Have we spent so much time talking about AI models, data, and agents that we've overlooked the networks carrying everything between them? In this episode of IT Infrastructure as a Conversation, I sit down with Jamie Pugh, CTO of Globalgig, to discuss why connectivity is becoming one of the most overlooked challenges in enterprise AI adoption and why network resilience could determine whether AI initiatives succeed or fail.
As organizations distribute AI workloads across multiple clouds, regions, and edge environments, network traffic patterns are changing dramatically. Jamie explains how traditional architectures were built for people consuming applications, while today's AI systems are increasingly driven by systems communicating with other systems. Agents, inference engines, data lakes, and cloud platforms are generating entirely new traffic demands that many existing networks were never designed to handle.
We explore how bottlenecks emerge when enterprises rely on single cloud on-ramps, fragmented connectivity strategies, or network designs optimized for yesterday's workloads. Jamie shares why bandwidth alone is no longer the primary concern, and how latency, carrier diversity, observability, and intelligent routing are becoming business priorities rather than purely technical considerations.
The conversation also examines the growing importance of predictive and self-healing networks, the role of network-as-a-service platforms, and why infrastructure, security, and AI teams can no longer operate in isolation. Jamie offers practical advice for organizations looking to audit their current network readiness and prepare for a future where AI-powered productivity depends on reliable, real-time connectivity.
If AI is becoming part of the infrastructure that powers modern business, what happens when the network underneath it can't keep pace? Join us as we discuss the hidden foundation of enterprise AI and why resilient connectivity may become one of the biggest competitive advantages of the decade. What role do you think networks will play in determining the winners and losers of the AI era? Share your thoughts with me.
00:00:00 --> 00:00:02 And if you are running a business right now,
00:00:02 --> 00:00:04 you may have noticed there's a quiet shift happening.
00:00:05 --> 00:00:07 One that most people are still underestimating.
00:00:08 --> 00:00:10 And that is, your company doesn't live inside
00:00:10 --> 00:00:14 your network anymore. It lives inside the browser.
00:00:15 --> 00:00:18 That's where your SaaS apps sit. That's where
00:00:18 --> 00:00:21 your data moves. And increasingly, that's where
00:00:21 --> 00:00:24 attackers are focusing their attention. So Nord
00:00:24 --> 00:00:27 layer has just launched its new business browser.
00:00:27 --> 00:00:30 and it's designed specifically for small and
00:00:30 --> 00:00:33 medium sized companies that need visibility and
00:00:33 --> 00:00:37 control without the overhead of enterprise security
00:00:37 --> 00:00:40 tools. What I like here is the balance. You get
00:00:40 --> 00:00:43 advanced protection, better compliance and full
00:00:43 --> 00:00:46 visibility into how your team is working online
00:00:46 --> 00:00:49 but without slowing anyone down or forcing them
00:00:49 --> 00:00:52 to learn anything new. It feels like a practical
00:00:52 --> 00:00:55 step forward rather than another security layer
00:00:55 --> 00:00:57 that adds friction. So if you want to see more
00:00:57 --> 00:01:01 about how it works please head over to Nordlayer
00:01:01 --> 00:01:04 .com slash browser and check it out and let me
00:01:04 --> 00:01:07 know your thoughts. But now on with today's show.
00:01:08 --> 00:01:11 What happens when AI ambition runs straight into
00:01:11 --> 00:01:14 the limits of the network? Well today I'm joined
00:01:14 --> 00:01:18 by Jamie Pugh CTO of Global Gig. We're going
00:01:18 --> 00:01:21 to enjoy a conversation that looks at the part
00:01:21 --> 00:01:24 of the AI story that many businesses still overlook.
00:01:25 --> 00:01:28 Yeah, this year I've spoken lots about agents,
00:01:28 --> 00:01:34 models, data, governance and ROI. But Jamie is
00:01:34 --> 00:01:36 going to shine a light on the connectivity layer
00:01:36 --> 00:01:41 that makes all of this possible. Now, GlobalGig
00:01:41 --> 00:01:44 helps multinational enterprises build composite
00:01:44 --> 00:01:47 global networks across carriers, countries, clouds
00:01:47 --> 00:01:51 and security environments. But he will describe
00:01:51 --> 00:01:53 this today as creating a single programmable
00:01:53 --> 00:01:57 one rather than leaving organisations with a
00:01:57 --> 00:02:00 patchwork of regional contracts, providers and
00:02:00 --> 00:02:02 blind spots. Because we're going to talk about
00:02:02 --> 00:02:06 how AI workloads are changing network traffic
00:02:06 --> 00:02:09 patterns. because for years enterprise connectivity
00:02:09 --> 00:02:12 was largely built around people accessing SAS
00:02:12 --> 00:02:16 tools and cloud applications. But as agents begin
00:02:16 --> 00:02:20 talking to tools and models and data lake and
00:02:20 --> 00:02:23 inference engines, not to mention more agents,
00:02:24 --> 00:02:27 traffic is becoming more distributed, bursty
00:02:27 --> 00:02:31 and system to system. So he will explain today
00:02:31 --> 00:02:34 why cloud on ramps, single carrier strategies,
00:02:34 --> 00:02:37 asymmetric, broadband links, poorly coordinated
00:02:37 --> 00:02:40 network designs, how all these things can quickly
00:02:40 --> 00:02:43 cause hidden bottlenecks. And he'll share why
00:02:43 --> 00:02:47 resilience now requires diversity, better observability,
00:02:47 --> 00:02:51 network as a service fabrics, SD -WAN and security
00:02:51 --> 00:02:54 policies that can adapt as traffic moves across
00:02:54 --> 00:02:57 regions and clouds. And we'll also discuss why
00:02:57 --> 00:03:01 milliseconds now matter to productivity, AI performance
00:03:01 --> 00:03:04 and business continuity. Because if people increasingly
00:03:04 --> 00:03:07 rely on AI tools to research, analyse, automate
00:03:07 --> 00:03:11 and support decision making... even the smallest
00:03:11 --> 00:03:13 connectivity problem can have a much, much bigger
00:03:13 --> 00:03:17 impact than they did in the old world of simple
00:03:17 --> 00:03:21 web access. So as enterprises race to scale AI,
00:03:22 --> 00:03:25 are they paying enough attention to the networks
00:03:25 --> 00:03:28 carrying the workload? I'd love to hear your
00:03:28 --> 00:03:30 thoughts on this one. But enough for me. Let
00:03:30 --> 00:03:33 me officially introduce you to the guests right
00:03:33 --> 00:03:39 now. So a massive warm welcome to the show. Thank
00:03:39 --> 00:03:41 you for joining me today Can you tell everyone
00:03:41 --> 00:03:43 listening a little about who you are and what
00:03:43 --> 00:03:46 you do? My name is Jamie Pugh. I'm the CTO for
00:03:46 --> 00:03:49 global gig You can think of global gig is building
00:03:49 --> 00:03:53 composite global networks for multinational enterprises
00:03:53 --> 00:03:59 We assemble underlays from around 200 or so global
00:03:59 --> 00:04:04 ISPs across 180 countries We layer on cloud and
00:04:04 --> 00:04:06 DC on ramps through network as a service fabrics
00:04:06 --> 00:04:09 and wrap it with SD WAN, SASE providing, you
00:04:09 --> 00:04:14 know, and SASE providing secure access and observability.
00:04:14 --> 00:04:18 So the result there is a single programmable
00:04:18 --> 00:04:22 WAN instead of a patchwork of regional contracts.
00:04:23 --> 00:04:26 And my role spans the technology platform, the
00:04:26 --> 00:04:29 SE organization, sales engineering organization,
00:04:29 --> 00:04:33 the AI direction and automation. Well it's a
00:04:33 --> 00:04:35 pleasure to have you join me today and when we're
00:04:35 --> 00:04:38 talking about things like network traffic it
00:04:38 --> 00:04:40 is one of those things that we I think we all
00:04:40 --> 00:04:42 take for granted when we're downloading I don't
00:04:42 --> 00:04:47 know a 4k video off Netflix or a hundred gig
00:04:47 --> 00:04:49 game download on Xbox, you look inside the average
00:04:49 --> 00:04:52 house, there's so much bandwidth being consumed
00:04:52 --> 00:04:55 and then now we're throwing in AI workloads into
00:04:55 --> 00:04:57 the mix. So I'm curious though, more recently,
00:04:58 --> 00:05:00 where are you seeing global networks struggling
00:05:00 --> 00:05:04 under modern AI workloads? Is there a big difference
00:05:04 --> 00:05:06 here? What are you seeing from your side? Yeah,
00:05:06 --> 00:05:11 I am. There's definitely a change. The traffic
00:05:11 --> 00:05:13 used to be really kind of designed. The architecture
00:05:13 --> 00:05:16 was designed around users consuming. In the last
00:05:16 --> 00:05:18 several years, we've seen that's been, instead
00:05:18 --> 00:05:21 of the data center hub spoke over MPLS model,
00:05:22 --> 00:05:25 it's been SD -WAN and out to SAS applications.
00:05:26 --> 00:05:30 And that's an interesting download pattern, where
00:05:30 --> 00:05:34 it's very download -centric. But now, traffic
00:05:34 --> 00:05:37 is actually being you know, sent between systems.
00:05:37 --> 00:05:39 So it's systems talking to systems. And when
00:05:39 --> 00:05:42 I talk about that, it's more like, you know,
00:05:42 --> 00:05:46 think about agents. So AI workloads are creating
00:05:46 --> 00:05:51 massive traffic now between the model to the
00:05:51 --> 00:05:55 data store, the agent to the tools, inference
00:05:55 --> 00:05:59 engines to data lakes. And that often spans across
00:05:59 --> 00:06:04 regions and clouds. Most architectures that we've
00:06:04 --> 00:06:07 built in the last 10 years weren't designed for
00:06:07 --> 00:06:10 that pattern. And the Cloud OnRamp is now what
00:06:10 --> 00:06:13 we see as the bottleneck. Let me expand on that
00:06:13 --> 00:06:18 a little bit. So the SD -WAN economics, we built
00:06:18 --> 00:06:21 a lot of networks saying, hey, we're going to
00:06:21 --> 00:06:23 replace these expensive NPLs and give you some
00:06:23 --> 00:06:28 diversity with some internet circuits. Typically,
00:06:28 --> 00:06:30 that was a DIA at a location that had symmetric
00:06:30 --> 00:06:34 bandwidths. And then we added in some broadband
00:06:34 --> 00:06:36 connectivity. And I think broadband was perfect
00:06:36 --> 00:06:38 for SAS. And it was a low cost option to put
00:06:38 --> 00:06:41 some resiliency in there. But what we're seeing
00:06:41 --> 00:06:45 now is that these agent to agent or agent to
00:06:45 --> 00:06:50 MCP, because right now on my workstation, for
00:06:50 --> 00:06:51 instance, I've got several agents that are doing
00:06:51 --> 00:06:56 research for me. When I go to research something
00:06:56 --> 00:06:58 like something that I want to purchase personally,
00:06:58 --> 00:07:00 it's now going out there and it's looking at
00:07:00 --> 00:07:02 about a thousand websites. I'd only look at five
00:07:02 --> 00:07:04 or six, right? But it's like looking at a thousand.
00:07:04 --> 00:07:07 So that's an interesting, you know, bandwidth
00:07:07 --> 00:07:10 characteristics that's changing. And as I start
00:07:10 --> 00:07:13 to do that kind of research and infer back to,
00:07:13 --> 00:07:14 you know, models that might be hosted in the
00:07:14 --> 00:07:17 cloud or hosted in my data center infrastructure,
00:07:17 --> 00:07:20 there's a lot of upload traffic going as well.
00:07:20 --> 00:07:23 And those broadband circuits, I think What we're
00:07:23 --> 00:07:25 seeing is they're not suited for that type of
00:07:25 --> 00:07:30 traffic at all. They're asymmetric, and their
00:07:30 --> 00:07:34 upload bandwidth characteristics are our best
00:07:34 --> 00:07:38 effort. So we're seeing where the SD -WAN is
00:07:38 --> 00:07:43 struggling with those circuits. So we're typically
00:07:43 --> 00:07:46 having to re -engineer them to think of download
00:07:46 --> 00:07:48 -only links, because they're causing they're
00:07:48 --> 00:07:51 causing some kind of degradation. Listening to
00:07:51 --> 00:07:53 your one example there, your personal use case
00:07:53 --> 00:07:57 of how your bandwidth has changed is a great
00:07:57 --> 00:07:59 example, especially when you think about scaling
00:07:59 --> 00:08:02 that across an entire workforce and everyone
00:08:02 --> 00:08:05 working the same, an entire state, even or even
00:08:05 --> 00:08:09 country, and continents around the world. So
00:08:09 --> 00:08:12 if we look under the hood at the IT infrastructure
00:08:12 --> 00:08:14 that makes this possible that start in the field
00:08:14 --> 00:08:17 of strain, when enterprises begin distributing
00:08:17 --> 00:08:20 AI workloads across regions, clouds and edge
00:08:20 --> 00:08:23 locations. What is it that breaks first in that
00:08:23 --> 00:08:26 network layer? And why are so many teams only
00:08:26 --> 00:08:29 just discovering this after deployment? Yeah,
00:08:29 --> 00:08:31 again, I think that goes back to that cloud on
00:08:31 --> 00:08:33 ramp. You know, that's where the apps are. That's
00:08:33 --> 00:08:36 where the inference is likely going to be. or
00:08:36 --> 00:08:40 also the data center on -ramp. Wherever the inference
00:08:40 --> 00:08:42 is going to be or the models are hosted, that's
00:08:42 --> 00:08:46 where it's going to be. Most enterprises still
00:08:46 --> 00:08:50 have that single on -ramp regionally, and sometimes
00:08:50 --> 00:08:53 the single on -ramp to the Cloud, period, and
00:08:53 --> 00:08:56 that's it. I think that worked great for human
00:08:56 --> 00:08:58 -paced application traffic, but AI workloads
00:08:58 --> 00:09:04 will create a sustained bursty traffic pattern.
00:09:05 --> 00:09:08 that will overwhelm that single on -ramp. So
00:09:08 --> 00:09:12 the way we design and the way I think the fix
00:09:12 --> 00:09:15 is, is to build multi -cloud network as a service
00:09:15 --> 00:09:18 fabrics with this elastic capacity component
00:09:18 --> 00:09:23 on both ends that allows teams to, you know,
00:09:24 --> 00:09:27 it allows the network to be more dynamic as far
00:09:27 --> 00:09:30 as, you know, Hey, I'm starting to see traffic
00:09:30 --> 00:09:32 patterns. I need to adjust and accommodate. You
00:09:32 --> 00:09:35 can do that easily across a NAS fabric where
00:09:35 --> 00:09:39 you can't do it with these static circuits that
00:09:39 --> 00:09:41 you have today. And before talking to you today
00:09:41 --> 00:09:44 I was reading that you've warned previously about
00:09:44 --> 00:09:48 the risks of relying on single carrier or fragmented
00:09:48 --> 00:09:51 network models. I also suspect that many people
00:09:51 --> 00:09:53 listening are waking them up at the moment to
00:09:53 --> 00:09:55 some of the risks around those bottlenecks. So
00:09:55 --> 00:09:58 can you walk me through how a connectivity failure
00:09:58 --> 00:10:02 in one region could cascade into wider disruption
00:10:02 --> 00:10:05 across AI systems just to bring to life the problem
00:10:05 --> 00:10:08 we're talking about here. Yeah, I think you know,
00:10:08 --> 00:10:10 there's a couple things that often get conflated
00:10:10 --> 00:10:14 here, you know fragmented is an interesting word
00:10:14 --> 00:10:18 because if it's poorly coordinated that's bad,
00:10:18 --> 00:10:21 but carrier diversity You know is also when you
00:10:21 --> 00:10:24 when you have you know You need fragmented networks
00:10:24 --> 00:10:27 in order to actually build that carrier diversity,
00:10:27 --> 00:10:30 right? But what you need is to strategically
00:10:30 --> 00:10:32 onboard them and orchestrate them through a single
00:10:32 --> 00:10:35 control plane Or have a provider that can do
00:10:35 --> 00:10:38 that for you And that's resilience, not fragmentation.
00:10:39 --> 00:10:43 So what we've seen is in a lot of instances where
00:10:43 --> 00:10:46 we take over somebody's network, they've bought
00:10:46 --> 00:10:51 what they thought were fragmented or diverse
00:10:51 --> 00:10:54 solutions. But in the end, there was some sort
00:10:54 --> 00:10:57 of component that was shared across them. So
00:10:57 --> 00:11:01 that could have been a last mile that was actually
00:11:01 --> 00:11:03 the same carrier. They just didn't know it because
00:11:03 --> 00:11:07 they bought from two different ISPs. or they
00:11:07 --> 00:11:10 share the same path or the same peering relationships
00:11:10 --> 00:11:15 upstream. And when a fiber cut happens somewhere
00:11:15 --> 00:11:17 in the network, all of a sudden they've got an
00:11:17 --> 00:11:19 additional 100 to 200 milliseconds of latency
00:11:19 --> 00:11:24 to get somewhere. And that characteristic shared
00:11:24 --> 00:11:29 across either surviving provider. So that's where
00:11:29 --> 00:11:32 you really have to get in and understand that
00:11:32 --> 00:11:35 these carriers nationwide, understand their peering
00:11:35 --> 00:11:39 relationships, understand their fiber paths so
00:11:39 --> 00:11:41 that you can design something that's more resilient.
00:11:42 --> 00:11:45 And latency and reliability are also increasingly
00:11:45 --> 00:11:48 becoming tightly linked to business outcomes.
00:11:48 --> 00:11:51 So on that side of things, how should infrastructure
00:11:51 --> 00:11:53 teams be maybe rethinking network architecture
00:11:53 --> 00:11:57 when milliseconds could directly impact not only
00:11:57 --> 00:11:59 AI performance, but the user experience as well?
00:12:00 --> 00:12:01 Should they be doing things differently or what
00:12:01 --> 00:12:04 should they be doing differently? Yeah, I think
00:12:04 --> 00:12:06 that, you know, there's some architectural shifts
00:12:06 --> 00:12:10 there, designing for predictable paths, not just
00:12:10 --> 00:12:14 available ones. You know, that's what we were
00:12:14 --> 00:12:17 kind of talking about a little bit before. So
00:12:17 --> 00:12:21 because the AI workloads are path aware, you
00:12:21 --> 00:12:26 know, the routing needs to account for not just
00:12:26 --> 00:12:30 the current performance, but also the policies
00:12:30 --> 00:12:33 that you're gonna be putting in place, you know,
00:12:35 --> 00:12:37 to support those AI workloads, right? I mean,
00:12:37 --> 00:12:39 to me, I mean, we've always done this, right?
00:12:39 --> 00:12:43 We've designed kind of these application performance
00:12:43 --> 00:12:49 profiles. But the problem now is that AI is bringing
00:12:49 --> 00:12:52 a whole new performance characteristic, and the
00:12:52 --> 00:12:55 inference is all over the map. So I don't think
00:12:55 --> 00:12:58 that all the tools that are out there today have
00:12:58 --> 00:13:03 all the profiles in order to be able to do that.
00:13:03 --> 00:13:05 So if I think about this personally, let me just
00:13:05 --> 00:13:12 get to a personal example. I am a big user of
00:13:12 --> 00:13:16 AI tools, right? And I've got subscriptions to
00:13:16 --> 00:13:21 Perplexity, to ChatGPT, to Claude. And I put
00:13:21 --> 00:13:29 a lot of my research and I guess a lot of my
00:13:29 --> 00:13:31 Excel work now, so a lot of my busy work, it
00:13:31 --> 00:13:35 actually goes into tools like Claude. And it's
00:13:35 --> 00:13:38 made me so much more efficient to the point where
00:13:38 --> 00:13:42 when there's a network issue that's keeping me
00:13:42 --> 00:13:46 from using those tools, I feel like I've lost
00:13:46 --> 00:13:51 my best employee. And my productivity is just
00:13:51 --> 00:13:54 shot. It's incredible now. I get frustrated so
00:13:54 --> 00:14:01 quickly. because of this and. Typically, in the
00:14:01 --> 00:14:03 past, I've been able to kind of say, you know
00:14:03 --> 00:14:05 what the Internet's down. I'm going to go find
00:14:05 --> 00:14:08 some busy work, you know, so I'll go grab an
00:14:08 --> 00:14:10 Excel workbook and I'll work on some formulas
00:14:10 --> 00:14:13 and I'll go drive in it or I'll I'll I'll take
00:14:13 --> 00:14:16 some downloaded content that I downloaded and
00:14:16 --> 00:14:18 I'll start reading it from a research perspective,
00:14:18 --> 00:14:23 but. Now that is so inefficient because I know
00:14:23 --> 00:14:25 as soon as the network comes back up that the
00:14:25 --> 00:14:27 Excel work will be faster So it's just wasted
00:14:27 --> 00:14:29 time for me to go in there and actually call
00:14:29 --> 00:14:34 that busy work and then the The downloaded material
00:14:34 --> 00:14:36 that I might have read is probably already aged
00:14:36 --> 00:14:38 out because you know, everything is happening
00:14:38 --> 00:14:41 so fast So I'm my mind wants real -time data.
00:14:41 --> 00:14:45 So You know, I feel like this is going to be
00:14:45 --> 00:14:48 a new frustration point across the enterprise
00:14:48 --> 00:14:52 as you know, as people have poorly architected
00:14:52 --> 00:14:54 networks and they're sitting in an office or
00:14:54 --> 00:14:58 maybe they're on ramps through their. their endpoint
00:14:58 --> 00:15:01 agents that are taking them into the VPN infrastructure,
00:15:02 --> 00:15:04 and they're having issues with that. I think
00:15:04 --> 00:15:06 the frustration levels are going to be much higher
00:15:06 --> 00:15:09 now going forward than they have been in the
00:15:09 --> 00:15:12 past. Just because people get so reliant on having
00:15:12 --> 00:15:15 such a good set of tools that can help them accomplish
00:15:15 --> 00:15:18 goals, and they're used to going faster, and
00:15:18 --> 00:15:20 when they have to slow down, it's going to be
00:15:20 --> 00:15:24 quite disruptive. 100 % with you there. And we
00:15:24 --> 00:15:27 are seeing a shift from reactive network management
00:15:27 --> 00:15:30 to predictive and self -healing systems. And
00:15:30 --> 00:15:32 saying that out loud reminds me of the Christine
00:15:32 --> 00:15:34 movie, the Stephen King film all those years
00:15:34 --> 00:15:36 ago, where the car heals itself, and he just
00:15:36 --> 00:15:39 says, show me. It would be great if we could
00:15:39 --> 00:15:41 get to that point on networks. But what does
00:15:41 --> 00:15:43 it all look like in practice? What technologies
00:15:43 --> 00:15:47 are making this possible today? Yeah, you know,
00:15:47 --> 00:15:49 I mean, SD -WAN kind of, you know, I mean, let's
00:15:49 --> 00:15:53 just look at the buzzword, right, of You know
00:15:53 --> 00:15:55 the self -healing. I've been hearing about self
00:15:55 --> 00:16:00 -healing for at least 10 years, right? So, and
00:16:00 --> 00:16:02 I've even implemented a lot of this early on
00:16:02 --> 00:16:07 with the DMVPN days, you know, and then the early
00:16:07 --> 00:16:10 SD -WAN. So, you know, you've got kind of this
00:16:10 --> 00:16:13 telemetry -driven, you know, routing decisions
00:16:13 --> 00:16:16 that are being made. That's great in the SD -WAN
00:16:16 --> 00:16:19 platform, but that's got to extend now into the
00:16:19 --> 00:16:22 data center to data center. And if you look at
00:16:22 --> 00:16:24 data center to data center, then you got cloud
00:16:24 --> 00:16:27 to cloud. So this is where you bring in that
00:16:27 --> 00:16:29 network as a service backbone between the two.
00:16:29 --> 00:16:32 And it also has to have that same type of intelligence
00:16:32 --> 00:16:36 for self -healing and predictive route changes,
00:16:36 --> 00:16:38 et cetera, within the different fabrics that
00:16:38 --> 00:16:42 you've got there. And if you look just recently,
00:16:43 --> 00:16:47 you can see that the big ISPs are, see that that
00:16:47 --> 00:16:49 is the fundamental change, right? It's because
00:16:49 --> 00:16:51 systems are gonna be talking to each other much
00:16:51 --> 00:16:53 more frequently, you're gonna get a lot more
00:16:53 --> 00:16:55 cloud to cloud, you're gonna get a lot more site
00:16:55 --> 00:17:00 to site, east -west traffic patterns instead
00:17:00 --> 00:17:01 of just north -south, which is what we've been
00:17:01 --> 00:17:05 designing for for the last 10 years. We just
00:17:05 --> 00:17:10 saw a recent acquisition of from Lumen to Alkira,
00:17:10 --> 00:17:13 for instance, the Lumen -Alkira, it's really
00:17:13 --> 00:17:16 a signal to the direction the industry's looking.
00:17:17 --> 00:17:19 Where's that single control plane that can manage
00:17:19 --> 00:17:22 traffic flows, not just at the SD WANs or the
00:17:22 --> 00:17:24 north -south layer, but how do we actually look
00:17:24 --> 00:17:28 at east -west as well? Big thank you to Denodo
00:17:28 --> 00:17:30 for supporting the Tech Talks Network and making
00:17:30 --> 00:17:34 these conversations possible. Because when your
00:17:34 --> 00:17:37 lakehouse stores the data, The real challenge
00:17:37 --> 00:17:40 is getting that data where it needs to go and
00:17:40 --> 00:17:44 faster. And your lake house stores the data,
00:17:45 --> 00:17:48 but Denodo helps deliver it faster. So with real
00:17:48 --> 00:17:51 -time access, built -in governance and a business
00:17:51 --> 00:17:54 -ready data marketplace, Denodo can help your
00:17:54 --> 00:17:58 teams unlock insights without costly duplication.
00:17:58 --> 00:18:01 And you can learn more by simply visiting denodo
00:18:01 --> 00:18:04 .com. For people listening in organizations that
00:18:04 --> 00:18:07 are running hybrid or multi cloud environments
00:18:07 --> 00:18:10 How do you design connectivity that is both resilient
00:18:10 --> 00:18:14 and flexible, but also without adding unnecessary?
00:18:14 --> 00:18:17 Complexity or costs and I appreciate that question
00:18:17 --> 00:18:20 is almost a podcast episode on its own But anything
00:18:20 --> 00:18:22 you can share around that all kind of advice
00:18:22 --> 00:18:24 that you would offer for people in that situation
00:18:24 --> 00:18:28 listening Yeah, you know, I think it's you have
00:18:28 --> 00:18:30 that design principle that you know, it's integration
00:18:30 --> 00:18:33 and management competence across many underlays
00:18:33 --> 00:18:36 So you shouldn't be choosing between resiliency
00:18:36 --> 00:18:39 and simplicity here Yeah, I think I think you
00:18:39 --> 00:18:42 know your concrete building blocks that I'm seeing
00:18:42 --> 00:18:44 and what we're putting into ours are those kind
00:18:44 --> 00:18:46 of network as a service fabrics I mentioned them
00:18:46 --> 00:18:51 before You know multi -carry underlays that's
00:18:51 --> 00:18:53 where you're orchestrating the SD -WAN or the
00:18:53 --> 00:18:56 on -ramp to those network as a service fabrics.
00:18:57 --> 00:19:01 And then you have your consistent security policy.
00:19:01 --> 00:19:03 So you have to start thinking of not just infrastructure
00:19:03 --> 00:19:07 as code, but policy as code across these, so
00:19:07 --> 00:19:10 that you're designing not connectivity, but you're
00:19:10 --> 00:19:12 also when the network changes, it's easy to make
00:19:12 --> 00:19:18 sure that your policies are adopting the changes
00:19:18 --> 00:19:23 well. So that's really kind of it. It's really,
00:19:23 --> 00:19:25 to me, it's those three building blocks there.
00:19:26 --> 00:19:28 The network as a service, multi -carry underlays,
00:19:28 --> 00:19:31 bringing in that centralized single control plane.
00:19:31 --> 00:19:35 And that right now, that's kind of more of a
00:19:35 --> 00:19:38 goal than a product because there isn't really
00:19:38 --> 00:19:41 something that's taking your SD -WAN SASE and
00:19:41 --> 00:19:46 your multi -cloud routing overlay and then your
00:19:46 --> 00:19:48 network as a service fabric and looking at it
00:19:48 --> 00:19:51 holistically. I mean, I think MSPs can come in
00:19:51 --> 00:19:54 and actually perform that for you as kind of,
00:19:54 --> 00:19:57 hey, we know this vendor, we know the network
00:19:57 --> 00:19:59 as a service layer, we know the cloud routing,
00:19:59 --> 00:20:01 we can build an architecture and we can observe
00:20:01 --> 00:20:04 it for you, and we can better manage that, but
00:20:04 --> 00:20:06 it's still more of an architectural goal than
00:20:06 --> 00:20:11 a product today. So that's kind of how we build
00:20:11 --> 00:20:15 today, right? We build them that on -ramp SD
00:20:15 --> 00:20:19 -WAN, SASE, and then we build a network as a
00:20:19 --> 00:20:21 service that's connecting multi -cloud all other
00:20:21 --> 00:20:24 cloud providers resiliently into this resilient
00:20:24 --> 00:20:28 fabric. So it's typically two multiple network
00:20:28 --> 00:20:30 as a service provider. So you get that resilience.
00:20:30 --> 00:20:32 So you get that cloud to cloud in that data center
00:20:32 --> 00:20:34 on ramp or that data center and data center.
00:20:35 --> 00:20:37 And then you've got the cloud workflows and the
00:20:37 --> 00:20:41 East West and the cloud. So if it's Azure, it's
00:20:41 --> 00:20:43 like is the cloud provider and you've got multi
00:20:43 --> 00:20:45 -regions of Azure, it's best to keep that inside
00:20:45 --> 00:20:48 the cloud. the CSP's fabric, it's going to be
00:20:48 --> 00:20:50 lower cost and it's going to be lower latency
00:20:50 --> 00:20:52 than to take it out. But when you go cloud to
00:20:52 --> 00:20:55 cloud, so like Azure to AWS, for instance, you
00:20:55 --> 00:20:57 want to take that through your network as a service
00:20:57 --> 00:21:00 layer. But being able to put all that together
00:21:00 --> 00:21:03 and observe it all, and then understand where
00:21:03 --> 00:21:06 things are potentially at capacity strengths
00:21:06 --> 00:21:09 or where you need to augment and dynamically
00:21:09 --> 00:21:12 improve, that's where an MSP brings a lot of
00:21:12 --> 00:21:15 value, and that's what Global Gate does. I think
00:21:15 --> 00:21:17 without looking into a crystal ball I think the
00:21:17 --> 00:21:19 one thing that is incredibly easy to predict
00:21:19 --> 00:21:23 as we look ahead is the amount of bandwidth that
00:21:23 --> 00:21:25 we all use whether it be in the workplace or
00:21:25 --> 00:21:28 at home is going to scale dramatically and you
00:21:28 --> 00:21:30 mentioned that great example at the beginning
00:21:30 --> 00:21:33 of you using agents and how many websites those
00:21:33 --> 00:21:36 agents access but What do you see a future -ready
00:21:36 --> 00:21:39 global connectivity strategy looking like for
00:21:39 --> 00:21:42 enterprises that want to scale AI safely? And
00:21:42 --> 00:21:45 what would you say are the maybe first technical
00:21:45 --> 00:21:48 steps teams should be taking right now to begin
00:21:48 --> 00:21:51 moving in that direction? Yeah, again, I think
00:21:51 --> 00:21:53 it's that programmable network, right? That one
00:21:53 --> 00:21:55 control plane. That's the future. Everybody's
00:21:55 --> 00:21:57 got to get there. There's going to have to be
00:21:57 --> 00:22:00 products that actually are evolved there to get
00:22:00 --> 00:22:04 us there sooner. Um, the underlay diversity thing
00:22:04 --> 00:22:07 we talked about, um, you know, verified physically
00:22:07 --> 00:22:11 though, not just, not just by, you know, Hey,
00:22:11 --> 00:22:13 he said, she said, you know, that this is a,
00:22:13 --> 00:22:16 that this is two diverse carriers. It's actually
00:22:16 --> 00:22:19 getting the KMZ files and looking at the paths
00:22:19 --> 00:22:21 and then looking at the peering relationships
00:22:21 --> 00:22:26 and making a decision there. Inference, where's
00:22:26 --> 00:22:29 the organization placing it? I think jurisdiction,
00:22:30 --> 00:22:32 latency, data governance, all those things are
00:22:32 --> 00:22:36 going to play. So it's not just going to be driven
00:22:36 --> 00:22:38 where compute is cheap. It's going to be all
00:22:38 --> 00:22:40 these things are going to play. So you have to
00:22:40 --> 00:22:43 have a network that's adapted to that. Then again,
00:22:43 --> 00:22:47 the invisibility, so the observability piece
00:22:47 --> 00:22:54 of it all. Then leveraging AI. to assist humans
00:22:54 --> 00:22:58 in the beginning to run it and then ultimately
00:22:58 --> 00:23:01 automating a lot of those functions. You mentioned
00:23:01 --> 00:23:07 technical steps as well. So I think the place
00:23:07 --> 00:23:12 to start there would be with an audit. I mean,
00:23:12 --> 00:23:15 I think it's kind of, is the... Are the tools
00:23:15 --> 00:23:17 that you've got in place? So that's the SD -WAN
00:23:17 --> 00:23:20 fabric. Do you have multi -cloud infrastructure?
00:23:21 --> 00:23:24 Are they capable of getting you to the next steps?
00:23:25 --> 00:23:29 You know, audit your circuits as well. How many
00:23:29 --> 00:23:31 of them are broadband circuits? How many of them
00:23:31 --> 00:23:33 need to be, you know, that diverse circuit that
00:23:33 --> 00:23:36 you're buying need to be upgraded to a symmetrical
00:23:36 --> 00:23:40 type service? whether that's a FTTX or DIA light
00:23:40 --> 00:23:44 or even a full DIA based on what sites you believe
00:23:44 --> 00:23:46 are gonna be there. So I think you need to get
00:23:46 --> 00:23:50 that audit done on the underlay as well. And
00:23:50 --> 00:23:53 then really kind of just working across your
00:23:53 --> 00:23:56 teams. So that's infrastructure, whoever owns
00:23:56 --> 00:24:00 your AI strategy and inference, understanding
00:24:00 --> 00:24:04 where they're going so that you can make sure
00:24:04 --> 00:24:06 that you're ahead of that on the network side.
00:24:07 --> 00:24:09 I think a lot of these teams still work in silos
00:24:09 --> 00:24:13 and where a lot of them may be taking shortcuts,
00:24:13 --> 00:24:16 because there's a lot of risk kind of that I
00:24:16 --> 00:24:19 believe organizations are taking right now because
00:24:19 --> 00:24:22 their CEOs are wanting to be first. They're wanting
00:24:22 --> 00:24:25 to be fast. They're telling people to embrace
00:24:25 --> 00:24:30 AI. There is a lot of that inferences going out
00:24:30 --> 00:24:35 to the big players right now. people are starting
00:24:35 --> 00:24:38 to redline these MSAs with their customers that
00:24:38 --> 00:24:42 forbid them from taking PII and customer identifiable
00:24:42 --> 00:24:45 information to these public LLMs, these public
00:24:45 --> 00:24:48 models. People are gonna start bringing that
00:24:48 --> 00:24:52 stuff inside so that it's gonna become part of
00:24:52 --> 00:24:54 their infrastructure. And I think the network
00:24:54 --> 00:24:57 team and the security team need to get engaged
00:24:57 --> 00:25:01 now to understand what that looks like so they
00:25:01 --> 00:25:06 can start building. That's what we do. We bring
00:25:06 --> 00:25:09 those teams together. We have these conversations.
00:25:09 --> 00:25:11 We let them know what the industry is doing,
00:25:11 --> 00:25:14 what our customers, our other customers are doing,
00:25:14 --> 00:25:17 and then we come up with a plan and help them
00:25:17 --> 00:25:20 get there. Well, I've had so many conversations
00:25:20 --> 00:25:23 around AI this year and everything from agents
00:25:23 --> 00:25:25 to the importance of data and the models, et
00:25:25 --> 00:25:28 cetera. But I love how you've been able to shine
00:25:28 --> 00:25:31 a light on the importance of the networks that
00:25:31 --> 00:25:33 we all take for granted today. And for anybody
00:25:33 --> 00:25:35 listening that would like to continue that conversation
00:25:35 --> 00:25:38 with you, maybe you've set off a few light bulb
00:25:38 --> 00:25:40 moments. Maybe somebody's listening thinking,
00:25:40 --> 00:25:42 at last somebody's talking about this. Where
00:25:42 --> 00:25:44 would you like me to point them so they can find
00:25:44 --> 00:25:47 out more information about Global Gig and maybe
00:25:47 --> 00:25:52 connect with you? Yeah, for sure. I'm not big
00:25:52 --> 00:25:55 on social media outside of the industry standard,
00:25:55 --> 00:25:57 LinkedIn, you know, so you can do Global Gig,
00:25:57 --> 00:25:59 Jamie Pugh, and you'll probably find me. I should
00:25:59 --> 00:26:04 be the first result there. And alternately, our
00:26:04 --> 00:26:06 website is a great source of information too.
00:26:07 --> 00:26:11 So that's www .globalgig .com, G -L -O -B -A
00:26:11 --> 00:26:15 -L -G -I -G dot com. Awesome. Well, I'll add
00:26:15 --> 00:26:17 links to everything that you mentioned there.
00:26:17 --> 00:26:19 I, for one, love how you're helping transform
00:26:19 --> 00:26:23 networks into a strategic advantage via managed
00:26:23 --> 00:26:25 network and security solutions. I'd urge anyone
00:26:25 --> 00:26:28 listening that's enjoyed today's conversation,
00:26:28 --> 00:26:31 please go check them out. Connect with you. And
00:26:31 --> 00:26:33 I'd love to hear back from you on anything that
00:26:33 --> 00:26:34 you thought. But more than anything, thank you
00:26:34 --> 00:26:36 for starting this conversation today. Really
00:26:36 --> 00:26:39 enjoyed talking with you. Yeah, thanks again
00:26:39 --> 00:26:41 for having me, Neil. Appreciate it. One of the
00:26:41 --> 00:26:43 things I loved about this conversation with Jamie
00:26:43 --> 00:26:46 today was how it shifted the AI discussion away
00:26:46 --> 00:26:49 from the obvious talking points and into the
00:26:49 --> 00:26:52 infrastructure that many people only notice when
00:26:52 --> 00:26:56 it fails. Yeah, we often talk about AI in terms
00:26:56 --> 00:26:59 of models, data, agents, applications, but I
00:26:59 --> 00:27:03 think Jamie reminded us all today that all these
00:27:03 --> 00:27:06 things depend on connectivity. Connectivity that
00:27:06 --> 00:27:09 can handle new traffic patterns, new latency
00:27:09 --> 00:27:12 demands and new levels of business reliance.
00:27:12 --> 00:27:15 And his point about AI changing traffic from
00:27:15 --> 00:27:18 people to applications into systems to systems
00:27:18 --> 00:27:22 must have set off a few lightbulb moments. Because
00:27:22 --> 00:27:25 agents querying tools, models connecting to data
00:27:25 --> 00:27:29 stores, inference engines reaching across clouds
00:27:29 --> 00:27:33 and employees relying on AI assistance. All these
00:27:33 --> 00:27:36 things place... more pressure on networks. Networks
00:27:36 --> 00:27:39 that were often designed for a very, very different
00:27:39 --> 00:27:42 era. And the lesson for me is that resilience
00:27:42 --> 00:27:44 cannot be treated as a backup plan. They need
00:27:44 --> 00:27:47 to understand their underlays, validate carrier
00:27:47 --> 00:27:51 diversity, review cloud on ramps, assess broadband
00:27:51 --> 00:27:54 usage and bring network security infrastructure
00:27:54 --> 00:27:58 and AI teams all into the same conversation before
00:27:58 --> 00:28:01 AI workloads move deeper into production. So
00:28:01 --> 00:28:04 I'll add links to Global Gig and Jamie's LinkedIn
00:28:04 --> 00:28:06 profile in the show notes, but I'd love to hear
00:28:06 --> 00:28:10 what you think. Have you or your business spent
00:28:10 --> 00:28:14 too much time talking about AI models while underestimating
00:28:14 --> 00:28:18 the network needed to support them? TechTalksNetwork
00:28:18 --> 00:28:20 .com, that's where you can find me, talk to me,
00:28:21 --> 00:28:23 work with me, browse through 4 different
00:28:23 --> 00:28:25 interviews. Love to hear your thoughts on this
00:28:25 --> 00:28:28 one, but that's it for today. I'll be back again
00:28:28 --> 00:28:30 real soon with another guest But thank you for
00:28:30 --> 00:28:32 listening as always and I'll speak with you then.
00:28:32 --> 00:28:33 Bye for now