Amazon Web Services (AWS) may have been caught off guard by the sudden rise of generative AI (genAI), but now it’s coming for the competition from every possible angle.

That was the message from the opening moments of CEO Matt Garman’s keynote at AWS re:Invent 2024, which reminded the audience of AWS’s breadth of services, with a security-centric pitch that set the stage for everything that came after. There was the usual blizzard of announcements, big and small (more on those below). The team of Forrester analysts onsite and beyond identified these key takeaways:

  • The second round of cloud AI competition has begun. As genAI early adopters contemplate scale-out strategies, AWS has a message for both long-standing customers and prospects reeling from VMware price increases: We can make genAI work with the data and juiced-up versions of services that customers already have. Amazon CEO and ex-AWS boss Andy Jassy took the keynote stage to announce Nova, a new set of models. For sheer power, AWS is building a new super cluster for AI training for its AI partner, Anthropic, the recent recipient of $4 billion in AWS investment, using AWS’s proprietary Trainium chips as a work-around for NVIDIA’s GPU dominance. Meanwhile, the Bedrock managed AI service will serve as a marketplace for AI models.
  • Mainstream AI service adoption will be abstracted and serverless. The keynote by Swami Sivasubramanian, vice president of AI and data at AWS, rolled out a series of closely intertwined enhancements to existing services like SageMaker and Kendra to tackle genAI challenges such as retrieval-augmented generation (RAG), a counter to Microsoft and Google’s top-to-bottom AI cloud solutions. AWS also pushed Amazon Q as an all-purpose generative AI assistant, with more third-party integrations, expanded development language support, and natural language automation for everything from data AI readiness to modernizing workflows.
  • AWS doubles down on data and storage for enterprise AI. AWS understands that its customers’ data has gravity — and wants to entice them to add more. The company showcased table buckets and queryable metadata updates to S3 that make it an ideal platform for data lakehouse architectures, especially with SageMaker for AI application development. Other updates include FSx for Lustre intelligent tiering and new storage-focused instances with high-speed Nitro SSDs for modern AI applications. Related announcements included federated data strategy with AWS Clean Rooms and a physical data transfer terminal service.

Here’s our take on key news from AWS re:Invent by category:

  • AI. AWS continues to augment its AI services across the full lifecycle of genAI. The new Nova foundation model series includes four for languages and two for computer vision. As Forrester predicted in its Predictions 2025 report for cloud computing, AWS announced RAG capabilities, including structured data retrieval, GraphRAG, and Kendra GenAI Index for enterprise data. This broad-spectrum approach includes multiagent collaboration, Bedrock model distillation, automated reasoning, intelligent prompt routing, and multimodal toxicity detection.
  • Data and analytics. AWS pushed the boundaries of data infrastructure with Aurora DSQL, a distributed and scalable SQL database. SageMaker got a boost with Unified Studio for integrated environments and HyperPod for orchestration and governance of model training, fine-tuning, and inferencing. Partners play a role via SageMaker third-party apps and Bedrock Marketplace.
  • Infrastructure. The common theme here is enablement for AI, HPC, and database workloads, with AWS Trainium2 and NVIDIA H200 GPU options and storage optimization in the spotlight. The announced P5en instances with NVIDIA H200 GPUs include third-generation Elastic Fabric Adapters to reduce latency. New storage services include optimizations for analytics and autotiered file storage plus support for Pure and NetApp storage, apparently aimed at VMware migration.
  • Application development. AWS continues to push its “well-architected” philosophy into its stack of cloud-native development and integration capabilities. Enterprises modernizing applications will be able to use services such as Step Functions and EventBridge to orchestrate workflows and connect resources across VPC and AWS account boundaries, easing integration of on-premises legacy apps.
  • Security. AWS initially focused on the security of the cloud, relying on partners to provide the security in the cloud. Today, the AWS security portfolio is much broader. The newly enhanced GuardDuty will help users walk through the MITRE ATT&CK chain, while various AI-oriented security announcements focused on data lineages. AWS made further accommodations for securing multicloud environments, too.
  • Sovereign cloud. AWS emphasized the launch of the European Sovereign Cloud, planned for Q4 2025 and backed by €7.8 billion in investment. This allows AWS to offer a single-provider multicloud environment in Europe. All cloud regions are powered by the secure Nitro hardware; pricing was not disclosed.
  • Cloud sustainability. Power usage effectiveness (PUE) value of AWS data centers has been decreasing, and the announced new data center design is aimed at bringing data centers’ PUE below the market current average. AWS expects the new data center design to translate into a 14% reduction in carbon intensity, a 46% reduction in mechanical energy used, and 35% less embodied carbon. From a silicon standpoint, Inferentia2 now delivers up to 50% better performance per watt than its previous generation while Trainium2 is three times more energy-efficient than Trainium1.
  • Cloud cost management/FinOps. AWS announced a slew of new capabilities, including adding genAI-enabled cost search function with Amazon Q Developer for chatbot-powered cost analysis, deeper anomaly detection with root-cause analysis, and a more accurate AWS Pricing Calculator that can ingest commitment purchases. AWS continues to lead the market in native cloud cost management, though competitor Microsoft Azure is close behind.
  • SAP deployment. AWS and SAP announced GROW with SAP on AWS, enabling rapid deployment of SAP’s ERP solution with AWS’s cloud benefits. This collaboration simplifies the adoption of SAP S/4HANA Cloud Public Edition and introduces new AI-assisted innovations. Customers will benefit from unified billing, existing AWS credits, and enhanced performance with AWS Nitro and Graviton ARM chips.
  • Network performance observability. A long-standing item on many network engineers’ wish lists is finally here: a holistic correlation of cloud networking performance and end-user experience. The CloudWatch Network Monitor solution will monitor network performance between AWS compute instances using flow monitor agents collecting TCP-based performance metrics, integrating CloudWatch Internet Monitor and Network Monitor. Network engineers should bring this data into their organization’s observability platform alongside network underlay and security observability data.