AWS re:Invent 2025 Recap for Automotive and Manufacturing

AWS re:Invent 2025, our flagship annual conference, took place December 1–5, 2025, delivering five days of keynotes, breakout sessions, product announcements, and live demos, introducing several new services and features. This recap focuses on the most relevant highlights for automotive and manufacturing: the core announcements, real-world customer examples, and featured demos. Everything is organized by strategic workload areas so you can jump straight to the topics that map to your current projects and priorities.

Autonomous Mobility

Autonomous vehicle (AV) and advanced driver-assistance systems (ADAS) development are some of the more demanding workloads both in terms of compute performance and storage resources. Matt Garman, AWS CEO, announced in his keynote the general availability of AWS Trainium3 UltraServers, powered by AWS’s Trainium3 chips, and shared his outlook for the next generation Trainium4 chips. Trainium3 UltraServers deliver high performance for AI training and inference workloads, offering up to 4.4x more compute performance, 4x greater energy efficiency, and nearly 4x more memory bandwidth than Trainium2 UltraServers. They are optimized for next-generation agentic AI, reasoning models, and reinforcement learning relevant for training autonomous driving decision-making systems, developing AI agents that can reason through complex driving scenarios.

AV and ADAS workloads benefit from a tenfold increase in the maximum Amazon S3 object size, from 5 TB to 50 TB, enabling easier storage and analysis of massive sensor datasets such as LiDAR point-cloud embeddings and camera feature vectors. Amazon S3 Vectors is now available and scales to 2 billion vectors per index with up to a 90% cost reduction, supporting perception systems trained on petabytes of data.

AWS also introduced serverless GPU acceleration and auto-optimized vector indexes in Amazon OpenSearch Service, enabling large-scale vector databases to be built up to ten times faster at lower cost for real-time similarity search. In addition, AWS Clean Rooms privacy-enhancing synthetic data generation enables organizations to create synthetic training data for edge-case scenarios, while Amazon Nova 2 Omni (Preview) enables multimodal analysis and reasoning across text, images, video, and speech for perception and decision-support workflows.

In breakout session AMZ304, Zoox shared how it uses Amazon SageMaker HyperPod to train foundation models for autonomous robotaxis, running multimodal models that process camera, LiDAR, and radar data to handle complex edge cases. By combining Amazon SageMaker’s Hybrid Sharded Data Parallelism (HSDP) and tensor parallelism, Zoox achieves 95% GPU utilization across more than 64 GPUs, while ingesting up to 4 TB of vehicle data per hour via AWS Data Transfer Terminals at speeds of up to 400 Gbps. Zoox also demonstrated its autonomous robotaxi service, which officially launched in Las Vegas during re:Invent.

Software Defined Vehicles (SDV)

AWS released Kiro, an AI IDE that helps developers go from concept to production with spec-driven development in July 2025. AWS also unveiled three frontier agents, a new class of AI agents: Kiro autonomous agent, AWS Security Agent, and AWS DevOps Agent that work for hours or days as an extension of your software development team. Kiro autonomous agent could be used to support customers as a virtual developer in helping accelerate their needs in software-defined vehicle development.

In his keynote, Matt Garman also introduced Graviton5: the company’s most powerful and efficient CPU. The new Graviton5-based Amazon Elastic Compute Cloud (Amazon EC2) instances deliver up to 25% higher performance than the previous generation and offer a 5x larger cache.

In session IND382, Nissan Motor Co. shared how it is accelerating software-defined vehicle development on AWS through its new Nissan Scalable Open Software Platform, which delivers 75% faster testing and provides a unified cloud environment for more than 5,000 developers worldwide to collaborate on software creation, data management, and vehicle operations, enabling faster feature updates. In session CMP360, AWS also presented insights into the design and performance of Graviton5, sharing real-world workload results and guidance from customers such as Siemens, Synopsys, Honeycomb, Airbnb, and SAP on successfully migrating and running workloads on the Graviton platform.

Connected Mobility

All AWS customers benefit from elastic and reliable compute for their workloads, including automotive customers for their connected mobility backends. AWS launched new features for AWS Lambda (Lambda), Amazon Elastic Kubernetes Service (Amazon EKS), and Amazon EMR that will be relevant for connected mobility use cases.

AWS announced Lambda Managed Instances, a new capability that allows customers to run Lambda functions on their own Amazon EC2 while maintaining serverless operational simplicity. This enhancement addresses a key customer need: accessing specialized compute options and optimizing costs for steady-state workloads without sacrificing the serverless development experience. Lambda Durable Functions enable developers to build reliable multi-step applications and AI workflows that automatically checkpoint progress, suspend execution for up to one year during long-running tasks, and recover from failures. Amazon EMR Serverless now offers serverless storage that eliminates local storage provisioning for Apache Spark workloads, reducing data processing costs by up to 20% and preventing job failures from disk capacity constraints.

Two new capabilities for Amazon S3 Tables namely Intelligent-Tiering storage class that automatically optimizes storage costs when data access patterns change, and replication support to automatically maintain consistent Apache Iceberg table replicas across AWS Regions and accounts can enable managing connected vehicle data across geographies. AWS also announced the preview of AWS Interconnect – multicloud, a managed private connectivity service that enables AWS customers to create high-speed network connections between their AWS Virtual Private Clouds (VPCs) and their VPCs on other public clouds, facilitating multi-cloud architectures.

In session IND308, BMW showcased how they modernized their Connected Drive remote services backend by migrating from a monolithic Java EE application to an event-driven serverless solution using Amazon API Gateway, AWS Step Functions, AWS Lambda, Amazon Simple Notification Service (SNS), Amazon Simple Queue Service (SQS), and Amazon DynamoDB. The new solution gives BMW 60% faster time to market for new features, 20% AWS infrastructure cost reduction, and reduced infrastructure maintenance effort. The solution now handles over 16.6 billion requests per day, processing more than 184 terabytes of data and 100 million API calls with sub-second latency, supporting BMW’s 24.5 million connected vehicles.

Digital Customer Engagement

Digital customer engagement is driven by our customers’ need to engage their end users through seamless, personalized, and reliable interactions across voice, chat, and digital channels, while maintaining brand integrity, compliance, and operational governance. The announcements center on conversational AI models and production-grade agents.

The Amazon Nova 2 model family expands customer interaction options, including Amazon Nova 2 Sonic for speech-to-speech voice experiences, Amazon Nova 2 Lite for expanded reasoning with a million-token context window, and Amazon Nova 2 Omni (Preview) for multimodal inputs across text, images, video, and speech. For taking action in customer journeys, Amazon Nova Act helps customer developers build, deploy, and manage reliable UI workflow automation (e.g., forms, search/extract, booking, QA).

To build, deploy, and operate effective agents securely at scale, Amazon Bedrock AgentCore adds quality evaluations, policy controls, improved memory, and natural conversation capabilities for deploying agents across the enterprise, while Amazon Bedrock’s expanded catalog of 18 fully managed open-weight models offers more choices to customers for balancing quality, latency, and cost.

In IND320, Toyota Motor North America and Toyota Connected shared how they built an AWS agentic AI platform using Amazon Bedrock to deliver a RAG-driven dealer assistant that provides instant access to official vehicle information and supports over 7,000 interactions per month. Toyota’s platform is evolving into a next-generation system in 2026, adding AgentCore Runtime, AgentCore Identity, and AgentCore Memory to securely scale and enable actions such as local inventory checks.

In IND3329, Cox Automotive showed how it moved agentic AI from prototype to production in weeks, deploying five agentic AI products with Amazon Bedrock AgentCore (AgentCore) and Strands Agents. Its fully autonomous virtual assistant now handles after-hours sales and service conversations without human oversight, supported by strong guardrails, evaluation, and cost controls. In SPS313, Volkswagen Group explained how it scaled global marketing with custom GenAI, combining custom fine-tuned diffusion models trained on proprietary image libraries with Amazon Nova to automatically enforce brand guidelines across markets.

IND3326 and INV204 focused on digital engagement at scale, with Formula 1 using AWS Media Services and Agentic AI to deliver synchronized multi-view streaming and automate operational root-cause analysis, while Lyft transformed customer support with conversational agents that cut resolution times to minutes and resolved over half of interactions without human agents.

Manufacturing and Supply Chain

Generative artificial intelligence (GenAI) and specifically agentic AI are transforming manufacturing and supply chain management. In Matt Garman’s Keynote, AWS highlighted that modern AI agents with the ability to reason and act are taking on tasks that until recently required expert human judgment and manual execution. Amazon Bedrock AgentCore added quality evaluations, policy controls, enhanced memory, and natural conversation abilities for deploying trusted AI agents, enabling manufacturers to confidently scale AI solutions for predictive maintenance, quality control, and shopfloor optimization. In addition, with Edge Device support for Strands Agents, customers can use the Strands Agents SDK to create autonomous AI agents that can run on small-scale devices, unlocking new agentic use cases in automotive, manufacturing, and robotics.

In IND367, Audi presented how it uses AI-powered quality inspection models trained on AWS and integrated into production quality processes to inspect welding joints at a vastly higher level of coverage than manual sampling, effectively enabling near-100% weld inspection while significantly reducing manual effort and improving quality monitoring. In HMC217, Rivian described how it runs mission-critical factory applications at the edge using AWS Outposts Gen2, supporting workloads such as SCADA (Supervisory Control And Data Acquisition), MES (Manufacturing Execution System), and robotic control in a cloud-native hybrid environment while reducing operational overhead and simplifying capacity planning.

In PEX305, Toyota (with partners including IBM) showcased how they use AWS services such as Amazon SageMaker AI to build predictive models for delivery ETAs across vehicle manufacturing and logistics. In API206-S, Fujitsu showed how it uses Amazon Bedrock AgentCore to power agentic supply chain workflows, with a guardian agent that continuously monitors agent outputs and corrects agent drift.

Product Engineering

An automotive company’s product engineering team relies on fast cycles for concept design, generative optimization, simulation, and cross-site engineering collaboration. AWS announced the availability of new memory-optimized, high-frequency EC2 X8aedz instances, with 5 GHz processors and 3 TiB memory which can be used to help support memory-heavy engineering workloads such as simulation pre/post processing and large engineering datasets.

Amazon SageMaker HyperPod checkpointless and elastic training can be applied to large-scale training and iteration on engineering AI models. To manage large CAD, simulation, and test artifacts across global teams, Amazon FSx for NetApp ONTAP integration with Amazon S3 enables file-based engineering workflows while seamlessly tiering, sharing, and analyzing data at S3 scale.

In CMP340, Toyota showed how AWS Parallel Computing Service (PCS) reduced high performance computing (HPC) setup time from six weeks to just 30 minutes allowing researchers to spend more time focusing on their work. Researchers can launch large-scale CPU and GPU simulations on demand, run long-duration workloads, and automatically shut down resources when jobs finish, removing vendor delays.

Migration and Modernization

AWS Automotive and Manufacturing customers are using AI-powered services to accelerate application migration and modernization. AWS expanded AWS Transform with agentic capabilities to modernize Windows .NET applications, VMware environments, and mainframes, helping customers analyze over 1.1 billion lines of code and save more than 810,000 hours of manual effort. AWS Transform custom accelerates organization-wide code and application modernization across any code, API, framework, runtime, architecture, language, and even company-specific programming languages and frameworks. With pre-built and custom transformations, AWS Transform enables consistent, repeatable modernization across diverse codebases, while Amazon EKS Capabilities simplify workload orchestration and cloud resource management for modernized platforms.

In IND218-S, Mercedes-Benz shared how it used GenAI and agentic refactoring on AWS to modernize its global ordering system, a mainframe-based system, converting 1.3 million lines of COBOL to Java and achieving a successful zero-incident go-live in less than six months from first commit to production.

In INV212, BMW and AWS highlighted how domain-specific agents in AWS Transform accelerate discovery, planning, refactoring, and testing, showing how its migration factory (a team of experts enabled by the AI capabilities of AWS Transform) reduced test creation time from days to hours achieving over 75% time and efficiency savings and increased test coverage by up to 60%. BMW returned to the stage in session MAM205, to dive deeper into how agentic AI-powered refactoring has helped them de-risk their mainframe migration. In addition, session PEX203 explained how AWS Transform for VMware and .NET enables customers to migrate full-stack Windows applications to Linux-based architectures on EC2 and Aurora PostgreSQL, with Toyota Motor North America accelerating mainframe migration for supply chain applications into modern architectures by more than 50%.

Conclusion

This blog post summarizes AWS announcements that are relevant for AWS customers in the Automotive and Manufacturing industries along with innovative customer stories from BMW, Toyota, Rivian, Nissan, Mercedes-Benz, and Zoox. We encourage you to review these announcements to identify capabilities that could support your workloads. If you’d like to learn how these new capabilities can support your organization’s agility and efficiency, AWS is here to help.

Learn more about our offerings at the AWS for automotive page, or contact your AWS team today.