Nobody Negotiates Their Way Out of Being a Supplier

What two days in Amsterdam taught me about intelligence, trust and transformation

On Tuesday afternoon in the Tobacco Theater, straight after lunch, I put one sentence on the screen: nobody negotiates their way out of being a supplier.

I’d been carrying that line around for months. Paris, Madrid, New York, London, Copenhagen: in almost every room this year, I heard some version of the same sentence, from operators, cloud comms platforms and contact centre players alike. The AI industry needs us. AI runs over our networks. AI has to reach the customer, and we’re the ones who reach the customer.

It’s true. It’s also exactly how a supplier talks.

That tension was the real story of CASA26, the CPaaSAA summit I’ve just spent three days hosting in Amsterdam, with GSMA, Sandbox Industries and McKinsey as strategic partners. The programme was built as one story in three acts: Intelligence, Trust, Transformation. Now the room has emptied, I want to write down what I think that story actually says.

1. The buyer isn’t short of intelligence. They’re short of intelligence they can use.

Monday was AI, all day. Leonard Lee of neXt Curve opened with an outside-in map of who actually owns intelligence, and by the afternoon the pattern was unmistakable. Intelligence is no longer a layer you bolt on top. It’s sinking into the network, the device, the conversation itself. The sharpest evidence came from voice, which is interesting again, not as a legacy channel to defend, but as the interface AI agents actually speak through.

Models are commoditising. Intelligence isn’t scarce. Usable intelligence is.

A large enterprise can hire a consultancy or a systems integrator to stitch models, data and workflows together. The plumber can’t. Neither can the florist, the builder or the physio with three staff. Nor can most of the workforce, who don’t sit behind a computer at all. They’re on sites, in vans, on their feet all day. They want AI to make their day easier, but they’re not going to wire it together in Zapier or n8n. They want something ready-made, smart and easy to use, that they can try and keep if it works. 8×8 made the point that customers don’t buy categories. They buy something that solves the problem in their own line of business.

It’s very basic. That’s exactly why so much of the AI industry misses it.

KPN and ElevenLabs showed on stage what the alternative looks like: AI built into the mobile service people already use, as a small monthly add-on. No new app, no integration project, nothing to build.

That’s where this industry comes in, and Dstny put it plainly: small businesses buy from their communications provider. Their internet, their mobile lines, their 5G, their phone system, all from a telco or comms provider they know. That relationship is a distribution channel the AI companies can’t buy. The trust isn’t spotless, and messaging fraud and spam have done this industry no favours. But the relationship is there, and it’s local.

The question has shifted from who has the best model to where do I buy the right intelligence. For most businesses on the planet, the answer should be: from the people they already buy their communications from.

2. The pendulum is swinging back to where the data and the trust live

For fifteen years everything moved to the centre, into a handful of hyperscale data centres where compute was cheapest and models were biggest. Training AI reinforced that. Using AI is starting to pull the other way.

AI needs to run close to the data, and much of the data that matters doesn’t sit in a hyperscaler, if it ever did. The sovereignty debate of the past year has made enterprises and governments far more aware of where their data lives and who can reach it. And a lot of what AI now has to process is local by nature: the voice conversation, the video feed from a site, the customer record in a regional system. A conversation happens in a place, in real time. You don’t want to ship it to the other side of the world, into someone else’s cloud, with all the compliance and privacy baggage that brings, just to answer a customer.

So the AI comes to the data, and the pendulum swings from the centralised, maximum-performance data centre to local, good-enough models. Most communications use cases don’t need the frontier model. A smaller model that’s nearly as good is often the better product, and the safer one, if it runs close to the conversation, answers faster and keeps the data in the country. Radisys took that argument all the way into the operator’s IMS core: AI inside the network, not beside it.

The mid-sized company has its own version of this. It wants to run its own choice of models, on its own terms, where nobody else can get at its data. That is pulling AI workloads back on-premise, into a local or regional data centre, or simply onto hardware in the building where the data already sits. The box runs locally, and a telco connects it, configures it, keeps it updated and bills for it. That’s not a return to the old on-prem world. It’s on-prem delivered as a service, and it opens a market for everyone from the chipmakers to the companies building the equipment at the edge of the network.

That argument didn’t only come from me. On the vCon panel, alongside Crexendo and Simwood, BT’s Richard Atherton underlined how much sovereignty and local solutions matter today. It’s telling when an operator of that scale says it out loud.

Put it together and you get a perfect storm: regulation, latency, sovereignty, and the fact that voice, the interface everyone suddenly wants, already runs through telco networks and comms platforms. That’s why voice is exciting again. Talking to AI is convenient, and it lives where this industry lives.

A personal aside, because I’m living this. The deck of nearly 80 slides I used as MC was built almost entirely by talking to an AI. So is this blog. Website updates, the CRM, pulling statistics, scheduling posts: all of it now runs through a conversation. Vibe coding came first. Now it’s vibe marketing, vibe-PowerPointing, vibe-blogging. The real gain isn’t speed. It’s that everything gets connected, with content, data and knowledge reachable by talking. Communications needs exactly that: every conversation, channel and signal brought together behind one natural interface.

That was the thread through Tuesday morning. Andrew Collinson launched his research for CPaaSAA on the value of network signals, with Miriam Sabapathy of STL Partners alongside. Proximus Global, Vonage, XConnect and the GSMA Open Gateway session came at it from the network, numbering and operator sides. They all landed on the same point: customers don’t want to buy APIs. They want better business. The network’s signals (identity, location, device, consent) are what make intelligence trustworthy enough to use.

For a decade this industry has treated trust as overhead. In an AI economy it’s the opposite. Intelligence is abundant; trust and proximity aren’t. A hyperscaler can train a model. It can’t quickly replicate a national licence, a verified subscriber base, a footprint close to the user, or decades of carrying regulated conversations without breaking them.

Intelligence without trust is a demo. Trust without intelligence is a utility. Put them together and you have something a buyer will actually pay for.

3. “AI needs us” is a supplier’s sentence

Here’s the uncomfortable part. Owning a scarce asset doesn’t make you a partner. It makes you worth partnering with, which is not the same thing.

This industry has two layers: the operators, and the cloud comms platforms we used to call over-the-top. Both are now being challenged by a wave of AI-native startups, and the capital behind that wave is enormous. Venture investors put more than $7 billion into voice AI startups in the first quarter of this year alone. That’s a pace of $28 billion a year into one slice of AI, and that’s assuming it doesn’t accelerate.

Now look at who shows up at this industry’s events. Walk the floor at the big comms shows and you’ll struggle to find the AI companies. Two worlds are building the same future, barely in the same room. CASA26 was an attempt to start closing that gap, with ElevenLabs and a set of speech, voice and AI-in-the-core players sitting next to operators and platforms. But being in the same room is the easy part.

The AI companies have the money and the momentum. What they don’t have is the hard part underneath: the regulation, the sovereignty, the consent, the carrier-grade foundation that real conversations run on. That was the point I tried to make on stage, and it landed.

But if all you do is carry the traffic, you’re not a partner to the AI economy. You’re a user of it, or a supplier to it. You’re building pipes, and this industry already knows how that story ends: it lived it with the internet.

The summit’s audience vote showed the instinct. The room broadly believes the value of intelligence stays in communications. The operators voted conservatively: most chose the future where platforms run the engagement and operators supply the signals underneath. I understand why. It’s the future that asks the least of them. But it’s also the supplier position, chosen by the suppliers themselves.

4. Partners take a stake

So how do you become a partner rather than a pipe? You take a stake. You create value for the AI companies, not just capacity. That is starting to happen, and it’s the most encouraging thing I saw this year.

Radisys is a good example. With V.AI it brings ElevenLabs, Hiya, ReadSpeaker, Sensory and Speechmatics into one ecosystem, combined with its own network, device and media processing capabilities. The speech model makers get distribution, and operators get access to all of them at once. That’s a partnership, not a procurement.

Alianza is going after the layer in between, and it made that case on stage. Crux sits in the call path, handles consent and policy, and hands the conversation to whichever AI experience the line is subscribed to. The AI companies get operator reach, and the operator keeps the customer. That’s orchestration as a business model, not a feature.

Crexendo is taking a stake from the cloud comms side. It’s building vCon, the emerging standard for capturing conversations in a portable, AI-ready format, natively into its NetSapiens platform, which more than 250 service providers run on. On the Monday of CASA26 it announced a partnership with Meetric to add conversation intelligence on top. Structured, consented, compliant conversation data is exactly the foundation the AI companies can’t build for themselves.

A year or two ago, the typical approach was to plug a frontier model into a bot, call it an AI strategy, and move on. That was using AI. What’s emerging now is skin in the game.

That’s what my Tuesday slide was really about. Three things change a structural position: scarcity, demand and risk. None of them is a better contract. Scarcity this industry has. Demand means starting from the buyer’s job to be done, not the capability you already own. Risk is the hard one. The supplier is the party that took none, which is exactly why the supplier gets squeezed.

Nobody takes that risk alone. Operators need speed they can’t hire. Startups need distribution and trust that no funding round buys. That’s why the last afternoon in Amsterdam, hosted with Sandbox Industries, wasn’t another panel about ecosystems. It was founders pitching live to operators and investors and answering the hard questions in public. That takes nerve. The room loved it.

The room was the test

As MC, my job wasn’t to make the argument on my own. It was to narrate the story everyone else was telling: operators, cloud comms platforms, security players, founders and investors, session after session, without being asked to agree with each other. I built the frame. They filled it in.

That made CASA26 a test. If the story was wrong, this was the room to say so: more than 150 people, some of the sharpest in this industry, and not a shy crowd. The analysts alone would have found the hole. Andrew Collinson, Dean Bubley of Disruptive Analysis, Amy Cameron and Miriam Sabapathy of STL Partners, Nick Lane of Mobilesquared, Raúl Castañón of 451 Research, Dave Michels of TalkingPointz, Žiga Lešnik of Simon-Kucher and Leonard Lee of neXt Curve are paid to find what’s missing in an argument. Nobody said I was missing a piece, or that it doesn’t work like that. The vote pointed the same way. There was real debate about who captures the value, and the operators were more cautious than I’d like, but nobody disputed the direction. AI is being embedded in communications, and the open question is who takes the stake.

What I’m taking home is about my own work as much as anyone’s. Research, community and insight are necessary, but not sufficient. The value sits at the table where corporates, startups and capital commit to something together and carry the risk of it. That’s where I’m spending the next year.

The AI industry does need us. It needs us as partners, not as pipes. Getting there isn’t a negotiation. It’s a decision.