AWS walked into Las Vegas with a full house of cards. The cloud titan absolutely unloaded with a series of announcements spanning silicon, software and networking at its annual re:Invent conference. And the key takeaway? It’s not playing catchup on AI anymore…it’s leading the pack.
For those new to the scene, events like re:Invent – and their accompanying deluge of news – can be a LOT to take in. That’s even more true for those who attend in person, especially given distractions like the onsite tattoo parlor, giant slide, barber shop, haunted house and more.
But while in town, Fierce sat down with a series of experts to help sift through the noise and home in on the rollouts that really matter.
Diana, Fierce Network Editor in Chief Elizabeth Coyne, Moor Insights and Strategy's Jason Andersen and Tekonyx Founder Sid Nag dive into AWS' big announcements around AI factories, APIs, Interconnect and more.
Catch the video at top, listen to the audio edition on the go or read through our transcript below.
New this season: You view this and future episodes on YouTube right here!
This podcast is written and hosted by Diana Goovaerts. It is edited by Diana Goovaerts and Matt Rickman. Liz Coyne is our executive producer. Special thanks to our guest, Nvidia SVP of Telecom Ronnie Vasishta, and to Nokia for providing a clip of their Chief Technology and AI Officer Pallavi Mahajan.
To learn more about the topics in this episode:
- AWS steals telcos' thunder with debut of AWS AI Factory
- Network APIs could find new life with help from AI agents
- AI is forcing a multi-cloud reckoning
- AWS debuts new Frontier Agents. Amazon workers aren't thrilled.
- AWS re:Invent: A tattoo parlor, barber shop and legit AI vision
- Microsoft preps cloud for AI with Corning, Heraeus hollow core fiber deal
This transcript has been lightly edited for clarity.
Diana Goovaerts (DG): AWS walked into Las Vegas with a full house of cards. The Cloud Titan absolutely unloaded with a series of announcements spanning silicon software networking and more at its annual reinvent conference. And the key takeaway, it's not playing catch up on AI anymore, it's leading the pack. I'm Diana Gobert and this is the five nine.
For those new to the scene events like re:Invent and their accompanying deluge of news can be a lot to take in. That's even more true for those who attend in person, especially given distractions like the onsite tattoo parlor, the giant slide, the barbershop, the haunted house and more. But while in town Fierce sat down with a series of experts to help sift through the noise and hone in on the rollouts that really matter.
And with me to walk through those today is our editor in Chief, Liz Coy.
Liz, it's so great to be back on the pod with you.
Elizabeth Coyne (EC): It is Diana. We're having a little bit of a Lunch Ladies reunion for those who know what Lunch Ladies is.
DG: I know I miss that show so much, but I'm so glad I'm back here with you now.
EC: Me too. We need to do this more often. For sure.
DG: Yeah, absolutely.
EC: So, let's get down to business. I know you were just out in your favorite place, Las Vegas, uh, for AWS re:Invent, and you sat through the keynote session, you had a bunch of briefings. So, I wanted to get some of your key takeaways from the show. Um, I know you spoke with AWS's telecom chief while you were out there. Did he say anything interesting?
DG: Yeah, actually, so, uh, his name is Jan Hoffmeyer and I had a really interesting conversation with him on a couple fronts. Um, some of the things that stood out to me was. Uh, we got on the topic of telco APIs, right? We know telco APIs have been a hot mess for quite a while now. I know that I wrote about them being a hot mess after I think it was DTW Ignite back in 2024. And part of the problem with telco APIs is it's twofold, right? Because for telco APIs to really take off, you need a developer community and also for telco APIs to work, they have to have the same function on every network, and that's hard to do when networks are at vastly different stages, right? There are a couple people still lingering in 3G, but mostly people are on 4G. But then in, even in the 5G realm, you have people with standalone and non-standalone. So it's hard to get everybody on the same page and also things are architected differently. One of the things Jan said to me that I found really interesting is he actually thinks AI agents can help with that.
So the way his vision works is that the AI agent, um, would be out front and it would be liaising with agents from other operators. And then behind that would be the model context protocol server, which is basically just the thing that allows agents to get the information they need from other networks. And behind that would be the telco API. And the whole idea behind the agent being the point person, if you will, is that they're a little bit more flexible.
They can reason, whereas the API is kind of a static thing and it can't.
He seems to think that letting agents kind of talk to each other might make that work in a way that's smoother. Granted, you still need the API, right? Because you need to be able to call the network, but I'll be really interested to see how that plays out.
There was one other cool thing, Liz, that I, I think you're gonna wanna hear about. Do you wanna hear about it?
EC: I do, I do.
DG: Okay. It was something called network language models, and it was something I hadn't heard about, right? Like we've heard of SK Telecom doing some stuff with LLMs,right? We've seen DT and SK telecom partner, but what he was talking about is really a small language model rather than a large language model.
So not only are you custom training the model to be pointed directly at the network and at an operator's unique network at that, but you're also condensing it, so it's only focused on the network.
So the idea behind that is that that kind of model will be more easily deployed on the infrastructure that's out there because it's small, right?
It takes up less space and also because it is trained on the intricacies of each operator's network, it'll actually be more functional. And my thought based on what he described is maybe that could make autonomy really become a thing. Like telcos have been very hesitant to make things autonomous, in part because they think that AI doesn't understand the network.
In theory, this could solve for that, but you know we'll have to see.
EC: For sure. I mean, you know, the telcos are always saying there still needs to be a human in the loop, but what Hoffmeyer is saying is maybe there doesn't if we have these small language models.
DG: Yeah, eventually. I think for the time being, you would need a human in the loop just to at least, uh, train the model even farther, like fine tune it. But yeah, I see a very interesting potential in that idea. And the cool part, Liz, is that he said. Because AWS just rolled out a new tool that will speed along not only condensation, but also fine tuning that they could come up with these models in a couple weeks to a couple months.
So like, you know, what's coming up in a couple months, don't you, Liz? Mobile World Congress.
So I will be watching to see if we hear more about that.
EC: Yep, for sure. So, I think you also had some big announce, or I think AWS also had some big announcements around interconnect. Can you talk a little bit about that?
DG: Yeah, that was one of the few things that grabbed my attention. So it kind of came before the keynote, so it got buried, especially under all the AI stuff.
But AWS Interconnect was something that they announced in collaboration with Google Cloud. Google Cloud for the record has been working on multi-cloud for quite a while, but what it and AWS have done is come up with an open API that allow their clouds to natively connect on the networking layer.
And so, what this means in normal people language is that it makes basically the connecting of the networks together, the connecting of the clouds, the physical infrastructure, it makes it like click and you're done.
EC: Wow.
DG: Yeah, you can go onto a cloud console and just say, I need X amount of bandwidth and I need it at this time, and you can click and be done. That seems like a huge step forward in terms of the multi-cloud journey that we've been on. And one of the other things I should note is that AWS said that they're going to be extending this to Microsoft Azure.
When I talked to Google Cloud about it, they were less definite on that front. So I'm not sure, but we'll see. But I did speak to Tekonyx Founder and Chief Research Officer Sid Nag a bit about why it matters – beyond just my thoughts – for the multi-cloud movement and AI. So here's what he had to say.
Sid Nag (SN): There's been this talk about multi-cloud forever in the industry, right?
But what we, what we noticed is that enterprises have a difficult time trying to deploy multi-cloud. So while it's a desirable end state, it's very hard to get to it. So, what the cross cloud architecture talks about is essentially how do you stitch all these different clouds together in sort of a layered cake, starting with the networking layer and where these clouds natively connect to each other, which is important, right?
Because we've had solutions in the past where clouds will connect to each other through some sort of exchange provider of sorts. And it's kind of klugy, right? In the telco world, I would, I would characterize it as a giant PBX, right? If you know what a PBX is, I'm sure you do, but, you know, but how do you actually connect these clouds without the middle sort of entity through native APIs, through native IP protocols without the need for the whole DYI approach with manual circuits and co-location routers and long lead times inconsistency in operations. So that's something that, uh, that, uh, is important to get rid of that stuff. And I think the AWS Cross Can Interconnect solution has
You know, it's a very open architecture, open API specification. They've also published this on GitHub. So any provider or partner wants to adopt this standard and sort of cross connect clouds. So I think it really helps, you know, on a number of fronts. AI and analytics across clouds become much easier. Connecting GPU clusters is a great example of that. Low latency pipelines, everything in a secure model.
And then, you know, you have all these application architectures, right? Apps that split across clouds where you know, you have the notion of composable applications today. How do all these different elements and composable applications that may be running in multiple clouds connect with each other?
There is an example or a use case of where, you know, you may have data sitting in one cloud and the application sitting in another. So all of that connectivity becomes extremely painless and seamless.
And I think AWS interconnect is indeed the right sort of offering to make that all happen.
EC: Well, that's super interesting and some great insight from Sid. Um, you also wrote up a story about AI factories and whether or not AWS was trying to steal the Telco's thunder, um, and forcing them back into that dumb pipe situation that we've heard about for decades.
So tell me more about the AI factories.
DG: Yeah, so I remember sitting there going, “Oh my gosh. They just did what the telcos have been trying to do for the past, you know, year plus.” You know, we've seen a bunch of operators come out with this. I can think off the top of my head of conversations I've had with Telenor, of press releases I've seen from like Swisscom. The whole idea behind telcos doing this is that they, like you said, they don't wanna be the dumb pipe.
You know, they ran into that problem with all the OTT streaming services. They ran into that with the cloud era where everybody's using their networks to get to the cloud.
And so the idea behind AI factories is they were looking at using the infrastructure they have and the positioning of that infrastructure to meet emerging AI needs.
So whether that be edge, whether that just be distributed compute, but it seems like AWS has just swept in and delivered the entire package that they were envisioning. Uh, so I am not quite sure, uh, if this is going to be competition or collaboration.
But here's Sid Nag again to explain a little bit more about the AI factory solution from AWS.
SN: Tthat's what the CIOs of today are looking for: how do I take all these different technologies that are so discrete and put them in some sort of a pipeline of sorts that delivers that intelligence to me that I can then use to drive outcomes for my business?
So I think with AI factories, AWS has come one step closer to that, where they've taken, you know, all these different components of – think of it as a private AWS region with NVIDIA train, AI accelerators, high speed networking, their storage and database technologies, as well as full service integration with SageMaker, and the training stage in Bedrock and some of the implementation or production stage.
We keep hearing about this MIT study with 95% of AI implementations are failing and it's failing because people don't know how to stitch these things together. And I think this is, this is the step in the right direction.
EC: Wow, I really hope that you have some good news for telcos from the show. Um, or are we back in the dumb pipe situation again and once again, have the cloud providers swept in and done things way faster than the telcos could?
DG: Well, they definitely have done things way faster than the telcos. I mean, that’s one thing AWS is known for, right? Like Google Cloud's niche is R&D, especially around AI.
But AWS is known for moving fast and that actually could help telcos out because AWS just came out with an update to its AWS Transform product, which could actually help telcos eliminate the legacy mainframe and mainframe migration burden that has been kind of holding both telcos and large enterprises way back.
I sat down with Moor Insights and Strategies Jason Andersen at the show to talk about AWS Transform and whether and why it matters.
Jason Andersen (JA): I have a lot of experience with telco systems from some previous work I did at other companies and I know it's a huge burden because those systems are incredibly sensitive, right?
And they're expensive to maintain, often on some very old hardware where vendors are charging maybe predatory prices on support and maintenance. So, it's one of those things that's a very difficult problem to solve. What Transform does is it allows AI to look at the code base and begin to recommend ways to move it to more modern frameworks like, say, Java.
And that movement also kind of comes with the idea that also ending up in an AWS data center might be something of an opportunity, especially now with this edge capability where maybe there's a partnership between a telco and AWS to build applications and then have them managed on-prem somehow.
But I do think Transform's a very interesting technology. The one thing I will say about these solutions, 'cause I've been looking at code transformation solutions for a while now, is don't get too excited about the hype, right? The stories of thousands of applications being out moved in a couple minutes are really well tailored to that solution.
It is an architectural approach, so you can expect to see probably savings in the migration of, say, 60 to 70% for a solution as complex as a mainframe, which is still millions of dollars. And it just might be enough to get that business case over the line, whereas in the previous years, maybe a telco wouldn't have been able to do that because it's just simply too expensive to move it off and too much risk of course, as well.
EC: That's a really great clip from Jason. There was also some chip news at the show. Can you tell me a little bit about that chip news?
DG: Yeah, there was a really interesting tidbit. I guess more of an observation really, that I wanted to share. So the big news was AWS announced general availability of Tranium 3, and they gave the first details about Trainium 4. So I imagine Trainium 4 will come out, you know, for availability next year because follows the same pattern as Trainium 3.
But one thing I noticed AWS did not talk about its dedicated inference chip Inferentia on the keynote stage. Didn't come up at all.
So Trainium was originally designed for training workloads and Inferentia for inference. But I noticed Matt Garman during his keynote – Matt Garman is AWS's CEO in case you don't know – but during his keynote he was talking about how well Trainium performs for inference, which was really interesting.
So I am curious, well, whether it will eventually let Inferentia fade away. Will they just focus on Trainium to reduce manufacturing costs? Right. Because when you have two different chips, you have to pay them to build the two different sets of infrastructure to build those chips and plus having one chip, you get more scale than with two.
So I will be watching that. Granted, silicon isn't my particular expertise, but I think that's something to watch for sure.
EC: Yeah, absolutely. Might be good follow up story with the AWS for sure. Any final thoughts and observances from the show?
DG: Yeah. A particular little nerd out note that I have is a couple years back I was writing in the networking space about something called hollow core fiber.
Right? So, you know, fiber is just strands of glass and hollow core fiber is basically like a little glass tube and the idea is that it can help with transmission and distance.
I already knew Microsoft was working with this. I believe Masha Abarinova from our team wrote a story about it. But AWS has deployed it. I saw a sample of it on the show floor. It was really cool. I think I even have, I took a little video. You can kind of see it at the end of the little video that I took.
It looks like a little cluster of five little tubes altogether in a round case. It’s just really interesting. So I am definitely looking forward to following up with them on that. But yeah, that was my little nerdy thing. And I guess the only other cool thing I did was I went down a slide instead of an escalator.
That's all for now. Make sure you like and subscribe, and we will see you next week.