Next in Tech | Ep. 249: AWS re:Inforce conference

View Full Transcript

Eric Hanselman: Welcome to Next in Tech, an S&P Global Market Intelligence podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, Chief Analyst for Industry and Company Data at S&P Global Market Intelligence. And today, we're going to be talking about all the things going on at the AWS re:Invent conference. It was as usual a huge gathering out in Las Vegas and a lot of technology perspectives, a lot of infrastructure perspectives, all sorts of things were going on. And with me to discuss it, we have a great and varied panel, Melanie Posey, William Fellows, Henry Baltazar and Scott Crawford. Welcome back to the podcast to all of you. 
 

William Fellows: Good to be here. 
 

Melanie Posey: Thanks Eric! 
 

Henry Baltazar: Thank you, Eric. 
 

William Fellows: Thank you. 
 

Eric Hanselman: Great to have you here. There was a lot that was going on. I guess maybe to start, Melanie and William, a lot of the core infrastructure announcements and AWS activities were really front and center. What was your take? And what were the big activities that were going on? 
 

William Fellows: So, I'll start, if that's okay. Well, this is -- I've attended all the reinvent since 2012. And I think underlying this year's addition more than ever was better, faster, stronger, harder because I think AWS, there's a bit of a perceived sense in which AWS needs to play catch-up in some areas above and beyond the infrastructure layer, which still continues to own and dominate, and I think the key existential question in the industry really has been, well, would you go to AWS for your AI even if you have it for your infrastructure. And given the investments that have been made by the other hyperscalers, that is up for -- the jury is still out on that one as we speak. And AWS certainly did a good job on reinvent in trying to narrow that gap or change that perception. It was particularly hard and strong on those things, both from the point of view of the underlying infrastructure with the new GPU announcements all the way through to the uplifting of its whole Agentic core and the models that are supported on it. And I'm sure we'll get on to talk to also about, well, as the world turns, so things change, and AWS has now been towards franchising multi-cloud as a first-class citizen in the universe in which it exists. 
 

Melanie Posey: And I would add to that just to what William said that there was definitely more top of stack focus this year than we've heard in quite a long time. And AWS with the whole world of Gen AI and Agentic AI, they are facing a bit of a crossroads now because when you look at it, AWS is the incumbent here that the days of AWS being the scrappy startup in this brand-new thing called cloud, that was all 20 years ago. And so, AWS has pole position in terms of being enterprises go-to cloud provider for general purpose IT kinds of things like the websites and e-commerce engines and all of that good stuff that AWS innovated on 20 years ago. And to W's point, just because you have your general-purpose IT workloads in AWS, does that necessarily mean you'll have your agentic workloads there, too? Maybe because data gravity is a real thing, but not all the data is in the cloud. So that brings us to some interesting scenarios where AWS is now openly -- maybe a bit reluctantly but openly endorsing hybrid and multi-cloud now. 
 

Eric Hanselman: Well, speaking of all your data not necessarily being in cloud, Henry, we've talked about a lot of the concerns about migration and data placement. There were no shortage of data-specific activities going on at reinforce as well -- or sorry, reinvent as well. 
 

Henry Baltazar: Yes, I would agree with that. I mean, as we discussed in our previous podcast, I mean, data is clearly on the move. We saw a much higher percentage of organizations moving data multi-cloud. That used to be like 40%. Now it's over 70%. Also, the other thing that's concerning is these payloads are getting bigger and bigger. We're also seeing more organizations moving over a petabyte, which before was like more around the 30%, 40% range, but now also around 30%, 70% as well. So now is the time for us to really care about these things because data is clearly on the move and any of these architectures, multi-cloud, hybrid cloud, whatever, are going to have to have data migration, data mobility as a big part of them. So that should be a big concern. I think case in point here is also a little early in the discussion for this, but the egress charges continue to be a big thing. The egress charges, not just them, but also their major hyperscaler competitors have those. And they continue to be a big problem for organizations because they add so much to the cost of doing business when doing a multi-cloud type environment. 
 

Eric Hanselman: Well, and I'll bring up one of the other things that was an announcement of the interconnect capability is the idea that AWS is now going to connect with other cloud providers much the same way it had done direct connect years ago to enterprise and really starting initially with Google, with Azure on the horizon. bringing together what is this great grand multi-cloud future and improving pathways to data that may be in other locations. Interesting in where that goes. Scott, I had that Freudian slip of talking about reinforce, which is the security conference rather than reinvent, which is the large, big 1 conference, there was a lot happening on the security side as well, which is clearly occupying a chunk of my brain. 
 

Scott Crawford: Well, the data -- at least you didn't start the report with calling it reinforce reports, so I go back and global edit of that myself. But yes, the reinforce conference, as you brought it up, that's another annual AWS event, typically in the summer, so they're usually 6 months apart from each other reinvent and reinforce. Regardless, it's still really important that security people pay close attention to reinvent because the strategic announcements that affect the entire portfolio and AWS overall strategy will be made at Reinvent in addition to a number of strategic security announcements as well, too. And a couple of examples of the broader announcement that do have -- announcements that do have security impact. You already touched you and Henry Wiff, all of us have touched on interconnect multi-cloud. It was kind of an interesting announcement because it was off to the side almost because it was announced on Sunday, Google came out with its own announcement of the initiative of branded as branded on its side of things. But so many people and quite a lot of people in security saw that as being a pretty strategic thing. And we've seen that as well, too. As you know, we've identified this pattern in the security market at least anyway, but it goes broader than that, that we've been tentatively calling an ecosystem, meaning Providers are increasingly motivated to provide value-add services on top of things like high-performance interconnect between enterprises and third-party providers. And a good deal for the reason for this is, Melanie, your own research shows just how distributed AI workloads are across virtually every venue you could possibly imagine. So, AI workloads running here, integrating with applications running there. This pattern seems to be increasingly prevalent, and this is just another example of that. And there's security functionality that can be added to that value-add functionality, which we already have with some interconnect providers. But with the announcement of interconnect multi-cloud, obviously centered on Amazon VPCs, but also additional capabilities such as AWS Transit Gateway and AWS Cloud WAN. So, these strategic announcements are very important to security people. And similarly, with -- I think my colleagues will talk a bit about agent core, Bedrock Agent Core as well, too. There's policy provisions in agent core that could be directly relevant to security as well, too. So, while there's always a raft of security announcements that reinvent and even though reinforce is the primary security venue for AWS, there are still quite a few implications for broader strategy, broader announcements that security teams must pay attention to. 
 

Eric Hanselman: Well, and rounding out, I'm going to pick up some of what John Abbott has been writing about on the semiconductor side because, of course, underpinning all of this is the silicon that is helping to drive some of this and a lot of focus, of course, if you look at Graviton 5, their new ARM-based processor at the low end. The interesting thing, watching some of the AI-focused silicon advancements and seeing what has been the conversation around the specific AI-specific silicon that AWS has, both Trainium and Inferentia devices. They are counterparts to Google's TPUs. This evolution of how the various devices are actually being pitched into the market. Originally, there was a split between the Tiniumfererenia for what was being used for inference operations and what was being used for training. And in fact, they're now starting to identify that we're seeing enough horsepower being thrown at inferencing. Of course, the Trinium devices are really solid for that. And in fact, that brand looks like it may actually consolidate somewhere down the road. So interesting to see where these are going. and the fact that there is now a new version of the Ultra server. You don't often think about AWS providing large physical servers, but the Ultra server systems, which are great huge stacks of trainiums and the new Trinium 3s, being able to go build out these massive server environments is now one more big part of the AWS portfolio, which gets us right back around to thinking about that overall infrastructure piece, getting right back to a lot of the focus on what they were building out and some of those capabilities. So, Melanie, that's an area where the big picture really is looking to get more into agents, looking to become more of the place where agents really are going to live and, of course, a supportive environment in which to do development for the Agentic world. So, a lot that was happening trying to actually deliver and level AWS up into the agentic world. 
 

Melanie Posey: And that's a great point, Eric. And I think one of the larger subtechs around all of that is AWS is moving from becoming kind of a builder's paradise to an operator's paradise, not to confuse any of that with Ginkgster paradise. But it's the whole idea that people would come to AWS and build and then they would do that as DIY and then operate in DIY mode as well. But a couple of years ago, AWS said they work from the customer backwards in terms of what they put out there in terms of offerings. And increasingly, enterprise customers are asking for easy buttons in terms of we don't want to build all of this stuff ourselves. So, we've seen a lot more, I guess, you would say, sort of managed package productization coming out of AWS in recent years. And we've definitely seen that with some of the enhancements to Bedrock and also with Agent Core and also the Ciro IDE system where it operates more as or from a reference architecture or a standardized set of components that the customers can then customize to their own needs. It's definitely a shift in terms of how organizations look at cloud. It's not kind of a playground where you can go off and do anything you want. When you start getting into scale operations, which enterprise Agentic AI is definitely a scale operation, you need policy, you need guardrails, you need standardized infrastructure, you need standardized run books, you need management, you need monitoring. And AWS is really starting to deliver on some of that easy button capability for customers that they're going to need if agent AI is going to become a standard core feature of IT environment. 
 

William Fellows: Melanie, that goes right back to the sort of core DNA or principles of AWS and the developer focus because I think what AWS has done with the collection of things that you talked about there is to put the ball right back in the court of -- well, certainly Google and to some extent, Microsoft, which have quite diverse complex collections of tooling now that operate at lots of different pieces and parts. And there's really going to have to be some rationalization if those vendors are going to be able to capitalize on both their investments in and the design development of these -- of their AI infrastructure -- excuse me, their AI software offerings, which are sort of taking a leadership position in the industry. So, it's good to see AWS coming back with a really consistent core set of capabilities here. 
 

Eric Hanselman: I can really say that we're looking at suite day for a lot of the application side agentics development capabilities that are being pulled in here. 
 

William Fellows: Yes, for sure. 
 

Eric Hanselman: Sure. So, if we think about really some of the takeaways for where we've headed, I guess one of the things I wanted to think about in terms of the interpretations of what we heard, yes, we've gotten a multi-cloud. -- yes, they're being able to manage a lot of the capabilities of Agentic. What are the things that we're looking at it, we saw some of the more impactful aspects. I mean, what's the -- what are the kinds of things that when you think about really that belong together in a lot of these capabilities? 
 

Scott Crawford: I mentioned in my earlier comments about the Bedrock Agent Core, which is sort of a foundational implementation that's designed to abstract a lot of the deployment and operational management issues that Melanie spoke of. separately from agent code at the agents themselves. And that's becoming increasingly apparent, that abstraction in other areas, most notably around the sheer number of MCP servers we've seen appear throughout the industry just in the last several months since MCP has been much a thing really, but a lot of functionality there in agent core that I think really speaks to -- kind of plays to Amazon's strengths like both Melanie and William were both saying, it's appeal to builders. So, playing to its strong suit in that appeal to builders. A couple of years ago, it didn't have some of the same sort of strengths that some of its direct competitors brought to the market by building its own generative AI models or I should say, generative AI models early on compared to some others, didn't have exclusive partnerships with those foundation model providers. So, playing to its strengths as a place where builders come to build out their cloud applications. which is kind of an evergreen appeal of the AWS platform. These are the things that it seems like AWS has really capitalized on and talked about a lot at Reinvent. And Agent Core is an example of that deployment, operational monitoring, policy, also some agent functionality to make sure that your deployments operate as expected. So, in security, we saw the introduction of the AWS Security Agent, which does automated assessment of the security of cloud-native applications deployed on the AWS platform. The DevOps agent, similarly, but looking more at operational integrity, reliability functionality for DevOps teams. So those aspects are playing to its strengths as the -- as the great attractor for builders of cloud-native applications. Those are things that were really in evidence in my view at re: Invent. 
 

William Fellows: It's also worth throwing into the mix here the advances that AWS has done in improving the ability for customers to understand what they're paying for this. In other words, it's introduced a new level of transparency in cloud financial management by changing the way that it organizing or supports the ability for partners to own and operate the billing account on behalf of customers. So, they've managed to engineer a way that now a customer has full transparency into the entire cost while at the same time, leaving room or the ability for partners to do this on behalf of customers, but without making that arrangement opaque. In other words, everyone can see what is going on. And that's been -- that's a really important change, especially for the -- as cloud bills rise, the importance of service providers and integrators and so on rises. It means that folks couldn't really get to grips with how they manage their costs. Now that things like the account spending is tied to marketplaces, so customers can choose to use up some of their commitments by spending on AWS on the AWS ecosystem. This is really becoming such a significant piece of how AWS is expanding its universe here. 
 

Eric Hanselman: Well, I mean, maintaining room for channel and partners in this ecosystem has been really important. I mean putting any roadblocks in there, cutting out what is that larger services community would be a no-no. 
 

William Fellows: For sure. 
 

Eric Hanselman: So, I was also interested in some of the models. And Scott, you touched a little bit about this and what's happening with Nova. And the thing I thought that was the most interesting in this was really the ideas around NovaForge, the point that now customers can go beyond just simply retraining and retuning models, but actually get access to the source data to be able to go blend the training data sets with their own data, which seem to be a really interesting step, helping organizations to actually now get to that next level of being able to build models that are much more focused around their specific training data. 
 

Scott Crawford: Yes. And that speaks directly to something that I mentioned earlier as far as 2 years ago; they hadn't been building their own models. They had an exclusive agreement with the model provider. And now having Nova Forge having the ability to do that does introduce them into that mix and also the ability to leverage that ability to add to data, add to training data. They announced that they were going to support, I believe it was 18 Open WA models in total. So, it's going to be a prevalent thing as far as choosing the models that work best for a given use case. But this capability with Novaforge is something that does introduce that point of variety in what organizations can use and choose to use from. And it also gives them a little bit better confidence in terms of not only the data to which the model -- on which the model has been trained or which is available to it in delivering output, which is can be a source of confidence for security teams, those concerned with data and content security, but also have greater control over the data that's exposed to those models. And yet again, another very important security concern among organizations. Actually, one of the most important concerns that they have is the exposure of sensitive data to generative models. So, these are a couple of points that, again, speaking to where security teams need to pay attention to AWS larger strategy initiatives put forward at reinvent. This is one of them and something that if organizations are looking to do bespoke training of their models that this is something that might be able to meet their needs and meet the needs of security teams who increasingly need a seat at the table in terms of there's been so much deployment that you kind of classify as FOMO in motion. Most organizations, even if they don't have a solid business case for where they're going with their models, they're spending and deploying them anyway because they can't afford not to or can't afford to be seen as not doing that at this point. So, this is something that I think will have appeal to security teams and having a little bit greater control over that data and that exposure in those environments. 
 

Eric Hanselman: Yes. That makes a lot of sense in terms of where that's headed. Well, one of the things I want to get from each of you, though, is given that we're here in the holiday gifting season, what would you like to see a turn up or wrap in your particular holiday of choice from the reinvent portfolio of capabilities. So, Melanie, why don't we start with you? 
 

Melanie Posey: Well, I think for a lot of enterprise customers out there when it comes to enterprise generative AI, all they really want for Christmas is digital sovereignty. So, AWS has spent a lot of time at Reinvent talking about digital sovereignty, particularly the AWS European sovereign cloud. And I think this is a really important area for AWS to lean into because, again, its what customers want, and they're working back from customer requirements as the guiding force of product development these days. And so, I think it's something that folks at AWS are pretty excited about. And I think it overall will help diminish some of the I don't exactly know if it's mistrust, but maybe apprehension about leveraging cloud platforms for generative AI initiatives. So having a sovereignty framework in place is definitely a great Christmas present for regulated industries outside the U.S., and it's also going to be a great Christmas present for AWS in terms of being able to participate on a parity level at the very least, hopefully more so with its hyperscaler competitors out there in the market who've been pounding the sovereignty drum for quite a few years now. 
 

Eric Hanselman: And environments when you may not have to part with a lot of Hanneka Gelt in order to actually get that capability all tied up with a beautiful bow that is going to keep your regulators happy. 
 

Melanie Posey: Absolutely. You have to take care of them at Christmas for sure. 
 

Eric Hanselman: All right. Who's up next? What other gifting opportunities came out of? 
 

William Fellows: I'm happy to. This is William. So, I spoke earlier about the billing data and reporting access changes. Well, following on from that, what is really required now is to actually leverage some of this AI and agentic technology to provide better cost forecasting on the basis of this new transparency and also extending the kind of time horizons for that so that you can do some looking over the horizon planning in terms of where this is going. And it's interesting if you then throw in the multi-cloud piece here as well, to what extent you're going to be able to now include that as well and what opportunity that provides to the rich ecosystem of FinOps folks. So, if I'm on the finance and the operations side, those things, I think I would like to come wrapped up in a bow for me. If I'm at the other end, if I'm an engineer or a developer, I think the build-out of the platform engineering kind of offering and how those things are curated, especially using the Kiro autonomous agent as the kind of front-end metaphor, if you look, I went and spend a bit of time with that at the show, and I was really blown away. I thought, well, if you could repeat this, cut and paste across some of these other disciplines, which still remain a bit complex and confusing, this is really going to help. So that would be my other choice for having something wrapped up and put under the tree for me. 
 

Eric Hanselman: Well, you make a good point with Ciro in that it may be a bit of a heartbreaker for some of the other AI-assisted ID and development environment vendors that are out there. But yes. 
 

Scott Crawford: We've seen among security practitioners actually a fair amount of positive feedback about CRO04. So that was really interesting, kind of arresting to me that it's gotten out there so quickly and received such a favorable hearing. I would like to see something similar just for shopping for the collection of services you have to put together in an integrated whole to achieve what you're trying to achieve as a builder. That's always been a challenge with a whole set of AWS offerings, how do we weave them together? How do you make the best use of them without knowing the entire catalog. And of course, their services teams are very good at helping connect customers with that. But that's always been a challenge, particularly with digest for analysts suggesting the announcements that come out of Reinvent. So, something like an agent that can help you find the pieces that weave all those components together, like Melanie, you mentioned initiatives towards European sovereign cloud. That's definitely a high priority for security teams. And there's functionality like they announced actually just before Reinvent, Identity Federation for external services with Java web tokens. And that could be really a very practical tool for authenticating multiple partitions for things like sovereign cloud initiatives if you know it's out there. So having something like an agent that can help you do your shopping pull services together. And then as far as their own services go, they announced an MCP server, the AWS API MCP server on marketplace a few days before Reinvent. And not just like 3 or 4 days later, they announced the AWS MCP server, not just API, but combines capabilities from the API, MCP server and the AWS knowledge servers into a unified interface, effectively deprecating the announcement of 4 days before. So now you got to kind of not only keep up with announcements as they come up, but as things come out that affect those very announcements within a few days, keeping up with that is a real challenge. In fairness, AWS, they're very big fail fast, develop quickly, implement things, get them to market, see if they succeed or not. That's always been a value that builders have valued about them as well, but it's very challenging to keep up with all that. So, an agent that can string together the full suite of AWS offerings for the sake of building a more coherent whole, I think, would be very welcome among practitioners. 
 

Eric Hanselman: All right. And Henry, you can close this out. What's the final gift you'd like to get? 
 

Henry Baltazar: So, I think in this case, I think Amazon did deliver a little bit early for the holidays this time around. I mean the event itself felt like it was a bit of a victory lap for the S3 storage team. Because if you think about it, just a year ago, they introduced S3 tables and in their keynote, they admitted they were afraid when they launched that. They were thinking, hey, should we be adding this capability on a storage level? And I think the positive response they got from that. And since then, all their competitors in on-prem and cloud have been looking to add those capabilities. Because of that positive response, I think that's going to unlock more of their innovation. I think case in point in terms of what they delivered this year, is just this summer when they announced the S3 vectors capability. And that's already gone on to field over billion queries in preview mode at the event, they announced the general availability of it and the expansion of that service from -- I think they had like maybe like 5 preview regions before. Now it's going up to 14 preview regions. But if you start looking at this, they're adding these capabilities I think it's going to make their infrastructure more compelling and add more value because they want to add some of these data intelligence capabilities, indexing capabilities, all this data intelligence type things into the data layer itself, where the data is residing. So, you don't have to drag that data out to analyze it and do these different things. And I think as we were talking about earlier in terms of how they make things easier and how they can make things more compelling, adding that value at the top and making that part of storage is really going to make it more compelling for people to build on top of that as well. So, I think that's good news for them, and it's going to be bad news for your competitors that can't keep up. So, it will be an interesting time seeing what happens in 2026. 
 

Eric Hanselman: Well, hey, they're certainly storage obsessed. And as they're focusing on all of the data pipeline and data capabilities, certainly a big part of that. Well, thank you all, and there's so much more that we could discuss on this, but we are at time for this episode. So, thank you all for being back on. 
 

Melanie Posey: Thanks a lot, Eric. Always a pleasure. See you next time. 
 

Scott Crawford: Thanks, Eric. 
 

Eric Hanselman: And that is it for this episode of -- Next in Tech. And as we exit the year, I want to give particular thanks to our production team, including Sophie Carr, Feranmi Adeoshun, and Kyra Smith on the marketing and events teams. Also, it's a wrap for 2025 -- we're going to be taking our usual end of the break. We'll be coming back with our 250th episode as we start out 2026. I hope you'll join us then because there is always something Next in Tech!