Telecom’s AI Future Will Depend On Moving Faster Together — Reflections From DSP Leaders World Forum

DSP Leaders World Forum in London last week brought together a strong mix of operators, vendors, analysts and ecosystem players for two days of discussions around the future of telecom infrastructure, AI, automation and digital transformation.

It was a well-organized event with some genuinely insightful speakers, strong hallway conversations and plenty of opportunities to reconnect with people across the telecom ecosystem. The informal networking discussions were often just as valuable as the panels themselves.

And one theme consistently surfaced throughout the event:

AI is now central to almost every telecom conversation.

Everybody Talks About AI

Two days in London reinforced something we increasingly hear across the telecom industry:

Operators clearly understand that AI matters.

But many are still stuck in a cycle of Proofs of Concept, experiments and pilot projects that are difficult to operationalize, scale and deploy in real-world telecom environments.

MWC Barcelona earlier this year was already dominated by AI discussions. The same themes have since appeared across virtually every major telecom event — from network automation forums to cloud, API and infrastructure conferences.

AI is no longer a side conversation in telecom.

It is now central to almost every strategic discussion happening across the industry.

Sovereign AI. Edge AI. Inference. Autonomous networks. AI-native telcos. AI-driven operations. AI-ready infrastructure. AI security. AI governance. AI regulation. AI-powered customer engagement.

AI is everywhere.

The next major event already being promoted heavily during the conference was TelecomTV’s upcoming AI-Native Telco Forum in Düsseldorf this September — focused entirely on questions like: what does it actually mean to become an AI-native telco? How do operators transition toward AI-native operations? What organizational and technical changes are required?

At the same time, new initiatives continue to appear across the industry. ANTA — the AI-Native Telco Accelerator — was discussed as another ecosystem initiative trying to help operators navigate the transition. GSMA is increasingly leaning into AI discussions as well, while programs like GSMA Fusion are helping industries define more concrete requirements and blueprints for what enterprises actually need from networks and APIs.

So clearly, the telecom industry is mobilizing around AI.

And that is a good thing.

These events matter. We probably need more of them, not fewer. It is important that operators, vendors, standards bodies, system integrators and ecosystem players come together regularly to compare notes, share progress and challenge each other.

But there was also a visible uncertainty running through many of the discussions.

Not because the industry lacks smart people.
Not because the technology does not exist.

But because telecom is fundamentally harder than many other industries now racing into AI.

Telecom Is Different

Outside telecom, many organizations are already moving from AI experimentation toward operational restructuring at enormous scale. AI is no longer a side project there — it is becoming the operating model itself.

Telecom cannot simply behave the same way.

Operators run national infrastructure. Critical enterprise systems. Emergency communications. Highly regulated environments. Security-sensitive workloads. Global connectivity platforms.

You cannot “move fast and break things” when millions of customers and entire economies depend on your infrastructure.

And yet telecom also cannot afford to move too slowly.

That tension was visible throughout the event.

Because the problem is no longer proving that AI can do interesting things.

The problem is building AI systems that can actually operate reliably inside highly regulated, mission-critical telecom environments at real scale.

And that is a very different challenge.

The POC Trap

One of the recurring themes throughout the event was the gap between Proofs of Concept and production deployment.

Most operators now have AI pilots running across networks, operations, analytics, customer service and security. But very few seem fully confident yet about how to scale these initiatives operationally.

And honestly, that makes sense.

Because the hard part is not building a demo anymore.

The hard part is everything around it:
data governance,
compliance,
privacy,
security,
operational integration,
observability,
resiliency,
organizational alignment,
and trust.

Several speakers made variations of the same point: if these realities are not designed into the solution from the beginning, many AI projects never become deployable products.

This is why the industry increasingly needs practical frameworks and operational blueprints instead of endless disconnected experiments.

Stop Reinventing The Wheel

What was interesting during the event was the repeated observation that the industry does not necessarily need to reinvent everything from scratch.

For example, practical blueprints and solution packages already exist for common operational challenges around automation, fault management and service assurance.

TM Forum has been heavily focused on Autonomous Networks for years and has already developed maturity models, scoring frameworks and operational solution packages to help operators move toward more intelligent and automated network operations.

Most operators are still somewhere around level 1.5 to 1.8 today, while many aspire to eventually reach level 4 autonomy.

In other words:
parts of the framework already exist.

And yet many operators still move cautiously, incrementally and often prefer to heavily customize or rebuild solutions themselves.

Partly, that comes from telecom culture.

Operators understandably want ownership and control over their infrastructure. Telecom has always had a strong engineering culture, and many operators remain cautious about becoming overly dependent on vendors or black-box systems.

That instinct is understandable.

At the same time, modern telecom environments have become far too complex for any single organization to build and manage entirely alone.

Distributed AI, edge inference, cloud-native architectures, orchestration, sovereignty, security and automation require much deeper collaboration between operators, vendors, cloud providers, systems integrators and increasingly startups.

The challenge is no longer whether the ecosystem should collaborate.

The challenge is how to collaborate faster and more practically.

AI Needs Ecosystems, Not Silos

Twenty or thirty years ago, networks were far simpler environments.

Today operators are dealing with distributed AI, edge inference, cloud-native architectures, real-time orchestration, APIs, security, sovereignty, automation, compliance and increasingly AI-native operational models.

The complexity level is fundamentally different.

And the reality is that operators themselves have spent the last two decades outsourcing and partnering extensively with technology vendors, cloud providers and systems integrators.

So the answer is probably not:
“let’s go back to how things worked twenty years ago.”

The answer is learning how to collaborate better.

Because if telecom can figure this out, the opportunity is enormous.

That was another important takeaway from the week.

Telecom is not “just connectivity” anymore.

Operators increasingly sit at the intersection of AI, identity, trust, security, communications, edge infrastructure, real-time orchestration, sovereignty and distributed computing.

If it was easy, everybody could do it.

But they cannot.

That is exactly why telecom still matters strategically in the AI era.

AI Is Not Magic

Another important realization slowly emerging across the industry is that AI itself is not magic.

Not every problem requires AI.

Some problems simply require better automation, better orchestration, better scripting, better operational processes or clearer data structures.

And even where AI does make sense, there are still difficult challenges around determinism, predictability, auditability, explainability, governance and operational trust.

That means operators increasingly need to take a use-case-driven approach.

Not:
“where can we add AI?”

But:
what problem are we actually trying to solve?
Does AI genuinely improve the outcome?
What are the operational and regulatory realities?
How do we make this deployable at scale?

Enterprises Need To Be In The Room

One of the refreshing aspects of the event was the presence of automotive players and discussions around software-defined vehicles, roaming complexity, global orchestration and connected infrastructure.

NTT Docomo discussing the strategic reasoning behind acquiring a bank was another interesting example of telecom operators thinking beyond pure connectivity.

But overall, enterprises were still underrepresented.

And that matters.

Because in many cases, the technology itself is not the bottleneck anymore.

The bigger challenge is adoption.

Getting solutions through procurement.
Through legal.
Through security teams.
Through compliance processes.
Through operational governance.

That is exactly why we have been leaning into enterprise-focused workshops at CPaaSAA.

Not technology-first discussions.

But problem-first discussions.

What are enterprises actually trying to solve?
Where can smarter networks, APIs, AI and trusted infrastructure genuinely help?
What are the operational blockers?
What are the trust requirements?
What would make adoption realistic?

Because if the pain is not large enough — or if the framework for trust is not mature enough — adoption will remain slow, regardless of how impressive the technology is.

Where Are The AI Startups?

Another thing I found particularly striking was who was not in the room.

There were very few startups.
Very few AI-native companies.
Very little discussion about actual LLM deployment strategies, orchestration frameworks or practical AI product ecosystems.

And frankly, almost none of the major AI platform companies were present either.

At times, the event felt a bit like an echo chamber:
telecom people talking to telecom people about how important AI is becoming.

Without enough participation from the companies actually building the next generation of AI infrastructure and AI-native software environments.

Even some of the more interesting technical observations were almost hidden between the lines.

For example, there were important discussions around inference moving toward the edge and the idea that many practical inference workloads may not require giant GPU clusters at all — that CPUs and more distributed architectures could play a major role in future deployments.

That is actually a strategically important observation.

But you had to listen very carefully to even catch it.

And maybe that summarizes part of the broader challenge:
the industry still lacks enough practical collaboration between telecom and the broader AI-native ecosystem.

Hybrid Changes Everything

Another major takeaway from the week is that the future will almost certainly not be fully edge or fully centralized public cloud.

It will be hybrid.

And “hybrid” is really another way of saying:
we do not fully know yet what should run where.

That will depend on workloads, latency requirements, sovereignty, economics, regulation, enterprise needs and operational realities.

Which means operators need to make careful bets.
Careful partnerships.
Careful investments.

Open source also plays an increasingly important role in this transition.

One of the recurring themes throughout the event — particularly from players like Red Hat and SUSE — was that operators want flexibility, portability and the ability to avoid becoming locked into a single cloud, infrastructure or AI stack.

That matters even more in a world increasingly shaped by sovereignty requirements, distributed inference and hybrid deployment models.

The future AI-native telecom environment will almost certainly rely on a mix of public cloud, private infrastructure, edge deployments and specialized AI environments working together.

Open-source technologies increasingly provide part of the connective tissue that makes that possible.

And increasingly, they need help from specialized startups solving very specific problems around orchestration, AI deployment, observability, distributed inference and sovereign AI.

The Industry Needs To Move Faster Together

That is also why CPaaSAA and Sandbox Industries were at DSP Leaders World Forum in the first place.

Not simply to attend another telecom conference.

But because we genuinely believe the industry needs stronger bridges between operators, enterprises, vendors, investors, startups and the broader AI ecosystem.

We sponsored the drinks reception not because branding matters that much, but because informal conversations and ecosystem connections increasingly matter a lot.

Some of the most valuable discussions during events like this happen outside the formal panels:
between operators and startups,
between enterprises and infrastructure providers,
between investors and founders,
between people trying to solve similar problems from different angles.

That ecosystem layer is becoming strategically important.

Because telecom’s AI future will not be built by a single vendor, a single operator or a single platform.

It will be built through collaboration.

This is also where startups and scaleups become extremely important.

Not because telecom should “fail fast.”
Operators cannot afford to fail fast in the same way consumer software companies can.

But startups can explore.
They can specialize.
They can experiment faster.
They can focus deeply on solving specific technical challenges.

That creates a very natural partnership model:
operators provide scale, trust and operational environments;
startups provide speed, experimentation and focused innovation.

And honestly, that startup ecosystem was still largely missing from this event.

There were operators.
Vendors.
System integrators.
Industry organizations.

But very few AI-native startups.
Very few frontier AI companies.
Very little direct participation from the people building the next generation of AI platforms and orchestration layers.

That absence was noticeable.

Because if AI is truly going to reshape telecom infrastructure, the industry needs to move faster together.

Not just with more discussions about AI.

But with more practical collaboration.
More enterprises in the room.
More startups in the ecosystem.
More operational frameworks.
More real-world deployments.
And more builders sitting at the same table.