless computing and machine learning (Artificial Intelligence).
AWS has settled on three computing models: EC2 instances (VMs you provision and manage), Containers (Fully-managed Docker containers you run your micro services in an isolated manner) and Lambda (code you just upload to AWS and run; AWS does all the provisioning and scaling). Lambda now supports C#, in addition to node.js, Java, and Python. We have been successfully using Lambda (Java implementation) since the beginning of this year. Having C# as an option, makes Lambda more useful for us since we can now refactor and reuse our current C# code.
With respect to machine learning, AWS introduced Rekognition (hosted image recognition service), Polly (speech to text engine), and Lex (text to speech engine). This provides the entire stack for creating speech-based systems. With the success of Amazon Echo (Alexa), I think speech-based systems are going to be big in 2017.
It appears relational database systems are still the storage solution of choice for most companies. AWS Aurora (a MySql-like hosted database service) was extensively covered. It is apparently Amazon's fastest growing service. There wasn't as much talk of DynamoDb (NoSql) database this year.