← AWS Podcast
AWS Podcast
#723: AWS News: Claude 4 Opus comes to Bedrock, AWS Transform accelerates modernization, and more
00:39:45
Claude 4 Opus and Sonnet, the most powerful models from Anthropic for coding and advanced reasoning,
Loading summary
Transcript30 lines
- [00:00]
A
This is episode 723 of the AWS podcast released on June 2, 2025.
- [00:09]
B
Welcome everyone to the AWS podcast. As you can tell, I am not Simon. Our fearless leader actually takes vacation. So today you get me and the one and only Shruti. Shruti. How's it going?
- [00:23]
A
Great. It's great to be here and yeah, I'm excited for some of the updates we have here.
- [00:30]
B
Me too. We've got a lot with the first one being a really big one that Claude 4 is in Amazon bedrock. This is Anthropic's newest AI model, Claude Opus 4 Claude Sonnet 4. And if you're not familiar with Amazon Bedrock by now, it is a fully managed serverless service that gives you access to the latest in advanced reasoning and agentic and capabilities. There's a lot of really interesting capabilities of Opus 4 and Sonnet 4. Shruti, what comes to mind for you?
- [01:04]
A
Yeah, I mean, you know, when you talk about Opus 4, it's the most advanced model from Anthropic to date, so it's extremely capable. It's designed for building sophisticated AI agents that can reason, plan and execute complex tasks with, you know, largely autonomously with minimal oversight. It's said to have best in market performance for coding according to Anthropics benchmarks. And it is really good with long running, long context tasks such as research synthesis or coordinating enterprise operations or refactoring large code bases. So that's Cloud Opus 4. Now Cloudsonnet, while not as big, has its own sort of unique use and value. Maybe you can talk a little bit about that.
- [02:03]
B
So Cloudsonnet 4 is going to complement Opus 4 in terms of balancing performance, responsiveness and cost. So this is going to be really suitable for, I'd say a lot of different types of applications that still want that reasoning, but you want to be able to have that performance. And so maybe you want like lower latency than maybe Opus that's going to take more time to actually do thinking before responding. So think about this is if you want it in your agent agentic types of systems, it also can do the same things at Opus. So the code reviews, bug fixes, real time feature development as well. Um, but I think the thing that stands out for me with Sonnet 4 is that it's actually the same price as Sonnet 3.7. So if you're already using Sonnet 3.7, or I know some people are still using Sonnet 3.5, Sonnet 4, you just switch the model and I'd Be really excited to hear what customers are doing with it.
- [03:05]
A
Yeah. The other interesting thing is this like notion of dual operating modes where you can either have near instant responses or extended thinking for deeper reasoning. And both models offer these and then being able to configure the thinking budget for balancing the latency of how fast you can get the response with the depth of how well researched and well thought out the response is. So I'm curious to see how our customers use that configurability and where that slider shows up for different use cases.
- [03:46]
B
Yeah, absolutely. All right, next one up is AWS Transform for mainframe and VMware. So AWS Transform is a new AI powered service that helps modernize mainframe and VMware workloads to cloud based architectures. There's a lot of different things here. Trudy, what comes to mind for you?
- [04:09]
A
Yeah, I mean I am super excited about this because this is really AI in action in a way that is going to help customers achieve their modernization goals. And we announced this, I think back last year at Re Invent. So this is the very first agentic AI service for mainframe modernization at scale. And so this is agentech AI that accelerates your modernization but also de risks it because it can automate assessment, planning, transformation, both mainframe as well as VMware workloads into cloud based architectures. And this is the kind of streamlined migration. It has code analysis and dependency mapping capabilities. It gives you comprehensive insights into your code base. Lots, lots going on. And then of course the other thing I really like is it does generate technical documentation because you know, if someone offered to like modernize your entire software stack, that is great. But like what happens when you need to go back and fix a few things or change a few things. So it preserves critical knowledge about sort of the features, the programming logic or the path of data flows and in this technical documentation so that then you can, you know what was done and what's retained. And of course you can interact with this technical documentation with AI. So yeah, it's a lot of, there's a lot packed here. And of course the benefits are that the modernization timelines can go now from literally from years to months.
- [06:08]
B
Crazy.
- [06:09]
A
It is, it is. And I mean we see so many customers kind of starting to wanting to do the modernization, but it is daunting just the amount of time it's going to take. But this is a great way to possibly start small on some of sort of isolated projects. See how AWS Transform does get used to this workflow and then you could probably accelerate large scale migration and modernization initiatives with it.
- [06:41]
B
Yeah, and on a similar topic, AWS Transform also is available for. Net, so you can get that agentic AI type of experience that you need to be able to modernize NET applications at scale and really at a fraction of the time. I like that you've got the web experience and then those who are huge visual studio fans, you can use it within there as well. Well, Shruti, what comes to mind for you at this one again?
- [07:16]
A
I mean, it's similar. It has all the key benefits of exciting timelines. Cost reduction, it's 40% lower operating costs by eliminating Windows licensing in applicable use cases. Performance improvement, possibly, you know, up to 2x faster application performance. There is better scalability of course, given that, you know, you can have up to 50% improvement in handling your growing workloads. So yeah, I really like the name Transform because it is truly transformational.
- [07:58]
B
It really is. And speaking of other things that are transformational, the EC2 P6B 200 instances by Nvidia Black Belt GPUs are available. And Shruti, I think you know a bit about this one, don't you?
- [08:17]
A
Yes, I do. I actually lead product marketing for the AI Compute portfolio and so I'm very well versed with these. And we are so excited that these are generally available now and available through Amazon EC2 capacity blocks for ML. So yeah, they are our instances powered by The Nvidia backward GPUs similar to our P5, P5E and P5EN instances that all had the Hopper family of GPUs within an instant. And eight of them. These instances also have eight Blackwell GPUs. They pack 1,440 gigabytes of HBM3E. This is like ultra fast GPU memory. And of course they have instant storage, like local instant storage as well. And the network bandwidth is 400 gigabits per second per GPU, so that amounts to 3200 gigabits per second. So really, really well suited for large scale distributed AI training as well as inferencing. And all customers who are either building or deploying Frontier foundation models want to start using these and get their hands on these. They are also well suited for HPC applications. And so that's the thing about GPUs is that while they're really great for AI, they're also general purpose in that sense, where they can be used very well for HPC as well. And one thing to note on the performance, these Deliver up to 2x the performance for AI training and inference as compared to P5EN instances which were the best GPU based instances in the portfolio up until Blackwell powered P6 came on the scene. So yeah, we are very excited they are available. You can start using them with DLAMIS, they're integrated with EKS, um, and the SageMaker hyperpart support is coming soon.
- [10:30]
B
I just, I can't fathom just like the kinds of applications that people would be building and just like how fast these things are. Like it just, it's mind blowing, that's all.
- [10:42]
A
It is, it is, it is. You know there are so many like especially the. When you start to get into the models that are like touching 1 trillion parameters or close to it, like you know, 500 billion parameters or more. These instances are really, really good for those both for training but also inference because of the amount of memory they pack. They can actually single instance can hold the entire model which means you are operating in a single sort of NVLink domain where all the GPUs are like connected with this ultra fast interconnect and that keeps all the inter gpu, you know like the all, all reduce all the sort of comms operations that need to happen when you're doing distributed training or inference. It makes it much faster to keep it within that ending domain. So yeah, it's exciting for sure.
- [11:39]
B
Really is. So yeah, let's talk about, we've got more updates so let's talk about analytics. Rudy.
- [11:46]
A
Yeah, so Amazon MSK adds support for Apache Kafka version 4 bringing the latest advancements in cluster management and performance to MSK. Provisioned Amazon OpenSearch ingestion increases memory for an OpenSearch compute unit to 15 gigabytes. It's been increased from 8 gigabytes to 15 gigabytes. AWS Glue Studio now supports additional file types and single file output. So the additional file types that are supported are Excel files as source and XML and Tableau's Hyper files as target. Amazon Kinesis Data streams expands IPv6 support to VPC endpoints. There it is again. IPv6. Yeah, those were all the updates under Analytics. Moving on, we have. As if the three updates that we just covered under AI were not enough, we have a few more under artificial Intelligence. Amazon Bedrock Data Automation now supports video blueprints so you can generate tailored, accurate insights in a consistent format for your multimedia analysis applications. BDA or Bedrock Data Automation automates the generation of insights from unstructured multimodal content such as documents, images, audios, videos for your generative AI applications with video blueprints, you can customize insights such as scene summaries, content tags and object detection by specifying what to generate, the output data type and the natural language instructions to guide the generation. We are also announcing that Amazon Bedrock now offers comprehensive CloudWatch metrics support for agents, enabling developers to monitor, troubleshoot and optimize their agent based applications with greater visibility. This is super exciting because it provides detailed runtime metrics for both invoke agent and invoke inline agent operations including invocation counts, latency measurements, token usage, so on and so forth. And this really helps customers better understand their agents performance in their production environments. Amazon Bedrock Guardrails now supports Cross region Inference, which is an optional feature that enables customers to seamlessly manage traffic bursts by utilizing compute across different AWS regions. It's so interesting. I was actually at the Customer Advisory Board meeting earlier this week and there were a couple of customers who brought this up and so now we are enabling it and that's exciting. So you know, if we are nothing, if we are not customer obsessed, it's like principle number one here at Amazon. Okay, Amazon SageMaker and Amazon Datazone announced a new data governance capability that enables customers to move a project from one domain unit to another. Domain units enable customers to create business unit or team level organization and manage authorization policies per their business needs. So now customers can take a project mapped to a domain unit and organize it under a new domain unit within their domain unit hierarchy. So this feature lets customer reflect changes in team structures as business initiatives or organizations change or shift and yeah, so quite useful. Amazon SageMaker Catalog launches Governance for S3 Tables making it easy to well, SageMaker Catalog integrates with S3 Tables which makes it easy to discover, share and govern S3 Tables for users to access and query the data with all Apache Iceberg compatible tools and engines. And the last update we have under AI is aws. Healthomics now supports output mapping files for CWL or Common Workflow Language workflows. So with this launch, healthomics now provides researchers and bioinformaticians with a complete catalog of all outputs generated by workflow runs along with their precise locations in Amazon S3. For everyone who may not know, Healthomics is a HIPAA eligible service that helps healthcare and life sciences customers accelerate scientific breakthroughs with fully managed biological data stores and workflows. And that's all we have.
- [17:02]
B
Next up is Business Applications simple email service Mail Manager announces the addition of a debug logging level for mail Manager traffic policies. This new logging level provides more detailed visibility on incoming connections to a customer's mail manager ingress endpoint, and makes it easier to troubleshoot delivery challenges quickly using familiar event destinations such as CloudWatch, Kinesis and S3. Now we've got one quick update in Cloud financial Management. AWS Cost Anomaly Detection now integrates with AWS user notifications via Amazon EventBridge. This enables customers to create enhanced alerting capabilities in the AWS user Notifications console. Now we've got compute. Amazon EC2 Mac instances now support configurable system Integrity Protection settings. So System integrity protection settings or also you might see it as SIP. This is a critical macOS security feature that helps prevent unauthorized code execution and system level modifications. This enhancement enables developers to temporarily disable SIP for development and testing purposes and install and validate system extensions and drive kit drivers. Optim oh Driver Driver Kit? Oh, let me do that again. This enhancement enables developers to temporarily disable SIP for development and testing purposes, install and validate system extensions and and DriverKit drivers, optimizing testing performance through selective program management and maintain security compliance while meeting development requirements. This new SIP configuration capability is available across all EC2 Mac instant families. AWS Deadline Cloud now Supports Foundry Nuke version 16 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 now enables automatic instance migration when you Reboot after receiving EC2 scheduled reboot event notifications. Amazon LightSail now supports IPv6 connectivity over AWS PrivateLink. AWS Parallel Computing service now supports accounting with Slurm version 24.11. Amazon EC2 makes it easier to launch Windows instances with EC2 fast launch EC2 fast launch reduces the launch times of Windows instances using pre provisioned snapshots. AWS Deadline Cloud now supports specifying a configuration script on both Linux and Windows Service managed fleets. The provided configuration script will be run with elevated privileges on each worker. All right, next topic is containers. Amazon EKS announces the general availability of EKS Dashboard. This is a new feature that provides centralized visibility into Kubernetes infrastructure across multiple AWS regions and accounts. EKS Dashboard provides comprehensive insights into your EKS clusters, enabling operational planning and governance. So some things you could visualize are your entire Kubernetes infrastructure without having to switch between AWS regions or accounts. You can gain aggregated insights into clusters, manage node groups and EKS add ons. This is going to include clusters running specific Kubernetes versions their support status, upcoming End of Life Auto Grades, Manage Node group, AMI versions, EKs add on versions and worm. So I'm definitely excited for this one and you can be able to access it in the US East 1 AWS region and aggregate those EKS cluster metadata from all commercial AWS regions. Really good one. Next ECS has added support for Amazon EBS Provisioned Rate for Volume Initialization. This feature helps you provision and attach fully performant Amazon EBS volumes from Amazon EBS snapshots to your Amazon ECS tasks, accelerating initialization for your ETL jobs, media transcoding and ML inference workloads deployed on Amazon ECS. All right, next topic is databases. DynamoDB Local is now accessible on the AWS Cloud shell and if you aren't using Cloud Shell you should. It is a browser based pre authenticated shell that you can launch directly from the AWS Management console and with DynamoDB local you can now develop and test your applications by running DynamoDB in your local development environment without incurring any costs. I think this is so cool. I mean as customers are looking to be able to just do things in a more cost effective way. Being able to test locally first before actually doing something in the cloud and realizing oh that didn't work without incurring those costs. I'm a huge fan of so definitely check that one out. Amazon RDS announces a new capability that helps you view engine lifecycle support dates for your databases. This new feature provides a centralized and convenient place to access engine support dates, offering greater control over your database lifecycle management. So now you have to start. You can better plan. There is no excuse to leave it to the very, very very last minute. Amazon RDS for MariaDB now supports community MariaDB miner versions 10.5.29 and 10.6.22. Of course we recommend that you upgrade to the latest minor versions to fix node security vulnerabilities in prior versions of MariaDB. This is me doing the reminder because Simon, who always chimes in here to say give those reminders and tell people to patch his stuff is not here. So I've got to take on that responsibility and be that person. Amazon RDS for Oracle now supports credential management with AWS Secrets Manager for databases that adopt Oracle Multi Tenant architecture. Oracle Multitenant Architectures enables customers to consolidate data and code from multiple databases into one database by setting up a multi multi tenant container database that can include multiple pluggable databases. So with this launch, customers can use AWS Secrets Manager to manage user credentials for their tenant pluggable databases. Amazon RDS for Postgres 18 beta 1 is now available in the Amazon RDS Database Preview environment. This allows you to evaluate the Pre release of Postgres 18 on Amazon RDS for Postgres. Amazon RDS for Oracle now supports the April 2025 release update for Oracle Database versions 19c and 21c. Amazon Aurora MySQL 3.09 is now generally available. Amazon Aurora and RDS for Postgres, MySQL and MariaDB now offer reserved instances for R8G. NM8G instances I definitely suggest taking a look at the Graviton instances if you're not familiar with them. They can offer up to 40% performance improvement and 29% better price performance compared to the equivalent Graviton 3 based instances, and they could even offer better price performance if you're using x86 based instances. So I definitely suggest looking at Graviton instances if you aren't familiar with them, and then you can also look in the console to be able to see if there's additional cost savings if you can use reserved instances if you're not using them already. Customers running Amazon Aurora and RDs for Postgres, MySQL and MariaDB databases can now purchase reserved instances for R7i and M7i instances. These instances are powered by custom 4th generation Intel Xeon scalable processors and provide larger sizes up to 48x large with an 8.1 ratio of memory to VCPU and the latest DDR5 memory. Amazon RDS for Oracle now supports the Spatial Patch Bundle for the April 2025 release update for Oracle Database version 19C. This update delivers important fixes for Oracle spatial and graph functionality, helping ensure reliable and optimal performance for your spatial operations. AWS Database Migration Service now supports Data Resync. This is a new feature that automatically corrects data inconsistencies consistencies identified during validation between source and target databases. Data Resync integrates with your existing DMS migration tasks and it supports both full load and change data capture phases. It uses your current task settings including connection configurations, table mappings and transformations to apply correctness automatically, and this is going to help ensure there's accurate and reliable migrations without manual intervention. With Data Resync, AWS DMS can detect and resolve common data issues. This might be like missing records, duplicate entries, or mismatched values based on validation results. This one's really exciting. I think anything that can help make a migration just a lot easier.
- [27:15]
A
Awesome. Next up, we have a few updates under Developer Tools AWS Code Build now supports new IAM condition keys, enabling granular access control on Code Build's resource modifying APIs. These new condition keys cover most of Code Build's API request context, including network settings, credential configurations, and compute restrictions. AWS CodePipeline now supports Deploy spec file configurations in the EC2 Deploy action, enabling you to specify deployment parameters directly in your source repository. You can now include either a Deploy Spec file name or a Deploy configurations in your EC2 deploy action. The action accepts Deploy Spec files either in the YAML format or rather in the YAML format and maintains compatibility with existing CodeDeploy app set app spec files. AWS CodePipeline now supports deploying to AWS Lambda with Traffic Shifting. This is a new Lambda deploy action that simplifies application deployment to AWS Lambda, and this feature enables seamless publishing of Lambda function revisions and supports multiple traffic shifting strategies for safer releases. So for your production workloads, you can now deploy software updates with confidence using either Linear or Canary deployment patterns. The new action integrates with CloudWatch alarms for automated rollback protection. If your specified alarms trigger during traffic shifting, the system automatically rolls back changes to minimize impact. AWS Code Build announces support for remote Docker servers, allowing you to speed up image build requests. You can provision a fully managed Docker server that maintains a persistent cache across builds.
- [29:23]
B
One update in end User Computing Amazon Workspaces Pool now support always on running mode so Workspaces Pools enables customers to reduce costs by sharing a pool of virtual desktops across a group of users who get a fresh desktop every time they log in and then always on for Workspaces Pools. This is available in all different regions and this is going to allow users to have their own virtual desktop session provisioned in seconds and allowing them to be more productive immediately because it's always on. All right. Management and Governance AWS Control Tower introduces account level reporting for baseline APIs, so now customers can programmatically view the statuses for their governed accounts via these baseline APIs. This is going to contain best practice configurations, controls and resources required for governance. When you enable this baseline on an organizational unit, member accounts within the organizational unit are enrolled under governance. So now with this new experience you can use baseline status to view enrollment for your accounts and and use Drift status to identify when account and organizational unit baseline configurations are out of sync. In addition to seeing statuses for your accounts and organizational units in the AWS Control Tower console, you can use the List Enabled Baselines API to view statuses for your enabled baselines and to view statuses for individual accounts, you can use the Include Children flag. So it sounds really cool this one. I mean just being able to have a better visibility of like the overall status, especially as you have multiple accounts, some customers have like thousands of accounts and being able to just programmatically be able to manage governance and control access. That's the only way to do it. Amazon CloudWatch application signals introduces auto monitor support for EKS workloads. This is going to make it easier for your EKS applications through a single configuration flag within the CloudWatch observability add on. So once you have enabled this you can gain access to pre built standardized dashboards within CloudWatch application signals and track application performance against key business or service level objectives. AWS Organizations now supports IPv6 AWS CloudWatch Synthetics adds safe Canary updates and automatic retries AWS announces the general availability of Migration Assessment Capabilities in AWS Transform Migration Assessment in AWS Transform analyzes your IT environment to simplify and optimize your cloud journey with intelligent data driven insights and actual recommendations. You just have to upload your infrastructure data and AWS Transform will deliver a comprehensive analysis that typically takes weeks and just minutes. This sounds super useful. I definitely would check this out if you are planning on migration.
- [32:44]
A
Next up we have a couple of updates under Migration and Transfer. So AWS announces the general availability of migration assessment capabilities in AWS Transform. Migration Assessment in AWS Transform analyzes your IT environment to simplify and optimize your cloud journey with intelligent data driven insights and actionable recommendations. You simply have to upload your infrastructure data and AWS Transform will deliver a comprehensive analysis that typically takes weeks in just a few minutes. And of course this is powered by agentic AI and AWS Transform removes weeks of manual analysis by providing you near instant visibility into your infrastructure and automatically discovers cost optimization opportunities. It helps you produce a business case including key highlights from your server inventory, a summary of current infrastructure, multiple different total cost of ownership scenarios with varying purchase commitments, operating system licensing options and more. So this is definitely extremely useful. You know before you take on a migration project, getting this sort of assessment with all these insights packed and getting them in just a few minutes instead of having to do it yourself and typically something that would take you weeks. Awesome. And then the second update we have is AWS Transfer Family now Supports MLKEM or FIPS203, a post quantum algorithm standardized by the National Institute of Standards and Technology or NIST for SFTP file transfers. Quantum resistant Public key Exchange helps protect transfers of data files that require long term confidentiality against harvest now decrypt later kind of threats. In such scenarios, an adversary may be recording present day traffic for decrypting once crypto analytically relevant quantum computers become available. Oh wow. I didn't even know these kind of threats exist.
- [35:09]
B
Crazy.
- [35:09]
A
And yeah, I mean it's really cool that we are kind of preempting some of that with a capability like this.
- [35:20]
B
All right, next we've got one quick update in networking content delivery. Amazon VPC has enhanced CloudTrail logging to include VPC resources created by default during a VPC creation. This enhancement offers improved visibility for VPC resources and aids in auditing and governance.
- [35:42]
A
Great, we have a few updates under Security, Identity and Compliance. Amazon Cognito now supports OIDC prompt parameter or the OpenID Connect prompt parameter incognito Managed Login. Amazon Inspector enhances container security by mapping ECR images to running containers. And that's it. I think those were all the updates we had under Security, Identity and Compliance. We do have a few service changes to announce and we understand that the decision to end support for a service or a capability really impacts customers. So we approach these decisions only after a lot of thoughtful debate and careful consideration. So when end of support is necessary, we like to provide customers with detailed guidance. And so we are going to include some of that guidance here, but we encourage you to go check it out online as well. So there are services that are closing access to new customers starting June 20, 2025. And Amazon Timestream for Live analytics is the only service that we are listing at this time that is closing access to new customers, but existing customers will be able to continue utilizing this service. Next, we have services that are announcing end of support, so we will be ending all support for these services. You should go online and review the specific end of support dates and migration paths for each of the services below. Or rather each of the services I'm about to tell you. Amazon Pinpoint AWS IQ AWS IoT Analytics, AWS IoT Events, AWS SimSpace Weaver, AWS Panorama, Amazon Inspector Classic Amazon Connect, Voice ID and AWS DMS Fleet Advisor. So once again, all of the services that I just called out, we are announcing end of support review the end of support dates online and also the migration paths that have been outlined. And finally we have a few services that have reached their end of support date and can no longer be accessed. And Those are AWS private 5G and AWS datasync discovery. Once again, to learn more about sort of you know the AWS service changes, please visit the AWS Product Lifecycle page. And I think those are all the updates we had today.
- [38:38]
B
Yeah, Shruti, where can people go if they want to contact you about the AWS podcast?
- [38:45]
A
Yeah, they can find me either on LinkedIn or on X.
- [38:49]
B
As Shruti Gopakar Same with me. You can find me on LinkedIn. Jillian Ford and if you want to reach the whole AWS podcast team, go to aws.Amazon.com podcast or just go into your favorite browser. Search for the official AWS podcast and there is a big button right on the page that says submit questions and feedback. We actually read them, surprisingly enough. So please submit your feedback. We would love to hear more feedback from everyone. And until next time, it'll probably be Simon. But who knows? Keep on building.