Telcos Take Charge On AI

The energy on the MWC floor said it all: AI for telecom has moved past “someday” and into “show me how fast.” We sit down with Broadcom’s Anupama Mahabhashyam to unpack what sovereignty really means for carriers, why local control and continuous compliance matter, and how operators can turn regulated infrastructure and rich datasets into a durable AI advantage.

We trace the shift from AI hype to execution, highlighting the operational realities telcos face as they build private and sovereign AI clouds. That includes the unglamorous but essential work of model governance—curating approved catalogs, tracking versions and provenance, and enforcing access policies—so teams stop firefighting model sprawl. We also cover the emerging traffic pattern of AI workloads, why early signals may be invisible on traditional links, and how the rise of voice and video inference changes bandwidth planning and east‑west flows across the network.

From there, we dig into the architecture: intelligent orchestration that matches models to GPU capacity, keeps workloads close to data, and prevents oversubscription. Anupama explains how AI as a Service abstractions can remove plumbing while preserving control, letting teams focus on high‑value use cases like anomaly detection, automated triage, and customer care copilots. We emphasize data readiness as the make‑or‑break factor—organizing datasets, enforcing metadata standards, and eliminating silos so generative systems don’t amplify fragmentation. Finally, we connect the dots to outcomes: improved reliability, faster MTTR, better customer experiences, and new revenue streams such as GPU‑as‑a‑Service and compliant enterprise copilots.

If you’re a carrier leader, network architect, or product owner mapping a path to sovereign AI, this conversation offers a clear blueprint: build on governed data, enforce model discipline, and invest in an intelligent infrastructure layer that scales with demand. Enjoy the episode, then subscribe, share with a colleague, and leave a quick review to help more builders find the show.

Transcript

Setting The Stage At MWC

SPEAKER_00 0:01

And it's Evan here. I'm with Anupamal from Broadcom at MWC Barcelona. How are you?

SPEAKER_01 0:06

I'm doing well. Thank you, Ivan.

SPEAKER_00 0:07

It's been a great week. We're going to dive into all things MWC, but before that, introduce yourself, your role within Broadcom, and what is your team up to?

SPEAKER_01 0:15

Yeah, thank you. So my name is Anupama Mahabasham. I lead product marketing and the BCF division for service provider solutions. So we are really excited because we just announced our new uh BMB Telco Cloud Platform 9. So we are really excited to meet our customers, partners, and analysts.

SPEAKER_00 0:32

And you have a huge ecosystem of partners and relationships here at MWC. Tell us about the vibe here on the ground and some of the feedback from the big announcement.

SPEAKER_01 0:41

Yeah, so the uh what really struck uh in at MWC this year is it's quite different from what I have seen in the previous years. So AI has stopped, uh it's like the conversation is no longer about uh potential, right? It's about uh what it actually is can uh start doing things. So every operator that we spoke to, or what you're actually showing in the vibe that you see on the show floor, it's completely different. Uh they are not asking, uh they are not asking that, okay, what can uh should we actually do the AI? They're actually saying that uh we want to do it, how fast we can do it. So the conversations, it's impressive that the infrastructure conversations uh totally caught up onto AI conversations. And telcos especially, they understand they can't have one without the another. So network is critical, data is critical, and uh telcos uh have uh are in a very uh good position because they actually are sitting on both of them. So that's the different vibe that you have seen from the previous years.

Defining Sovereignty For Telco AI

SPEAKER_00 1:38

Oh, that's so well said, and really it looks like telcos are trying to take control of their own destiny, uh, not rely on necessarily big tech or the hyperscalers or others, but really um create their own sovereign AI clouds and ecosystems. Um, what needs to happen for telcos to really capture this opportunity?

Monetization Hopes And Traffic Reality

SPEAKER_01 1:57

Yeah, that's uh really question. So before I dive into what opportunity, right? So let's uh figure out like edit score sovereignty comes down to five different traits, in my opinion. So operational control is uh must be local, essentially, and then you should not have an external jurisdiction. All your workloads and your metadata, everything has to be within the borders. Then it the operations has to be vetted by the local staff. And then the compliance is not just a one-time certification check and you are done. It's an ongoing process and you need to be continuously aligned with the industry and uh national regulations. So this uh so I think telcos are in a very uh have a huge opportunity to become the sovereign AI infrastructure providers, in our opinion. The reason is that they are already well positioned to operate and are actually own the infrastructure which is regulated, trusted by the government and nationals, and that's a huge opportunity for them to leverage this opportunity and start generating or providing AI services to their customers. So that's a huge opportunity tap into. And that's where our role as the VMware Telco Cloud platform uh taps into. We are helping the operators to see this vision and helping them become the sovereign network and sovereign uh providers with our solution, essentially.

SPEAKER_00 3:18

Well, that's fantastic. And of course, you have to get the architecture right, you have to get the strategy right, and I know you're helping with that. And um, then hopefully AI will drive not just traffic on the network, but revenue. Uh what are you seeing? What are the operators telling you about their readiness to drive traffic and revenue around AI?

Infrastructure, Data, And Model Governance

SPEAKER_01 3:37

Yeah, that's a very interesting question. I think that's the argument that everybody is uh talking about, right? I think uh, in my opinion, I think you know, honest answer, the meaningful uh AI-driven uh traffic has not been completely materialized by the operators' networks. So, but uh I think it's slowly emerging, and I think there is uh we can see that there is a significant change that is happening beneath the surface. So, what I mean by that is it's really important to understand that uh the impact of AI doesn't come uh almost like a tsunami, right? It actually gradually uh comes, and I think uh the majority of AI infrastructure activity, it's happening in the networks that I think the traditional access bound operators uh do not have visibility into it. So I think uh so it's it what it helps is like I think they are thinking or they are actually lacking is that oh their network is not going to show up. So it might be a problem for them because there might be a burst of traffic that is actually happening elsewhere in the network. So it's important to understand that. Another important point is with this AI evolution, the traditional way of network traffic from AI applications like text-based applications to now video and voice applications, I think the amount of volume or the data uh is has increased tremendously. So, with this, there is so much of exchange of data between the network that is happening both the directions. So that means I think uh if uh the foundation is not laid right, I think the operators who are waiting for the traffic to come suddenly burst at them, I think they'll be at a risk and they will not be able to uh uh tap on this opportunity.

Intelligent Orchestration And AIaaS

SPEAKER_00 5:19

Yeah, that would be a shame. And data readiness is one of the keys to execution. Of course, there's skill sets and tooling and what what are some of the obstacles to overcome when it comes to execution, not just the vision or the strategy?

Data Readiness And Business Outcomes

SPEAKER_01 5:32

Yeah, I think the important barrier that I I think I see is that or we think is that it's they're confused about AI, uh thinking about AI and actually doing it. And then the real complexity when they actually go into the real engine, that's when the complexity hits. Uh it's uh it's just that I think let's take a few examples, right? I think it's again everything boils down to infrastructure and data management. For example, if you want they want to fundamentally run a simple model, right? You can't just go into an external repository, download a 200 gigabit model, and then uh assume that you're gonna repeat this uh process every single time you want to spin up a new instance. That is not scalable, first of all, it's not secure, and moreover, it's not reliable because it's just uh I think uh the amount of uh tra uh what you are uh actually tapping into, it's not even uh a thing. So you need a uh essentially the governance and model governance in order to do that. And then the next aspect is like I mentioned, to go into deeper about the model governance. There are so many, there's platforms like Hugging Face, uh, you have millions of models. So what model my organization is running and uh uh who is actually authorizing it or who is approving it, what version I should be using. All this, so in if you don't have that uh uh cloud maturity or model maturity model that is not sitting right next to your infrastructure, you lose control very quickly. So that's a big problem. And then the next problem that we can think about is about the orchestration of uh data, essentially, like how do you so essentially matching a model with the GPU capacity. I don't want to do over subscription, so I want to ensure that uh everything is uh going smoothly. So for that, you need to have uh some kind of uh correlation and uh uh connection.

SPEAKER_00 7:30

Yeah.

SPEAKER_01 7:30

Yeah, the connection essentially. So I think for that the uh intelligent orchestration, I think operators who are emerging from being this uh uh ambitious driven to actually uh implementing, they have started to realize that so having in order to implement this, having an intelligent infrastructure is equally important for as intelligent as the AI layer on top of it. So I think that's a big change that we are seeing, and I think that's where VMwate AI as a service actually really helps. So what it does is it removes the complexity that we just talked about, removes the plumbing that you need to do, and it abstracts the complexity and it gives you the services that instead of that, you can actually focus on the use cases that you want to deliver to your customers rather than uh the internal details that you don't need to worry about.

SPEAKER_00 8:18

That's fantastic. Any final takeaways from what you're hearing from the operators? What are they excited about this year as they get into private AI and sovereign and uh GPU as a service? They have so many ideas on the table. What do you what are you hearing and seeing?

SPEAKER_01 8:33

So I think what I uh what I see is that I think one of the uh important aspects uh before uh they implement all these different use cases and agent tick AI, there are a few things that they need to understand in order to tap the real opportunity, right? Uh you see that uh I I think telco started to realize, in my opinion, which I can clearly see in the MWC with millions of subscribers, uh thousands of uh uh services and uh uh so many operational teams, hundreds of operational teams doing it. They realize that AI can really help them uh triage the their issues and then find the root causes and provide solutions very easily, right? But I think the important thing that needs to happen before that implementing is the reason for all this is the data is like humongous. They have the diversity and also the scale of data that they own is tremendous. So it's important that first the data needs to be organized, it needs to be structured so that there's no uh information uh silos that you have in your organization. So then because Gen AI mimics the silos essentially as soon as you have it. So first that's the first step. And the second, like what we discussed earlier, model the governance needs to be there. It needs to know who is accessing it, who is what are they actually using it for, and then what is the outcome that is expected out of this, right? That's important. And then finally, I think uh it's about uh uh the connection of all these two things and then uh bringing in the you are use cases that actually matter to the business outcomes and uh business. So are you doing it for impact to your business uh strategy or is it a customer experience that you want to improve, or are you doing it for your operational efficiency? All these matter, and I think with so many things coming with the Agentique AI, I'm sure that we will have so much of autonomous uh uh networks that maybe the issues are resolved automatically without even a uh customer knowing it, actually. So it's a lot of people.

SPEAKER_00 10:32

Well, that would be that would be a dream. Um and you know, congrats on doing all of the hard lifting and and magic behind the scenes of of the telcos. Well done.

SPEAKER_01 10:40

Thank you. Thank you so much.

SPEAKER_00 10:41

And thanks so much. Have a great MWC.

SPEAKER_01 10:42

You too. Bye bye.