#734: AWS News: OpenAI, Amazon Elastic VMware Service, and Lots More.

0:00This is episode 734 of the AWS podcast released on August 25th, 2025. Welcome to the AWS Podcast. This is Jill here, and I am with the one and only Shruti. Shruti, how are you doing? Good, good. It's a wonderful afternoon here in Southern California, and I'm excited about the updates we have this week. Me too. No, Simon. Oh, I always miss him. Every time he's not here, it just feels like there's something missing in the show. But I will do our best with you every time because we've got a lot of really big updates. So let's get right into it. The first one is Amazon Application Recovery Controller now supports Region Switch.

0:51which I want to call this out. One, I'm a little bit biased because I did just do an episode on this that is going to be coming out. So, of course, a plug for that one. I also like this one because it's not a matter of are you going to adopt a multi-region strategy, whether that's for just overall high availability or disaster recovery. It's really a matter of when. And this just makes it simple. So this is an automated feature. and it's going to allow you to orchestrate the specific steps to switch operating your multi-region application out of another AWS region. And a lot of people have a lot of manual processes to do this and now making it automated, simple, easy, all the good things.

1:39Definitely like it. It's available in all commercial AWS regions, which makes it even more fun. So super excited for you all to start utilizing that. And of course, look out for the episode. that I'm going to be doing with the folks on the team who have been behind the scenes, the masterminds of it. So that was the first pick. And the second one is the OpenAI OpenWay models are now in Amazon Bedrock and in Amazon SageMaker Jumpstar. Shruti, so what stands out to you about this? Yeah, I mean, you know, first of all, it's that OpenAI models are now available on AWS. And we know that a lot of customers have asked us about it.

2:27And one of the primary value propositions for Amazon Bedrock is the choice it provides in terms of the, not only in terms of the available models, but also all the other tooling that goes around it. And OpenAI models have been, you know, much sought by a lot of our customers. So it's really exciting that these are now available. Now, these models, as you mentioned, are open weight models, the GPT-OSS-120 billion parameters and the GPT-OSS-20 billion parameters, which will, you know, again, allow customers to either sort of use them as they are via Amazon Bedrock or, you know, get started via SageMaker Jumpstart and, you know, build on top of them.

3:16Like maybe there'll be some fine tuning involved, so on and so forth. Maybe there'll be some other forms of customization. So the other thing that stands out, of course, about these models in particular is that they both feature 128K context window. and adjustable reasoning levels to help match specific requirements. So based on what type of application or what type of agentic workflow you are building, you can choose the reasoning level on these models to match your use cases. So, yeah, no, this is a really exciting launch for a lot of customers. It really is. And let's get on to the other exciting launches that we've got and announcements that we've got going on.

4:04So we're going to start with analytics. Amazon CloudWatch launches a natural language query generation, powered by, of course, generative AI, for OpenSearch, PPL, and SQL query languages in CloudWatch Logs Insights. CloudWatch Logs Insights, if you are using it, this enables you to interactively search and analyze your logs with Logs Insights with the query language, OpenSearch Service pipe processing language, and OpenSearch Service structured query language. Another one from OpenSearch. OpenSearch serverless adds support for hybrid search, AI connectors, and automations. This new set of APIs facilitates use cases such as RAG and semantic search.

4:53Neural search enables semantic queries through text and images instead of vectors. Neural Search uses a high-level API with connectors to Amazon SageMaker, Amazon Bedrock, and other AI services to generate enrichments like dense or sparse vectors during query and ingestion. Hybrid Search enables combining lexical, neural, and KNN vector queries to deliver higher search relevancy. Amazon OpenSearch serverless introduces automatic semantic enrichment. You can now boost your search relevance with minimal effort, eliminating complex manual configurations through an automated setup process. Semantic search goes beyond keyword matching by understanding the context and meaning of search queries.

5:44So an example of this is let's say you're searching for how to treat a headache. Semantic search is going to intelligently return relevant results about migraine remedies or maybe pain management techniques, even when these exact terms aren't present in the query. And previously, implementing semantic search, this required machine learning expertise or maybe hosting your own model. But now this is automatic, so this is going to make it much simpler. Another one from Amazon OpenSearch Serverless is they now support backup and restored. So system will automatically backup all collections, indexes in the account automatically every hour.

6:29And the backups are retained for 14 days and available to restore indexes using APIs. OpenSearch Serverless now supports KNN byte vector and new data types. So these are also going to map parameters such as strict allow templates, wildcard field types, and Kuroemoji completion analyzer. I don't know what that was. I thought it was an emoji that they were adding as part of the feature. I got really excited. But if anyone knows what that is, of course, let us know. We do have a big button on our website. Amazon QuickSight now supports connectivity to Apache Impala. Apache Impala is a massively parallel processing SQL query engine that runs natively on Apache Hadoop.

7:13QuickSight customers can now connect using their username, password credentials for Impala and import their data into Spice. Amazon OpenSearch service now supports fine-grained access control for OpenSearch UI when accessed through SAML via IAM Federation. Next topic is application integration. Amazon SQS increases the maximum message payload size to 1MIB, enabling customers to send and receive larger messages through their Amazon SQS standard and FIFO queues. Amazon SageMaker Lakehouse now automates optimization of Apache Iceberg tables stored in Amazon S3 with catalog level configuration, reducing metadata overhead and improving query performance.

8:05Now you can enable automatic optimization for new iceberg tables with a one-time data catalog configuration. Once enabled, for any new table or updated table, data catalog continuously optimizes tables by compacting small files, removing snapshots, and unreferenced files that are no longer needed, resulting in controlled storage costs and faster queries. Okay, next up we have some exciting updates under artificial intelligence. Amazon QBusiness launches agentic RAG or retrieval augmented generation for QBusiness applications. This new feature enhances the ability of QBusiness to provide more accurate and explainable responses to complex multi-step RAG queries.

8:59Amazon QBusiness also launches response events for enhanced chat transparency. So this new feature will provide real-time visibility into the assistance query processing steps. This capability will allow users to observe how their queries are processed and create transparency and trust in their interactions with QBusiness. Amazon SageMaker AI now supports P6E GB200 Ultra Servers in SageMaker HyperPod and training jobs. This is really exciting because the P6E GB200 Ultra Servers allow you to use 72 NVIDIA Blackwell GPUs under one NVLink domain. and it allows you to accelerate deployment and training of foundational models at the trillion parameter scale.

9:55So now that the support for these ultra servers is available in HyperPOD and training jobs, it means that you can use these powerful compute resources without having to manage the underlying infrastructure and you can benefit from HyperPOD's built-in features such as the security, the fault tolerance, the topology of air scheduling, all the monitoring capabilities, so on and so forth. So very exciting for customers who are either building their own models via pre-training or fine-tuning, especially those building these really, really large models. Amazon SageMaker HyperPod now provides a new cluster setup experience.

10:46This new experience sets up all the resources needed for large-scale AI workloads, including networking, storage, compute, IAM permissions, in just a few clicks. It introduces both quick and custom setup paths that make it easy for both beginners as well as advanced AWS customers to get started. Amazon SageMaker HyperPod now supports continuous provisioning for enhanced cluster operations. AI ML customers need to start training quickly, scale seamlessly, and perform maintenance without disrupting operations, and have granular visibility into cluster operations. They also require the ability to efficiently manage dynamic inference workloads where capacity needs change quite frequently.

11:41So with this new continuous provisioning, SageMaker HyperPod automatically provisions remaining capacity in the background while training jobs can begin immediately on available instances. Hyperboard will retry in the background when it encounters node provisioning failures and ensure that clusters reliably reach their desired scale without requiring any manual intervention. So for any of you who have tried to run training or inference jobs across a large cluster, you know how much work goes into provisioning these clusters and maintaining them and, you know, accounting for the faults or failures.

12:29And this new capability just kind of allows you to get started with the number of nodes you have, while HyperPort in the background is sort of trying to provision additional capacity when available. Amazon SageMaker HyperPort now supports custom AMIs or Amazon machine images. Customers deploying AI workloads on hyperpods sometimes need customized environments that meet strict security, compliance, and operational requirements while maintaining fast cluster startup times. And they can often struggle with complex lifecycle configuration scripts that slow deployment and create inconsistencies across cluster nodes.

13:15So with these custom armies, customers can benefit from all of HyperPod's capabilities while incorporating their customized security agents, compliance tool, proprietary libraries, as well as any specialized drivers. Amazon SageMaker Studio now supports trusted identity propagation. This enables admins to trace actions that are taken in SageMaker Studio back to a human user. It also enables administrators to manage permissions based on user identity to lake formation and S3 access grants. Anthropics Cloud Opus 4.1 is now available in Amazon Bedrock. Cloud Opus 4.1 is Anthropics' most intelligent model to date and an industry leader for coding and agents.

14:14And Opus 4.1 is a drop-in replacement for Opus 4 and delivers superior performance and precision for real-world coding as well as agentic tasks. Also available in Amazon Bedrock with the expanded context window is Anthropics Cloud Sonnet 4. The context window has been increased from 200 ,000 to 1 million tokens, representing a 5X expansion. This enhancement will allow Claude to process and reason over much larger amounts of text in one single request, opening up new possibilities for comprehensive analysis and generation tasks. Automated reasoning checks, a safeguard within Amazon bedrock guardrails, is now generally available.

15:08This safeguard uses formal verification techniques to validate the accuracy and policy compliance of outputs from the generative AI models. Automated reasoning checks deliver up to 99 % accuracy at detecting correct responses from LLMs, giving you provable assurance in detecting AI hallucinations while also assisting with ambiguity detection in model responses. So this automated reasoning checks, it provides a very different approach from traditional testing methods. Unlike sampling outputs for quality, automated reasoning checks offers mathematically rigorous guarantees that AI responses adhere to defined business rules and domain knowledge.

15:58This is especially valuable for enterprises and regulated industries that require unambiguous validation of AI outputs before deployment. This is a really, really cool feature, and I hope that everyone who is using Amazon Bedrock checks this one out. SageMaker HyperPod now supports topology-aware scheduling of large-language model tasks. This enables scientists, data scientists, applied scientists, to schedule their LLM tasks on an optimal network topology that minimizes network communication and enhances training efficiency. As a lot of you know, the training and fine-tuning tasks are often distributed across multiple accelerated computing instances that are connected through EFA networking, and they exchange large volumes of data between them.

16:57Multiple network hops between instances can result in a higher communication latency, which ultimately impacts your performance. And so now SageMaker HyperPort Task Governance allows data scientists to use the network topology information, which basically means like truly understanding what instance is connected to what other through how many hops on and so forth. It gives you that topology information and data scientists can use it to schedule specific tasks with those specific topology preferences. And that can help you reduce instance-to-instance communication and overall increase your training or fine-tuning efficiency.

17:42All right. And one last update in AI is SageMaker HyperPod now supports a fine-grained quota allocation of compute resources. So what this does is allows fine-grained compute quota allocation of GPU, Tranium Excerator, vCPU, and vCPU memory within an instance. Administrators can allocate these fine-grained compute quota across teams, optimizing compute resource distribution and staying within the budget. Next up is business applications. Amazon Connect Outbound Campaigns now supports account-based campaigns, allowing you to reach multiple people associated with the same account. This seems super handy, so I'll give you an example.

18:32Let's say there's someone who's calling a joint bank account. Let's say that per first person, they're unavailable. The system is then going to automatically try to reach other authorized members of this account. You can also define a prioritized contact sequence across multiple phone numbers. For example, maybe the mobile first and then home and then work. If the first person is unreachable, Connect will automatically try the next number in the sequence. Super useful. Amazon Connect now provides a new API that returns real-time position in the queue, enabling businesses to better estimate the wait time.

19:12Amazon SES announces the ability to provision isolated tenants within a single SES account and apply automated reputation policies to manage email sending. With this enhancement, customers can create multiple tenants in their SES account, each with dedicated configuration sets, identities, and templates. Next up, we've got Compute. AWS Batch now supports AWS Graviton-based spot compute with AWS Fargate. Definitely check this out if you are using Batch and are looking for ways to be able to cost optimize your overall AWS usage. This capability helps you run fault tolerant ARM-based applications with up to 70.

19:57Yes, that's right. That's 7-0 percent discount compared to Fargate prices. Definitely check it out. AWS Deadline Cloud introduces new cost-saving compute option, and this can reduce rendering costs with prices starting as low as 0.006 per vCPU hour. AWS Deadline Cloud now supports Autodesk vRed, as well as using the Deadline Cloud client. If you're not familiar with Deadline Cloud, This is a fully managed service that simplifies render management for teams, creating computer-generated graphics and visual effects for films, television and broadcasting, web content and design. AWS Elastic Beanstalk now supports VPC endpoints that have been validated under the FIPS 140-3 program.

20:54AWS Lambda now supports GitHub Actions to simplify function deployment. AWS Outpost Racks now support new Amazon CloudWatch metrics. AWS Outpost Servers now support ServiceLink static configuration. This new feature enables customers to configure the ServiceLink interface and DNS IP addresses of their Outpost servers with static IP addresses during installation, eliminating the requirement for dynamic host configuration protocol servers in their data centers. AWS Parallel Computing Service now supports IPv6. AWS Parallel Computing Service now supports Slurm Spank plugins. AWS announces the general availability of Amazon Elastic VMware Service.

21:42This allows you to run VMware Cloud Foundation directly within your VPC. With Amazon EVS, you can leverage the scale, elasticity, and performance of AWS while maintaining your familiar VCF, also known as VMware Cloud Foundation software, and existing skills, eliminating the need to replatform or refactor applications during your migration. Amazon EVS offers you choice, control, and flexibility in managing your VMware environments. Amazon EC2 single GPU P5 instances are now generally available. I know a lot of you have been waiting for this one because you see the NVIDIA H100 and you're like, oh, wow, if only I could, I just need one.

22:30I don't need all of them. And this one is for you. Definitely going to make it really cost effective for those that just need one GPU. So this new instance size enables customers to start small and scale in granular increments, providing more flexible control over their infrastructure costs. and this is the p5.4x large instance by the way if that's what you are looking for you can get them through amazon ec2 capacity blocks for machine learning amazon ec2 now supports force terminate for ec2 instances ec2 instances that get stuck in the shutting down state because of rare issues caused by frozen operating system or underlying hardware problems so when this happens customers use forced terminate and the instance will first attempt a graceful shutdown process and then if it's unsuccessful within the timeout period the instance proceeds with a forced shutdown so forced terminate this is available in the ec2 console or the abys cli amazon ecr now supports 100 000 images per repository this is up from the previous limit of 20 000 wow that's a lot Amazon ECS console now natively integrates with Amazon CloudWatch logs LiveTail, enabling real-time logs streaming directly within the ECS console.

23:59Amazon EKS now supports deletion protection, helping you prevent accidental termination of your EKS clusters. One enabled deletion protection requires explicit disablement before a cluster can be deleted, providing an additional safety control for critical environments. This is one where someone is like, oh, thank you. And we're just like, you're welcome. Amazon EKS expands support for Cilium as the container networking interface for Amazon EKS hybrid nodes. Cilium is a cloud-native computing foundation graduated project that provides core networking capabilities for Kubernetes workloads. Now you can receive support from AWS for a broader set of Cilium features when using Cilium with EKS hybrid nodes, including application ingress in cluster load balancing, Kubernetes network policies, and kube proxy replacement mode.

24:54We're announcing multiple enhancements to Amazon EC2 on-demand capacity reservations in cluster placement groups. So customers can now add on-demand capacity reservations belong to different cluster placement groups to resource groups, which will enable customers to manage and target groups of reservations spread across multiple placement groups. Customers can also share cluster placement group on-demand capacity reservations across multiple AWS accounts through AWS Resource Access Manager. Next up, we have databases. Amazon Aurora Serverless v2 now offers up to 30 % improved performance for databases running on the latest serverless platform version 3.

25:45Aurora Serverless V2 measures capacity in ACUs, where each ACU is a combination of approximately 2 gigabytes of memory, corresponding CPU, and networking. You specify the capacity range and the database scales within this range to support your application's needs. The version 3 serverless platform version supports scaling from 0 up to 256 Aurora capacity units. With improved performance, you can now use Aurora serverless for even more demanding workloads. Amazon DocumentDB with MongoDB compatibility offers extended support for version 3.6, allowing customers to maintain critical workloads on version 3.6 for up to three years beyond the standard support end date on March 30, 2026.

26:45This is designed for customers who may need more time to plan and implement version upgrades, especially when navigating application dependencies or managing enterprise-scale deployments. Amazon DB is announcing the support of Console2Code, powered by Amazon Q developer. Console2Code makes it simple, fast, and cost-effective to create DynamoDB resources at scale by getting you started with your automation code. Amazon DynamoDB now supports more frequent throughput mode updates from provisioned to on-demand capacity. This enhancement makes it simpler for customers who have use cases which require loading large volumes of data into their DynamoDB tables multiple times per day, or customers who want greater flexibility to manage their workload requirements and optimize costs.

27:40Amazon RDS or Relational Database Service for MySQL now supports MySQL's miner versions 8.0.43 and 8.4.6, the latest miners released by the MySQL community. Amazon RDS for Oracle now supports the spatial patch bundle for the July 2025 release update for Oracle Database version 19c. Amazon RDS for Oracle now supports the July 2025 release update for Oracle Database versions 19c and 21c. Amazon RDS for SQL Server now supports cumulative update CU20 for SQL Server 2022, RDS version 16.00.4205 .1.v1, and general distribution release for SQL Server 2016 SP3, RDS version 13.00.6460.7.v1, as well as the SQL Server 2017, RDS version 14.00.3495.9.v1, and SQL Server 2019, which is RDS version 15 .00.4435.7.v1.

29:16Those were a lot of new versions. In short, RDS now supports cumulative updates CU20 for Microsoft SQL Server 2022 and general distribution releases for Microsoft SQL Server 2016, 2017, and 2019. For those who want the detailed version information, the best way probably is to just go read the update on what's new posts. AWS Advanced Go Driver is now generally available for use with Amazon RDS and Amazon Aurora Postgres and MySQL compatible database clusters. This database driver provides support for faster switchover and failover times, federated authentication, and authentication with AWS Secrets Manager or AWS Identity and Access Management.

30:10Amazon CloudWatch Database Insights expands the availability of its on-demand analysis experience to the RDS for Oracle database engine. This feature leverages machine learning models to help identify any performance bottlenecks during the selected time period and gives advice on what to do next.

30:50Next up, developer tools. Amazon Location Service now supports multi-polygons and exclusion zones, simplifying creation and geofencing of complex boundaries. This allows customers to create geofences for non-contiguous areas. One example could be, let's say you want to define California's boundaries and include offshore territories like the Catalina Islands. These enhanced geofences are fully integrated with existing workflows and can be accessible through the console, programmatically through the API, and SDK. Next up is end-user computing. We have accelerated Amazon Workspace's deployment with streamlined Bring Your Own License process.

31:41So with this streamlined approach, customers can enable the Bring Your Own License feature in their AWS account without contacting AWS support. Amazon Workspaces has improved the Bring Your Own License process, offering customers a more efficient and faster way to import their Windows licenses to use with Workspaces. One update in gaming. Amazon GameLift Streams announces the launch of a new runtime environment, Proton 9. along with increased default service limits for all customers. Proton 9 is a significant update to the Proton compatibility layer, which allows Windows games and applications to run on Linux systems.

32:25This enhanced runtime environment provides game developers with more flexibility and options for deploying their games on Amazon GameLift streams, potentially expanding their reach to a broader audience across different platforms. Next up, we have a couple of updates under Internet of Things. AWS IoT Core now offers the Delete Connection API, enabling programmatic disconnection of MQTT clients using their client IDs. This capability allows developers to terminate MQTT connections with options to clear persistent sessions and suppress publication of last will and testament messages, which is messages that the MQTT broker automatically publishes on a client's behalf when it disconnects unexpectedly.

33:19AWS IoT SiteWise introduces asset model interfaces. Now, SiteWise is a managed service that simplifies the collection, organization, and monitoring of industrial equipment data at scale. With this asset model interfaces in IoT SiteWise, industrial customers can define and maintain standardized properties and metrics across similar equipment and process types while maintaining the flexibility to accommodate equipment variations. We have some updates under management and governance. AWS announces support for BillingView in AWS Budgets, enabling organizations to create budgets that span multiple member accounts without requiring access to the management account.

34:10AWS Config now supports 10 additional AWS resource types. This expansion provides greater coverage over your AWS environment, enabling you to more effectively discover, assess, audit, and remediate an even broader range of resources. AWS customers can now view and manage their support cases from the AWS Console mobile app. Customers can view and reply to case correspondence and resolve, reopen, or create support cases while on the go and away from their workstations. AWS Resource Explorer introduces support for filtering on multiple values in both the search and list resources APIs. This will allow customers to build targeted queries to find resources more easily.