There is a law firm somewhere that wants its client calls to land in its case management system. The way it does this today is that the lawyer, before every call that matters, opens a web interface, types in a three-way calling number, dials it into the conversation, and hopes the recording makes it through.
That is the state of the art in 2026. Not in a legacy corner of the market — in the part of the industry that has just absorbed tens of billions of dollars of investment in conversational AI. The intelligence is extraordinary. Getting it attached to an actual phone call is still a manual workaround performed by a professional who bills by the hour.
Dag Peak, chief product and technology officer at Alianza, told that story in a CPaaSAA Talk we recorded this week. It is the most useful thing in the launch, because it locates the problem precisely. The barrier between conversational AI and the world’s conversations is not model quality, and it is not the network. It is provisioning, consent, and the fact that a phone call can only ever be forwarded to one place.
That last constraint is worth sitting with, because almost every commentary on this launch has skipped straight past it to the word “orchestration.”
This piece draws on both halves of the week. Our Founding Partner, Rob Kurver, was in Scottsdale for Navigate, where Alianza Crux was formally launched — the keynotes, the architecture, the developer surface, the ecosystem pitch made to a room of operators — and the Talk with Peak was recorded a few days later, once the slides had settled and the questions the launch raised could be put to the person responsible for building it. What follows is our read of both.
The phone number has been a winner-takes-all asset
When we published AI Voice: Who Will Run the Conversation? in March, the question in the title was deliberate. The assumption across the market — from AI-native start-ups, from platform partners, from operators — was that someone would win the conversation, singular. One agent, one number, one relationship.
That assumption is not a strategic belief. It is an artefact of call forwarding.
Because a number can be pointed at exactly one destination, every company building conversational AI has been forced into a fight for sole possession of the call. Peak’s description of the resulting market was blunt: a zero-sum cage match, where the prize is the phone number and everyone else is locked out. The business consequence is that the buyer — the restaurant, the clinic, the 300-location field services business — must choose one AI capability and forgo the rest. Peak’s analogy lands it: imagine an enterprise being told it may run Salesforce or Gong, but not both.
Alianza Crux is interesting because of what it does to that constraint rather than what it adds on top of it. If multiple experiences can attach to the same live conversation under policy — a fraud check before, a live translation during, an analytics feed and a booking afterwards — then conversation access stops being rival. The question changes from who owns this number to who is worth attaching to it.
That is a market structure change, not a product feature, and it is the part of this launch the industry should be arguing about. It implies a catalogue rather than a conquest. It implies that the AI voice market fragments into thousands of specific, verticalised experiences — Peak’s favourite example is a company building an answering agent for dog washers, which sounds like a joke until you remember that the App Store looked exactly that silly in 2008. And it implies that whoever runs the attachment layer is in a structurally better position than whoever builds the best individual agent.
The hardest integration problem in AI voice is not technical
The standard framing of a platform like this is integration aggregation: one integration instead of a thousand, one API instead of a bespoke project per operator. The experience providers quoted in the launch make that case well — SoundHound on the configuration friction of activating voice AI on an existing number, MiaRec on every service provider historically being its own project, Revmo on time-to-value falling from months to days.
But the aggregation that actually matters is regulatory, and it is the thing our Trust pillar has been pointing at all year.
Conversations are the most heavily regulated data in communications, for good reasons that predate every current acronym. Peak was direct about how Crux was approached: not by recording every call and handing it to whoever wanted it, but by building a trust framework into Alianza Crux from the start — consent announcements that vary by jurisdiction and by call type, because a caller who dials an AI answering agent and expects a machine is in a materially different position from three humans on a call whose transcript is being sent to a third-party analytics system.
Now multiply that by the thousands of conversational AI companies currently in the market. Each of them, today, must independently work out compliance with the EU AI Act, with the Illinois Biometric Information Privacy Act, with California’s privacy regime, and with the rules of every other jurisdiction they touch. Most of them are excellent at models and inexperienced at regulated voice. Peak described the terrain as a minefield of complexity, and having spent months of his own time in it, he would know.
This is the sharpest argument in the whole launch, and it is one the AI-native side of our ecosystem consistently underrates: solving consent once, at the network, on behalf of an entire ecosystem is worth more than solving connectivity once. Connectivity is a hard engineering problem. Consent is a hard liability problem, and liability is exactly the asset class operators have been accumulating for a century and a half.
Which brings us to the argument at the centre of the whole launch.
The second chance is real — and it is not the same chance
The framing Alianza built Navigate around is three eras. Era one was connection: a hundred and fifty years of the network’s job being to get two people on the line. Era two was cloud and applications, where the network carried rich communications services but the value accrued to the companies building on top of it — Teams, Zoom, the UCaaS platforms, the CPaaS platforms. Operators funded the roads and watched someone else open the shops. Era three, cognitive communications, makes the network responsible for understanding and acting on the conversations flowing across it. Peak’s question to operators is simply whether they intend to sit this one out too.
It is a well-worn argument in our industry, and it is usually made badly — as exhortation, as a slide with a burning platform on it. Made properly, it is the most important thing in this launch, and it deserves better than the shrug it often gets. Here is why we think this chance is structurally different from the last one.
Era two asked operators to become software companies. That was the entry price, and most of them could not pay it — not because of a failure of will, but because the capability required was the one thing a capital-intensive infrastructure business is least equipped to grow quickly. While the industry was buying spectrum and building towers, others were writing the applications. The gap was organisational, and organisational gaps do not close on a two-year roadmap.
Era three asks for something different. The assets that decide this one are the assets operators already hold and nobody else can assemble: the phone numbers, the identity behind them, the regulatory competence to handle conversations lawfully in every market they operate in, the physical placement to run inference close to where the conversation enters the network, and — the one that is consistently undervalued — an existing billing relationship with millions of small businesses who would rather buy from the provider they already pay than from a company they have never heard of. None of that is a software capability. All of it is a distribution and trust capability.
That is the real shape of the second chance. Not “telcos can build AI too,” which is the version that has failed repeatedly, but: operators do not have to build the intelligence at all. Thousands of companies are already building it, extremely well, and are currently stuck trying to reach customers one call-forwarding configuration at a time. The role available to operators is to be the substrate those companies cannot route around — the place where conversations, identity, consent and policy already converge.
There is still an uncomfortable edge to this, and we would rather say it than let a member discover it later. Sitting out era two was not a decision anyone consciously made; it was the result of not having anything to sell at the moment the market arrived. Repeating that outcome requires no effort at all. The second chance converts into revenue only for the operators who turn up with a packaged, priced, sellable offer while the ecosystem on the other side is still forming — which is a commercial readiness question, not a technology one, and it is the question far too few operators are currently working on.
Sovereignty is what decides whether this travels
For European members, the single most consequential thing Peak said was about where Alianza Crux runs.
There are implementation flavours, he said, designed to sit entirely within the operator’s boundary. Nothing leaves it. Nothing touches the United States. Nothing touches a hyperscaler. And he was careful to broaden the point past the obvious one — this is not a European carve-out, it is a requirement for anyone, anywhere, who wants conversations governed inside their own network.
Anyone who has sat through a European tier-one’s procurement process knows why that sentence matters more than any capability slide. Conversation data governed under someone else’s jurisdiction is not a feature discussion; it is a reason the deal never starts. If Alianza Crux can be domiciled in full — the governance layer as well as the data — then it is sellable in markets where American-domiciled control planes are not. That is a strategically significant position, and it is one very few of the AI voice platforms currently raising money can occupy at all.
It also connects to something we have been arguing for two years about where inference goes. Peak described conversations with operators who are planning to put NVIDIA-powered compute into central offices and at the network edge, so that inference happens where the conversation enters the network rather than in a distant region. For real-time translation, live coaching, anything where latency is the product, that placement is a genuine structural advantage — and it is one of the few AI advantages that cannot be replicated by writing better software. Sovereignty and physics are pushing in the same direction, toward the network.
The question the launch has not answered yet
We would be doing our members no favours by pretending this is finished, and to his credit, neither does Alianza.
Alianza Crux entered controlled availability last week, with a founding partner programme, a handful of experience providers integrated, and broader commercial availability planned for the first half of 2027. Peak was explicit that parts of the trust framework still need to be fleshed out and that the commercial models are still being refined so that the numbers work for the operator, the experience provider and the end customer simultaneously. That is an honest position for a platform that is genuinely early, and far more useful than a claim of completeness.
But it leaves the central question open, and it is the question our members will ask first: what does the operator actually earn?
The directional answer — stem the decline in voice revenue, keep customers who would otherwise take their communications spend elsewhere, and sell new services to businesses that would rather buy from the provider they have trusted for decades — is a strategy. It is not yet a revenue model. There is a real difference between an experience that appears as a line item on the operator’s own bill, and a referral fee on a relationship the experience provider owns. The first is a proper CSP motion with margin and customer ownership attached. The second is disintermediation with better manners. Which one this becomes will do more to determine the outcome than any architectural decision.
There is a second open question underneath it. Alianza Crux is designed for an ecosystem of thousands of long-tail experience providers, which only works if onboarding is close to self-service. A controlled availability programme with a small set of hand-integrated partners is the opposite of that by design. The moment those two things have to meet is the moment we find out what Crux really is — a marketplace, or a small number of deep partnerships wearing a marketplace’s clothes.
And there is a reach question. Today this sits on Alianza’s own infrastructure, which is a serious base of more than a thousand service providers, but it is not where most of Europe’s conversations live. Peak committed on the record to extending integration to IMS for mobile operators and to UCaaS platforms — wherever conversations exist, as he put it. Roadmap, not product. But it was a commitment, made in public, and we will hold it up against what ships.
Bring the question to Amsterdam
What makes this launch worth a considered commentary rather than a paragraph is that it is one of the few answers to our own research question that has infrastructure behind it. The market has spent a year claiming the orchestration layer rhetorically. This is a partner building the unglamorous parts — consent by jurisdiction, provisioning without porting, governance that can be domiciled — and being candid about which parts are not built yet.
Peak’s own framing of what the trust layer has to answer is the cleanest we have heard anywhere this year: who gets access, to what, and who said so. Three questions, no jargon, immediately legible to a regulator. We intend to borrow it.
Brian Beutler, Alianza’s CEO and founder, is with us on the main stage at CASA26 in Amsterdam next week, and the second chance is his argument to make — he has been making it to operators for years, and he will make it better than a blog post can. Our three pillars map onto it almost exactly. The conversation-as-asset case is Intelligence. Consent and sovereignty are Trust. Whether operators turn up with something to sell this time is Transformation, and it is the whole ballgame.
So the room should hear the argument. Then it should ask him the revenue question. He should expect it.
Watch the full CPaaSAA Talk with Dag Peak on YouTube. AI Voice: Who Will Run the Conversation? was published in March 2026.