Next in Tech | Ep. 285: Resilient Digital Infrastructure

View Full Transcript

[00:00:00] Eric Hanselman: Welcome to Next in Tech, an S&P Global podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, chief analyst for industry research at S&P Global, and today we're talking about resilient digital infrastructure. When we start thinking about what's really needed to support a modern digital enterprise, we've started to look in new and different ways at really what's required and what the capabilities are, and what really has to be brought to the table to effectively support them.

[00:00:33] Eric Hanselman: And with me to discuss it are returning guests Melanie Posey and Jean Atelsik. Welcome back to the both of you.

[00:00:40] Jean Atelsek: Thanks.

[00:00:41] Melanie Posey: Hey there!

[00:00:43] Eric Hanselman: So, I guess before we get started in this, maybe talk a little bit about what resilient digital infrastructure means. And Melanie, you are heading up the resilient digital infrastructure team, so can you kick us off with a little bit of background on resilient digital infrastructure?

[00:01:02] Melanie Posey: Okay, sure. And thanks again for inviting me on the podcast, Eric, and I appreciate the opportunity to talk about this whole issue of resilient digital infrastructure. As we move from the era of cloud to the era of AI, a lot of the requirements start to change. And as organizations start to automate a lot of their workflows and in some cases automate what gets delivered to customers and partners, the need for resilience is more elevated in the sense that if your business runs on AI, the AI needs to run. Now, lest anybody think that AI, like cloud, is just something out there in the ether, and there are no wires and cables attaching everything to a power source and to a network, you would be wrong if you think that. So, the resilience part of it is making sure that all of those pieces come together in the way that they're supposed to, they're optimized, they're secured, and they're fit for purpose. I think one of the key characteristics of the AI is that there are lots and lots of these components. You don't just have the infrastructure layer. You've got runtime layers, you've got the specialized TPUs, you've got the AI models, you've got all the tools, you've got the platforms, you've got training in the core, inferencing at the edge. And that's a lot to put together, A. And two, it's a lot to make sure that it stays operational and up and running. So that's why we brought all of these elements together. Cloud, networks, AI infrastructure, inferencing and training, unified control planes, all of that is part and parcel of resilient digital infrastructure. And it's not optimized, it's not agile if you don't have something on top managing and orchestrating and automating all the different pieces. And that's where the unified control plane comes in. And in some sense, that's where you start with digital infrastructure, going from top-down rather than bottom-up.

[00:03:53] Eric Hanselman: So, this really is a coordination on a level that I guess maybe we should have always had but maybe didn't. But that now when we start operating at the scale that we start to get to with AI and agentic operations, with a lot of the coordination that has to happen for, what is a modern application, we've got to be operating at new and enhanced levels now that we're actually running maybe more mission critical.

[00:04:25] Eric Hanselman: But I don't know, we've certainly had to run mission critical infrastructure in the past. But the other thing I'll point out is we're actually now starting to get into a world in which there is a regulatory requirement for resilience. You look at the UK's DORA Act, now we've got, I guess some, some regulatory requirements that are looking to enforce the things that we should have been doing, historically.

[00:04:49] Melanie Posey: Absolutely. And that's one reason why this unified control plane is so important, that you can't manage what you can't see. Therefore, observability wrapped up into unified control planes that have management, orchestration, autonomous or somewhat autonomous provisioning. All of these things need to work in concert in order to ensure the resiliency of those mission-critical workloads, which have been running in cloud for almost 20 years now. But what AI does is expands the scale, expands the scope, and amplifies the speed at which everything needs to happen. And for that reason, it's really important to have A comprehensive view of what your resilient digital infrastructure looks like. And we've talked about this on the show multiple times, that the days where people were on premises or 100% cloud are long gone.

[00:06:03] Melanie Posey: It's a hybrid world. It's a multi-cloud world. And guess what heterogeneity introduces into any equation? Complexity. It’s all about, how do you reduce the complexity of setting up the infrastructure, changing it around if need be, optimizing it for specific workloads, expanding it to new geographies? How do you take all that complexity and shrink it into something that is actually manageable? And then beyond that, how do you ensure that the various regulatory requirements that you mentioned, Eric, are in place? Your industry regulations like DORA and then our old familiar friends, HIPAA and all the rest, in addition to EU sovereignty regulations, which are coming out of Brussels fast and furious these days. All of that needs to be essentially programmed in to reduce the complexity of complying with all the regulations alongside your typical, usual IT of uptime, making sure that there are no security breaches, making sure that the bandwidth and the capacity is going where it's needed and when it's needed.

[00:07:42] Melanie Posey: So, a shorthand for resilient digital infrastructure would be really darn intractable or something like that. And this is why you need management and orchestration to bring it all together.

[00:08:04] Eric Hanselman: Tt's something that I know from my side, on the security side, we've been starting to shift that conversation from just sort of raw security terminology to resilience for a while. I wrote a piece actually now a couple years ago about how we needed to think about security resilience, but now we're really bringing that to infrastructure.

[00:08:23] Eric Hanselman: But it's been top of mind because this last week the three of us were actually out at various infrastructure related events. Melanie, you and I were over at Equinix's Horizon event, and Jean you were at Broadcom's Explorer. And Broadcom certainly has a take on infrastructure and its capabilities.

[00:08:46] Eric Hanselman: I’m curious what you saw there, and, are they looking for, you know, darn resilient capabilities?

[00:08:56] Jean Atelsek: Well, I mean, speaking about a unified control plane, obviously for the environment, as far as Broadcom is concerned, the only possibility is VMware Foundation, which is how the company is going to market. And that includes, virtualization, it includes, storage, networking, Kubernetes and management, all of which were world-class tools that VMware, had built. But after Broadcom acquired VMware, they rationalized the portfolio, they cut out a bunch of stuff, they reduced the SKUs from nine thousand to nine, and they jettisoned a bunch of partners. And the vision is for this VMware Foundation to be the place that you orchestrate and manage your IT and now AI infrastructure. So, at the conference, there were, you know, a bunch of announcements, but really what they were focusing on is that they have the AI factory, VMware AI factory, and that includes VCF at the base. They have added some security features to vDefend and Avi, the web application firewall and the network load then on top of that, they have an agentic layer. In terms of resilience, their point is that in terms of cost, when you own the GPUs and you're creating tokens as opposed to getting them from an Anthropic or an OpenAI and getting billed based on usage, that is automatically more sovereign, more secure, and more cost-effective. Cost maybe containing because actually both of you have spent a lot of time on the FinOps side in the broad parts of infrastructure. And now, as we've talked on previous episodes, we're starting to see that same cost containment concern crop up particularly around AI and agentic

[00:11:08] Jean Atelsek: Particularly around agents that go and burn, they burn tokens when you're sleeping. And so I would say that an interesting thing to come out of, the VMware Explore conference in terms of the resilient infrastructure is they, Broadcom, is doubling down on saying, "We're a technology company. We are not a services company. Our expertise is in virtualization." That also includes that a feature of VMware AI Factory is multi-tenant, model as a service. So, the different business units can consume models without having to dedicate a GPU workload. And they were saying that their partners, their remaining partners, are the supply chain that delivers that.

[00:11:58] Jean Atelsek: So, they're not too concerned about the sovereignty stuff. They're like, "We supply a stable, manageable, secure, foundation and all the partners are going to deal with the integration and everything else with the outside world."

[00:12:17] Eric Hanselman: Well, and again, if you are controlling the environment in which all this operates, presumably that's the way that you achieve sovereignty because you know where all of your stuff is. Not this, you know, foofy cloud stuff where it could be any-all where. But yeah, again, that whole private AI kind of aspect has been something that's been a strong part of their story, really for quite some time now.

[00:12:41] Jean Atelsik: Yeah. And I think that given that a lot of folks renewed their VMware contracts after Broadcom made these very dramatic changes to the licensing and the pricing, that multiplied what they would have to pay. VMware is setting the table. They were talking about the roadmap.

[00:13:00] Jean Atelsik: It's not fully formed yet, this platform, but I think the plan is to have it in place by the time those contracts that were renewed in '23 and '24, by the time they come up for renewal, they will have a robust private AI offering that will make buyers think twice about abandoning,

[00:13:25] Eric Hanselman: That it will be a compelling enough story. And the other announcements that I also heard were a reiteration of a lot of the focus on cost management, in what is now a supply chain issue that I think everybody's been seeing, which is the cost and availability of memory has become huge. And so, managing memory within an environment is important, and they've been bringing back that, that key point that in fact they can help you manage memory much the same way they manage virtualized resources like CPUs and GPUs.

[00:14:00] Jean Atelsik: Right. In VMware Cloud Foundation 9.1, which came out several months back, they introduced NVMe tiering, which lets you offload what would otherwise require DRAM, which is in very short supply and very expensive, onto other tiers as a big cost saving, and a way to get more out of your current, assets than instead of having to buy new, which is proving very tough that's expected to be through, 2026 and into 2027.

[00:14:35] Eric Hanselman: Well, I guess they have a claim on some part of that universal control plane that's a key part of resilient digital infrastructure. Melanie, you and I were at Equinix, and they are also putting together pieces of a control plane capability in a couple different areas, mostly around connectivity. But that’s another important aspect of building out infrastructure. You got to be able to tie it all together, yeah?

[00:15:06] Melanie Posey: Exactly. And tying together all that infrastructure you've got multiple networks is problematic, and ensuring the security and the performance of all of those links is equally complicated. And what Equinix is doing is not necessarily dissimilar to what we're seeing from Dell and HPE around how you can build an infrastructure stack using agentic controls for putting all the different pieces together, from design and deployment all the way through to observability, collecting the telemetry, opening up tickets, remediation, diagnosis on what's wrong. But with Equinix, the interesting thing is they're doing this for the networks inside the data center and also inside the data center to far-flung points of Equinix's distributed platform. And one of the really important things around this is interconnection, being able to make these links between the different partners and customers that have chosen to co-locate their infrastructure in Equinix's data centers. And the ability to do this autonomously on the fly means that of the different capabilities that are out there in the AI ecosystem, when customers need them, they are available to them. And when opportunities come along to partner with other companies inside of Equinix data centers, you don't have that manual provisioning where you've got somebody in a cage and they're screwing things into racks and then testing it to make sure it all works. You've taken out the manual aspect of that. So basically, the infrastructure that you have inside of an Equinix data center can start realizing time to value much faster

[00:17:38] Eric Hanselman: So, it's something where, again, one of the things about colocation providers and Equinix in particular with its exchange model, they are the meeting point where all of these services can potentially come together. Even though we've now seen the cloud providers themselves start to get into the interconnect game, AWS Interconnect, they had announced initially that they were able to build connections dynamically, with Google Cloud Platform.

[00:18:06] Eric Hanselman: They just announced that they've extended that to Microsoft Azure. Everybody's got a piece of this, and I guess the question is, you know, from the Equinix side this is one more way to automate that process, to connect everything that is outside of your cloud to the interconnect capabilities that now can link your various cloud environments together.

[00:18:28] Melanie Posey: Yeah. And it's an interesting dynamic that Equinix has always had with the cloud providers. It’s what we used to call co-opetition, right? That, you're its customer, but it's also a competitor. And that's the thing to keep in mind about Equinix's business model is a big part of it is enabling other members of the AI ecosystem to set up and run their businesses in an optimal fashion. So, when you have a situation like that, some degree of co-opetition competition is inevitable. But one way I would look at it is Equinix, as part of their value proposition, says they're a vendor neutral ecosystem meeting point. So, it's all about giving customers choice, right? You can use the AWS interconnect multi-cloud or Google's version of that, which is particularly useful for a company that's mostly in one cloud, and they take some capabilities or store some data in another cloud. But Equinix's audience here is different, and I think it increasingly reflects what organizations are starting to look like, especially with all this interest in private AI, as you and Jean were talking about previously. It’s the whole idea that I could build these links to all my partners myself, or I can be in one place where all of these people live and have that happen now in a programmatic fashion now that Equinix has introduced this autonomous networking capability. And it all goes back to the same old story around outsourcing, right? Like you can do it yourself or you can pay somebody else to do it. And I'm the kind of person who hates doing laundry, so I pay somebody else to do it. But I get clean clothes at the end of the day, and that's what it's all about, getting what you need. How you get it, who you get it from, maybe not quite as important.

[00:20:57] Eric Hanselman: I don't know what kind of universal control plane you use to get your laundry done, but there's a possibility... and actually, that's a good analogy to bring up the ecosystem part of this. Because the other thing that I thought was interesting, that Equinix was also talking about, or another of their announcements, was their inference exchange.

[00:21:16] Eric Hanselman: And here's a point at which you actually...the ecosystem becomes ever more important because you've got to connect to whatever edge capabilities are, wherever you're actually going to be putting your AI capabilities to work. And that's a point at which you may be, you almost entirely are going to be working with other ecosystem partners who are going to be providing some of the model capabilities for inferencing, maybe providing other aspects of, you know, the data that you're actually working with.

[00:21:46] Eric Hanselman: There are a number of different parts to all of this, but it's one more place where that control plane has to extend to give you the ability to automate, and Melanie, to your point, to get that observability, the telemetry to understand what's going on in those environments. You have to be able to see what's going on.

[00:22:06] Eric Hanselman: You know, you have to be able to look at that laundry ticket and be able to say, "Wait, my clothes are going to be ready which day?" and be able to, you know, track and ensure that, yeah, you've got something reasonable to wear when actually get to the end of the week. But that's this other part of it.

[00:22:21] Eric Hanselman: And that interaction, I think, from those ecosystem places, or the players in each of those environments, is going to be playing out in the various places in which you're actually trying to get work done. And Jean, you know, this is, Broadcom would certainly identify themselves as the major participant for some of this.

[00:22:39] Eric Hanselman: But they are, you know, increasingly in this world in which there are lots of different capabilities that are being put to work. Despite Hock Tan's comments that you really should be entirely on Broadcom, or you're paying a cloud tax that is going to be long-term untenable, it seems like the reality most find themselves in is this, you know, heterogenous environment in which you got to be able to bring this all together.

[00:23:04] Eric Hanselman: You got to be able to ensure that it all plays nicely together. And I guess, Melanie, that gets back to our universal control plane and that darn tractable environment that we're trying to build.

[00:23:20] Melanie Posey: Right. Exactly. A machine with multiple moving parts is going to be inherently difficult to keep in good running order than something that only has a few parts. But another way to think about this is an ecosystem approach. And it kind of means that if all of a sudden one of your ecosystem partners isn't working out for you, the autonomous networking that provides the connectivity within the data center allows you to make that switch pretty quickly. So, I think another element of resilience which we don't necessarily think of top of mind is supply chain, right? And if you think about an ecosystem as a supply chain, then, to the extent that your available alternatives are all good enough, you've got some supply chain resiliency built into the system. And I think that's the one takeaway I'd like to give to folks when you're thinking about resilient digital infrastructure is, think about it in terms of performance and manageability, but also think about it in terms of change management as well. How easy is it for you to switch out just from a sort of physical point of view, not so much getting into vendor lock-in and everything goes along with that.

[00:25:07] Melanie Posey: But that also is a component of sovereignty as well, that folks who have to operate their infrastructure and their AI operations under a sovereignty regime, or in some cases might fall under a sovereignty regime at some point, need to think about some of these issues around supply chain.

[00:25:31] Melanie Posey: How easy is it to switch? How easy is it to expand? See, Eric, we were talking about the edge. The edge is everywhere. Ask 10 people to define the edge, you'll get 25 different definitions, because most people will double back and say, "No, it's also this." I think resiliency, it casts a wide net, and it's a deep stack. And at the top of that stack to make sure you have visibility in everything, and you have controls over everything, that unified control plane is an absolute must-have piece of the puzzle

[00:26:12] Eric Hanselman: You have brought us right back to the definition of resilience at its core. So, this has been great. Appreciate both of your perspectives, and great to talk about a little bit of what we saw in the last week and what we've been seeing in the head. But we are at time for this episode.

[00:26:29] Melanie Posey: Well, thanks a lot, Eric. It's always great talking to you. Always a good conversation.

[00:26:35] Jean Atelsik: Really appreciate it.

[00:26:36] Eric Hanselman: Thank you both, but that is it for this episode of Next in Tech. Thanks to our audience for staying with us, and thanks to our production team, including Dylan Schieble, Sophie Carr, and Feranmi Adeoshun on the marketing events teams. If you liked this episode, please subscribe or like us. I hope that you join us for our next episode because there is always something next in tech.