← AWS Podcast
AWS Podcast
#750: re:Invent 2025 - Day 3 Wrapup
00:22:42
It is the end of re:Invent! Simon and Jillian share some updates and also take a moment to reflect o
Loading summary
Transcript43 lines
- [00:00]
A
This is episode 750 of the AWS podcast released on December 4th, 2025.
- [00:09]
B
Hello everyone and welcome back to the AWS Podcast. Simon Alish here with it. Great to have you back. Joined by Gillian Ford, of course. G', day, Gillian.
- [00:16]
A
Good day, Simon. I'm. I'm kind of sad that this is like the last day of Re Invent.
- [00:20]
B
It is, it's. It's day three and it's the, the. We're coming to the end of the festival. I'm sure there's a lot of very tired people in Vegas at the moment. It's quite a. It's quite an experience. I always tell people you need to go to Re Invent at least once.
- [00:33]
A
You really do.
- [00:34]
B
Yeah, a lot of people like to go every year. I'm not one of those people. I think I've gone five times or four times. I've gone enough times that I don't feel the need to go. But it is an amazing thing. And so there've been some announcements overnight, not too many. So we're also going to talk a bit about the year. But firstly, the big thing was that we've introduced Graviton 5, which is our most powerful and efficient CPU. And this really comes because people always need more things are growing all the time. So the graviton based EC2M9 instances give you the highest CPU core density available in Amazon EC2. In fact, you get 192 cores in a single package. And what this does is it reduces the distance that data has to travel between cores. So your inter core communication latency drops by up to 33% while you're getting more bandwidth. Now it's interesting to think about. You know, we often think about latency between regions or availability zones or servers. This is between cores on the die. It's pretty, pretty cool. But this is really important when you're doing things like real time gaming, high performance databases, big data analytics, et cetera. The chip also has a 5 times larger L3 cache. So this is the memory buffer that keeps frequently accessed data close to the processor. And so each core has access to 2.6 times more L3 cache than Graviton 4. So you get fewer delays and faster memory speeds and a lot of work on the networking side as well. So up to 15% high network bandwidth, 20% higher EBS bandwidth across instant sizes on average, and up to twice the network bandwidth for the largest instances. So if you're moving a lot of data, this is, this is kind of important, Julian and also touches on energy use as well.
- [02:22]
A
It is. There's just like a lot of really important things to this. Like, I think for me what stood out is like the enhancement, enhancements with the Nitro system. That is just like some serious like engineering feat. I mean really being able to introduce the nitro isolation engine. So really what that is for folks. So we have like Nitro has been around for a couple of years, but I mean this being built into Graviton 5 instances, the 6th generation Nitro cards, they're going to offload the virtualization, storage and networking functions that are dedicated to, to the hardware. So you get the actual formal verification to provide mathematical certainty that your workloads are isolated from each other. And AWS operators. I mean that I want to know how they did that.
- [03:16]
B
Yeah, well, and that's the nice thing is just a couple of weeks ago we had Byron Cook on the show who talked to us about automated reasoning. And automated reasoning is the magical maths that helps us do this because it's not about just saying things are happening, happening, it's about being able to prove that they're happening, which is very, very cool. The other thing with this new CPU is it is using 3 nanometer technology. I can't believe that's the level we're down to now. Like, it's crazy in terms of the fine grayness of it all. So that's, that's the current generation as well. And what's interesting is that a lot of folks use AWS Graviton. In fact, for the third year in a row, more than half of the new CPU capacity added to AWS is powered by Graviton. And 98% of the top 1000 EC2 customers use Graviton. So folks like Adobe Airbnb, Atlassian Epic Games, Formula One, Pinterest, SAP, Siemens Snowflake, Synopsys, all these folks are using it. So it's one of those things. I think you and I talk about Graviton quite a lot on the show and often it's just a case of saying, hey, if you can use it, you should use it because it's going to make life better for you. I reiterate this point. It's kind of like a nice bonus when you need a bit more performance, you want better cost benefits as well. And with a lot of the services like rds, et cetera, it's just a checkbox. Like it's not a big job today.
- [04:44]
A
Totally. And another call out as everyone's getting really excited about this. So the M9G instances those are going to be available in preview. They are available in preview, the C9G instances and the R9G instances for memory intensive workloads. Those are planned for next year, so stay tuned. Other Cool Announcements Amazon Bedrock now supports Responsys API. This is a new OpenAI API compatible service Endpoint Responsys API enables developers to achieve asynchronous inference for long running inference workloads, simplifies tool use integration for agentic workflows, and also supports stateful conversation management instead of requiring developers to pass the entire conversation history with each request. I know everyone listening has probably tried that at one point.
- [05:42]
B
It ends poorly.
- [05:43]
A
Yep, the Responses API enables them to automatically rebuild context without manual history management. These new service endpoints support both streaming and non streaming modes, enable reasoning effort support with chat completions API, and require only a base URL change for developers to integrate within existing code bases with OpenAI SDK compatibility. So now starting today you can use this with the OpenAI GPT OSS 20B.
- [06:19]
Slash. The 120B models and there's support for other models are going to be coming soon.
- [06:26]
B
And speaking of AI, we're happy to announce TypeScript support in strands Agents in Preview and a whole bunch more now if you haven't come across the Strands Agents SDK. This is an open source Python framework that takes a model driven approach to building and running AI agents in just a few lines of code. And so now TypeScript support is available in Preview so you can choose between Python and TypeScript depending on what you like. Now I'm gonna say I'm a massive Strands fan. It is magical. It takes away so much of the heavy lifting of what you do and it just runs things really, really well. So the typescript support in Strands has been designed to provide an idiomatic TypeScript experience with full type safety, Async Await Support and Modern JavaScript Typescript patterns. Strands can easily be run in client applications in browsers, server side applications in runtimes like AWS Lambda and Bedrock Agent Core, and you can also build your entire stack in TypeScript using the AWS CDK. There's also some other updates for the strand SDK, so I'm excited. Firstly, Edge device support for Strands Agents is generally available, so this extends the SDK with bidirectional streaming and additional local model providers like LLAMA CPP that also let you run agents on small scale devices using local models. So you can get the power of this on a very small locally running model. Secondly, Strand Steering is now available as an experimental feature, giving you a modular prompting mechanism that provides feedback to the agent at the right moment in its lifecycle, steering agents towards a desired outcome without rigid workflows. And finally, Strands Evaluations is available in preview. This gives developers the ability to systematically validate agent behavior, measure improvements, and deploy with the confidence during your development cycles. It is so much fun using Strands, I have to say. It just lets you. Lets you build a lot of cool stuff with a fair bit of control, but without having to do all the boilerplate stuff you'd normally have to do. It just does it for you. So I'm a fan, Gillian.
- [08:31]
Very excited.
- [08:31]
A
For this one as well. I've seen a lot of startups that I've worked with who are typescript fans, and they've been waiting for this one. So I'm sure this is one that people are going to start being able to play around with.
- [08:42]
B
It's a nice thing. It's a nice thing. Now, as we draw to the end of the year and of course it's been a big year on the podcast. Lots of episodes, lots of new sort of topics, and Julie and I have got a few things cooking for the new year as well. We're sort of always working to improve the podcast and the content and the topics we get to talk about. But let's look back for a moment, Gillian. Let's, I guess, think about some of the things that have really stood out to us in terms of things we've covered, et cetera. I think for me, it would be remiss of me without for the very last time saying that it was indeed the year of IPv6.
- [09:21]
I think we covered IPv6 a lot this year.
- [09:24]
A
It was important for customers, and I don't think we could have like an end of year episode without you saying that.
- [09:29]
B
Well, I will not be saying it next year because it will not be the year of IPv6, because this was the year of IPv6. But, but in all seriousness, I mean, we've been talking as an industry about moving from IPv4 to IPv6 for a long time, and thankfully, I think all services are uplifted, have uplifted themselves to a point where you can pretty much use it across the board now. So if you, if your organization is really pushing to use IPv6 as the standard, I think you're looking pretty good now.
- [09:57]
What about you, Jillian? What leapt out at your, at your. I've got a, I've got a list, but I don't want to trample on your list.
- [10:04]
A
So what really stands out is Amazon Connect. I mean we've had so many update shows this year where over and over there's so many releases from that team. And I think what's interesting is that especially this year, what I've seen with a lot of customers is that they've really been interested in voice applications. Whether that's to improve like the customer experience, maybe being able to extend like to patients, whatever kind of the use case is. Voice, especially with AI, has become a trend. And now with Connect being really this ecosystem to build so many different types of voice applications instead of just kind of cobbling it together with the building blocks within aws, I'm very excited for customers that really just want to be able to improve their patient. I mean their. The customer experience. I work with healthcare a lot, so.
- [11:01]
B
I. Yeah, you think about patients a lot, which is fair enough. That's a. It's a big use case for Connect. I agree. They've been iterating at a massive pace. Speaking of iteration, I think for me one of the standout themes has been capacity increases. And the reason why I like this is I kind of have a deep appreciation of the absolute complexity and skill behind doing this. Like, you know, doubling the capacity of something sounds easy on paper, but there's a lot of engineering work that goes on to do that. And we've seen this happen across lots of RDS instance types where suddenly you can have up to 256 terabytes of storage. We've seen capacity increases for Kinesis, for SNS, for SQS, for obviously S3 just recently, 50, 50 terabyte maximum objects throughput increases of 10x for different services. Like this is amazing stuff. And what I love is that customers don't have to do anything to get the benefit of it. Like they just get it. And that's, I think often we don't talk about, you know, when people talk about why do you use the cloud? One of the reasons is, is that the stuff I use constantly gets better and I don't have to do anything to make it better. Whereas in the past if you wanted those types of increases, you're doing the engineering to make that happen. So I think it's a kind of, it's a big deal for a customer perspective.
- [12:23]
A
Yeah, that is true. And even kind of similar into that topic, I think 2025, 2026 are really the years of like the MCP ification of things. We started to see that through all these update shows and this is really the first year that I've seen, what's been announced was something that actually wasn't in the official AWS documentation. It was in our AWS Labs GitHub repo. And that's, I think, a testament to just the pace of innovation and also just that this is an area that I think is really going to influence how customers architect their applications.
- [13:08]
B
I think you're right. And I think it's interesting to reflect that pretty much all the best practices around AI and gen AI development at the start of the year are completely different now at the end of the year, like, it's, it's changed. Like the way we prompt, the way we manage context, the tools we use, the abstractions we have available to us, it's just completely changed. And it can be challenging in many ways. Like, I've been working on a particular application that I built for internal use, and I kid you not, I've rewritten this thing three times now because just when I get it to a point where I think, okay, it's nearly ready, something comes out that renders most of the work I did obsolete, and I shouldn't have done it now because I can use something else to do it. And so I think that's a testament to how fast this is all moving. And it's both challenging and exciting, I think. I mean, the speed of things, things like Kiro and Kiro cli, I live in those all the time now, whereas I wasn't doing that last year.
- [14:07]
A
Yeah. I think at the same time, though, I think AWS has historically just really heavily invested in educating customers early on, even workshops at Re Invent AWS summits. And I think it's just like, great to see, obviously, that we've continued to be able to do that as we adapt and evolve to what these new best practices are. Like the, the AI league this year, I mean, and the prizes that were awarded, I mean, that was absolutely amazing.
- [14:38]
B
That was cool. That was very cool. Very cool. I think one of the other elements that really appealed to me is again, back onto maybe something more not mundane, but foundational, which is storage. Storage continues to be fundamental to what we do. And the work that S3 has been doing in terms of providing things like S3 tables, a lot of the performance increases that we've seen, et cetera. I mean, it just creates that really solid foundation for folks to be able to build, from, which I think is vital if you're trying to do anything at scale and also efficiently, you know, the, the intelligent tiering stuff that continues to. Iterate just helps people build. You don't even think about storing data anymore. When you think about it, like I used to think about it a lot. Now it's like, oh yeah, I'll just shuck it in S3 and it's. Call it good.
- [15:22]
A
Yeah, it is interesting that you point that out because it's like on the one hand, there have been some releases this past week and Lambda Durable functions. It really stands out as it kind of changing the paradigm a little bit of what's in terms of a best practice architecture. Whereas you still have some of these fundamentals that we're. But now we're just expanding what the capabilities are like S3, obviously in terms of the object storage size increases and being able to do vectors with S3.
- [15:57]
B
So it's like, yeah, that's a huge. The vectors one was a huge one. You're right. That's a really good call out. And I think also, like looking back, some of the guests we've had on, some of the customers that have spoken about their experience, I think is always. Is always really powerful. I've loved having the Verna Vogel series around. The Frugal Architect, I think, has been very revealing in terms of how people think about problem domains, which has been great having Byron Cook on again to talk about automated reasoning whenever he speaks. It's fascinating because you're hearing someone come from a completely different discipline, like a deeply pure mathematical discipline, and being able to translate that into the practical world. I think speaking to customers and to engineers, Gillian, is one of the biggest rewards of this job.
- [16:41]
A
I agree. Yeah. There's just so much that you learn. None of us can really know everything. So just being able to hear someone else's experience.
- [16:52]
I always learn something. And so I'm sure the listeners do take something away from it as well.
- [16:59]
B
It's so true. And I think.
- [17:02]
As I sort of think about what's been going on, I think one of the other things that I always try and call out is to also not get overwhelmed with all the changes and everything that's going on. There's some fundamentals that will hold you in good stead. And I talk about these on the podcast a bit, but I'll talk about them again because that's the whole point. The first thing is patch your stuff. I mean, you know, the, the security profile of our industry tends to be quite poor. And it doesn't have to be that way. Particularly again, if you're using the cloud, if you're using aws, patching can be automatic or automated. And it should just be fundamental, like there's no excuse to be running things that aren't patched properly anymore. And I think that's really important. And it also means that if something is emergent where people figure, oh, there's this vulnerability, the speed of patching is also really, really fast in the cloud. Like it can be done in a morning type stuff rather than oh my goodness, now there's a six month project to do all this. I think that the second one is to always review the sizes of EC2 instances and other things that you're using to see is are they the best value for money at the moment. And like I say, if you're not doing that at least every six months and probably three months is the right way to go, you're just missing out. You're leaving efficiencies to be had because you take those efficiency savings and you invest them in other things you want to do. So not taking advantage of that is kind of missing out on an opportunity. And I think related to that is really reviewing your understanding of best practices. And as we touched on things are moving so quickly. What was the best way to do things a little while ago may not be the best anymore. You know, some of the recent changes with the amount of data Lambda can cope with and as you mentioned, that that ability to have long running tasks within Lambda changes the economics of Lambda, so it may change the way for your domain you operate. So it's both challenging but also exciting, I think, Gillian, because it sort of forces you to have that mental plasticity to change your mind about things.
- [19:01]
A
I agree, but I also think it's never just been faster to be able to run all of these tests, to be able to get the data that you need to be able to make these decisions. I mean, look, there were CloudWatch, Security Hub, both had releases that were related to helping customers be able to really understand holistically what's going on in their AWS environment.
- [19:27]
You've got the AWS MCP servers, one of which is related to cost optimization and understanding your costs and cost projections. So it's never been easier to be able to be able to make really quick decisions and be able to optimize your architecture.
- [19:44]
B
It's very true, very true. So Gillian, any final thoughts, reflections that you wanted to share with the listeners before we wind up for this year?
- [19:54]
A
You really nailed it. I think it's really, of course, keep listening to the AWS podcast.
- [20:01]
B
How would you know this great information.
- [20:05]
A
But seriously, I think it's like it's never been a better time to really experiment with, of course, AI using AI for whatever it is that your role is. I mean, I think it's interesting that the biggest announcements at re invent were AI for DevOps, AI for security, at least in terms of new service. There are other big announcements, but two new services. So it's really never been a better time to just learn and experiment, especially with using AI for your own productivity.
- [20:39]
B
Agree, agree. Gillian, I want to thank you for all the work you've done on the podcast this year. You've been a stalwart co host and done a huge amount of work in the back end. And I think what people may not be aware of is that for both Gillian and I, the podcast is not our job. The podcast is in many ways a passion project. Our full time job is being solution architect, and we love that job because it lets us talk to customers all the time. And it means that when we're doing the podcast, I hope it shows that we have some degree of understanding of how these things work, in quote, unquote, the real world. And I consider that a feature, not a bug in as much as that we're not talking from sort of some academic viewpoint. We're actually talking from experience. And. And Gillian, your experience has been, I think, invaluable for folks, and I think that it's really come through in a lot of the episodes you've done with guests as well.
- [21:30]
A
Thank you so much for saying that. And I would say the same for you. I mean, I've been a listener of the AWS podcast for eight, nine years. You've been here for like 10 years.
- [21:43]
B
Scary.
- [21:46]
A
I mean, you're such a leader, a role model. I always learn something from you every single time on the episode, and I'm sure listeners always do. And that's why the podcast has the following that it does.
- [21:58]
B
Well, we do hope you all have an amazing break. There will be a few extra episodes sporadically sprinkled through the holiday season, so there'll be a few there. But also it's always a great time to catch up on all those episodes you didn't get to listen to. I highly recommend listening to a lot of the Interview Pro episodes we did. There's a lot in there that you can get out of and learn from. And we have some absolute rippers coming up in the new year as well. They're cooking away at the moment, so we're very excited to release it. But thank you again for all your support. Remember, the biggest thing you can do for the podcast is tell someone else that the podcast exists. That's the main thing. And of course, as ever, and with all this cool new stuff, until next time, keep on building.