I just spent two of the more energizing, thought-provoking days I’ve had in a while at RAISE Summit in Paris — 9,000 people at the Carrousel du Louvre, one of the biggest AI gatherings in Europe, three years old and already dwarfing most telecom events I’ve stood in. Great conversations, inspiring energy, and enough new names and ideas that I’m still sorting through my notes.

I went in with one question: where does my own patch of the world — telecom, cloud communications, and the AI layer now growing between and around them, the Bermuda Triangle I spend my working life inside — fit into this new AI world? Telecom alone has been building trusted, regulated, real-time infrastructure at national scale for well over a century. Cloud comms and AI are the newer arrivals in that triangle, and lately, the loudest.

I came out with more answers than I expected, a notebook full of new names, and one big takeaway that I’ll get to at the end. This is a long one. It has to be. There’s a lot going on.

The short version, if you only have five minutes:

  • RAISE is on its way to becoming a defining event. From 2,000 people to 9,000+ in three years, built on one simple idea: put founders and capital in the same room. The energy, youth, and diversity of that room is something telecom should see up close.
  • The money is staggering. North of $6.5 trillion in market value stood behind the sponsor list alone — more than the entire telecom sector invests in years. NVIDIA’s celebrated $1 billion investment in Nokia is a rounding error against the capital in that one room.
  • AI is rebuilding the whole stack — at a real cost. 10x software output is real (Cognition showed the receipts), and so is a 108% year-on-year rise in new security issues (Snyk showed the bill). Telecom’s decades of five-nines discipline are exactly the muscle this new world hasn’t built yet.
  • The customer experience layer was almost empty. When RAISE says “customer experience,” it means web experience. Real-time voice, contact centers, CPaaS, latency, barge-in — the world where humans actually talk — was nearly absent, except for a few exciting builders: ElevenLabs, Gradium, Featherless, Gladia, and one CPaaS exception, Telnyx, whose CEO I ran into walking the floor.
  • Sovereignty is becoming real, and open is winning. Mistral’s €20B talks, the AI Alliance’s Project Tapestry, federated learning efforts like Flower, SambaNova’s $1B raise with JPMorgan as an on-prem customer — the “bring the model to the data” pattern is everywhere. The open-source playbook (free core, value on top) is coming for AI, and that’s good news for anyone who isn’t a trillion-dollar company.
  • Telco billionaires are already moving. Xavier Niel — Iliad, Free, a 16% stake in Vodafone — is backing exactly the right things, including the most exciting voice AI company I found all week. Inspirational, and a signal.
  • The big question isn’t “how do telcos use AI.” It’s: what do telcos have that this new world needs and cannot build itself — and how do we connect these worlds so the answer gets built? That’s the work. That’s why CPaaSAA exists, and it’s exactly what we’re shaping CASA in Amsterdam this September around.

1. The room: younger, more diverse, and absolutely buzzing

Before a single session, the room told you something. Younger than any telco or comms industry event I’ve been to — twenties and thirties everywhere, a full generation removed from the fifty- and sixty-somethings who populate most telecom gatherings. Roughly 15% female by my own rough count walking the floor — still nowhere near good enough, but markedly better than telecom’s usual numbers, and honestly refreshing to see. And investors, everywhere — not tucked into side rooms doing quiet deals, but on stage explaining why they’d backed a particular company, walking the floor with notebooks out, comparing notes between sessions. This wasn’t an industry conference with some capital sprinkled in for color. This was where the capital actually lives right now, dressed as a tech conference.

That’s part of why I was there in the first place — with Sandbox, looking at this very intersection of startups, scaleups, and the investors backing them. That dynamic, more than any single keynote, is what makes an event like RAISE special: watching capital and founders find each other in real time, at a pace and density telecom events simply don’t have. It was also good to catch up with friends from Orange Ventures, KPN Ventures, Swisscom Ventures and others around the summit, comparing notes on what each of us is seeing across the industry — a useful reminder that the telecom investment community is smaller, and more connected, than the 9,000 people on the floor might suggest. And it makes perfect sense that they’re paying attention: for a telco venture arm, this room is next door, whatever the rest of the organization thinks.

Worth sitting with the trajectory, too. RAISE started as a place to bring startups and investors into the same room — nothing more exotic than that. Three years ago it was roughly 2,000 people. This year, more than 9,000. From what I saw on the floor, I wouldn’t bet against it doubling again before next July. At that pace this isn’t just a good conference anymore; it’s on a trajectory to become one of the defining events in the space. There’s real craft in what the organizers have built, and I’ll come back to what we might learn from it.

Is there a downside to all that youth? Honestly, a little. The energy is remarkable, the brightness undeniable — and the depth of experience with critical systems, with things that genuinely cannot fail, is thin. We’re in a phase where speed matters more than five-nines, and this generation is built for it. But somebody in the room should be thinking about what happens when this software runs the things that can’t go down. More on that below — because a few speakers were.

2. The money, and what it puts in perspective

I didn’t want a vibe about the scale of this, I wanted numbers, so I went and got them.

I added up the disclosed or public market valuation sitting behind RAISE’s full sponsor roster. It comes out north of $6.5 trillion — and NVIDIA and OpenAI alone account for roughly 85 to 90% of that figure. More market value stood behind two days at the Carrousel du Louvre than the GDP of most G20 nations — and that’s just the sponsors, before you count the money walking the floor as attendees.

Two comparisons worth making. First: that figure comfortably exceeds what the entire global telecom industry invests in capex over multiple years. Second, and sharper: NVIDIA’s billion-dollar equity investment in Nokia last October — real money, a genuine vote of confidence in AI-RAN, enough to move Nokia’s share price double digits in a day, celebrated across our industry as a landmark moment for telecom’s relationship with AI — is, set against NVIDIA’s own roughly $4.75 trillion market cap, a rounding error. A fraction of a percent. Telecom’s biggest AI headline of the past year barely registers as a blip against the capital sitting in that one room in Paris. That’s not a criticism of the deal. It’s a calibration of scale.

One honest caveat on the sponsor math: there’s a fair amount of AI-washing in that roster. A meaningful share of those sponsors are traditional data center, storage, and hosting players who’ve repositioned as “AI infrastructure” because that’s where the demand is. My rough read walking the floor: maybe a third of the sponsor list is genuinely AI-native, close to half is legacy infrastructure riding the wave, and the rest sits somewhere in between. But the startups and scaleups themselves — the hundreds of young companies presenting, pitching, demoing across the floor — those are real. That’s not washing. That’s a generation rebuilding the entire stack, and they were the best part of the event.

And the neoclouds were everywhere — IREN as the event’s headline sponsor, alongside Nscale, Crusoe, Nebius, CoreWeave, and more, all racing to build the large-scale, distributed, sovereign compute infrastructure that “secure European cloud” conversations kept circling on stage. Several of these companies barely existed in their current form two years ago; a few were mining crypto before they were building AI data centers. Now they’re headline sponsors at one of Europe’s largest AI events, moving at a speed that should make everyone in telecom sit up — because this is the layer being built right now, and it’s being built fast.

3. Who wasn’t in the room — counted carefully

Here’s where my two worlds diverged. Sponsors, top tier to bottom: zero telcos, zero cloud communications platforms — at least none I could identify by name across the full published list. Partners — roughly ninety organizations spanning media, French government bodies, VC funds — no telecom associations that I found. Speakers, roughly 350 of them: as far as I could tell, well under 1% with any telecom pedigree, Nokia and Deutsche Telekom being the names I could confirm.

The honest caveat, stated plainly: an event this size, spread across three parallel stages plus an invite-only CxO Summit I didn’t have access to, is too large for any one person to audit with certainty. I’m sure telco people were walking the floor — I know some were, because I talked to them. But in terms of visible presence — booths, stages, sponsorship, program — the telecom and cloud communications footprint at one of Europe’s largest AI events was vanishingly small.

I wasn’t alone in noticing. Philippe Ensarguet, VP Cloud & Software Engineering and an Orange Fellow — also Co-Chairman of Project Sylva’s governing board and a Board Advisor to the Linux Foundation, so sovereignty is very much his home turf — posted his own recap the day after we spoke on the floor: “I was struck by the total absence of the telco ecosystem in this AI fireworks display, no booths, no speakers on stage. This industry usually lives ten years behind the curve, and I’m afraid that gap is about to double.” His closing suggestion, in the same public post: RAISE needs “a customer-case track to connect this tech ecosystem to the business it’s meant to serve.” A senior Orange technologist, in public, arriving independently at the argument I’d been building all week.

And here’s the thing I want to be careful about: this isn’t an indictment. It’s a map. These are, quite simply, different worlds that have grown up separately. Telco people gather at their own events — the big mobile congresses, the industry forums — and talk mostly about networks and spectrum. Cloud comms people have their own circuit, focused on UCaaS, CCaaS, CPaaS, enterprise engagement. And now there’s this third world, enormous and young and moving at frightening speed, with its own events, its own capital, its own vocabulary. When RAISE speakers said “customer experience,” they meant web experience — browsing, searching, web data. The real-time world where actual humans talk to businesses — voice, contact centers, latency, interruption, the entire domain where telecom and cloud comms live — simply isn’t part of their mental model yet. Not hostile. Just not connected.

That disconnect is precisely the space we’ve worked from the start of CPaaSAA. It used to be a two-way bridge: telcos on one side, the CPaaS and application layer on the other. Now there’s a third corner, and the triangle is where everything interesting is happening.

4. Maybe it’s not a triangle. Maybe it’s layers.

Walking that floor, another way of seeing it kept forming. Maybe the cleaner picture isn’t a triangle at all — it’s a stack.

At the bottom: telecom infrastructure. Connectivity, networks, real-time transport, regulatory trust. Built over more than a century, distributed by nature — hundreds of regional operators that together, and only together, deliver global coverage.

At the top: the application layer. Customer experience, contact centers, engagement. This is where businesses actually meet their customers — and contact centers, when you strip away the vendor language, really are applications.

And in the middle, growing explosively: everything RAISE was actually about. Call it the runtime layer — where intelligence actually runs, and where orchestration, privacy, quality, and governance all live. Compute, data platforms, model training and inference, agent orchestration, security. This is where the value is visibly moving, and both of my worlds already know it: telcos and CPaaS players talk about moving up from connectivity and channels into intelligence, while the CX and cloud comms world talks about moving down from applications into orchestration and control. Our own AI Voice report makes exactly this argument — the interesting battle isn’t the model or the app, it’s the layer in between that decides where intelligence runs, how it’s governed, and who captures the margin. A remarkable amount of that layer was in the room at RAISE — the harness and orchestration players, the inference platforms, the data platforms we used to call data lakes, companies like Glean building enterprise AI platforms — all of it assembling in real time, with $6.5 trillion of conviction behind it. Just not nearly enough of it connected to communications.

Seen that way, CPaaS has a very natural place: it’s the API access layer — the way developers reach down into communications infrastructure, and increasingly the way they’ll reach into the intelligence layer too. Which is why it made perfect sense to run into David Casem, co-founder and CEO of Telnyx, walking the floor as an attendee. Telnyx — a CPaaSAA member — is a communications platform in plain terms: programmable voice, messaging, and network infrastructure for developers, the same category as Twilio or Infobip. And they’re leaning in hard. Casem’s argument, made publicly, is that production-grade voice AI breaks the moment you stitch it together across half a dozen vendors — telecom, speech-to-text, the model, text-to-speech, orchestration, compliance, each from a different provider — and that real-time AI agents need telecom, edge infrastructure, inference, identity, and agent primitives built as one integrated stack. Telnyx is reportedly running well over a thousand autonomous bots internally, treating AI-native as an operating discipline rather than a pilot. One CPaaS player, at least, has read the map and started walking.

5. Ten times the software, and the bill that comes with it

The most consequential thread of the whole summit, for anyone who cares about critical systems, ran through three separate talks that probably weren’t coordinated but might as well have been.

Nokia’s Pallavi Mahajan, Chief Technology and AI Officer, opened it on the Master Stage, arguing that Nokia had to become AI-native — not bolt AI onto the network, but rebuild the underlying muscle. A hardware company teaching itself to think like a software company, with culture, not compute, as the real constraint. It’s early days, and it was a short panel — but it matters that a telecom vendor of Nokia’s weight was on that stage at all, making that argument to that audience.

Cognition’s Scott Wu, on the same panel and in his own session, showed what AI-native actually looks like when a company fully commits. Engineering headcount up 1.4x over six months; shipped code volume up 10x. Eighty-nine percent of Cognition’s own code now committed by their AI agent, Devin, with agent-initiated work having quietly overtaken human-initiated work inside the company. The productivity revolution is not hype. The receipts are real.

And then Snyk’s Manoj Nair presented the bill. New security issues introduced across Snyk’s 4,800-plus customers: up 108% year over year and accelerating. His best line of the two days, quoting Jen Easterly: “We do not actually have a cybersecurity problem so much as a software quality problem.” Rubrik’s keynote drove it home from the operational side — non-human identities now outnumber humans 109 to 1 inside the enterprise, only 23% of organizations have full visibility into what their own agents are doing, and when attack is deterministic while defense is probabilistic, the question stops being whether you’ll catch the breach and becomes whether you can recover from it.

Here’s what struck me, sitting in a room where almost nobody came from telecom: this is the conversation telecom has institutional memory for. Decades of five-nines discipline, of understanding that speed without operational rigor breaks things at national scale, of engineering for recovery because failure is assumed. Right now, in this phase of the cycle, speed beats reliability and that’s a rational choice — you apply the new tools selectively, where breakage is survivable. But the pendulum will swing, because it always does, and when this software starts running things that truly cannot fail, the discipline telecom takes for granted becomes the thing this young, fast world needs to learn. That’s not nostalgia. That’s an asset waiting for a market.

6. Voice, latency, and the intelligent engagement gap

Now to the part of the stack I care about most — and the part that was most conspicuously thin on the ground.

Real-time voice is genuinely different from the rest of AI. A chatbot can think for eight seconds; a human on a phone call cannot wait eight seconds. Latency, accuracy, barge-in — the ability to handle a customer interrupting mid-sentence — these are make-or-break constraints that simply don’t exist for regular LLM workloads, and they’re precisely where telecom and cloud communications have decades of hard-won expertise. Agents are patient. Humans are not. Across two days of an AI summit, that distinction barely came up.

The builders who do get it were the most exciting finds of my week. ElevenLabs had the only meaningful voice-AI floor presence, and CEO Mati Staniszewski put numbers behind it live on stage: a $600 million revenue run rate, said out loud at the Louvre — after 20 months to reach the first $100 million, 10 months to the second, and 5 months to the third, with valuation talks reportedly around $22 billion. The growth, he said, is coming from voice agents finally getting reliable enough for enterprises to trust. And his strategic framing deserves quoting, because it’s the whole argument of this piece in one line: owning the conversation beats owning the model. ElevenLabs pays the frontier labs tens of millions a year for their models — and is betting the durable value sits in the conversational layer built on top. But the discovery that most energized me was Gradium — a spin-out of the Kyutai research lab, now backed by NVIDIA after a seed round past $100 million, building ultra-low-latency, multilingual voice explicitly aimed at customer care. Featherless AI is pushing an adjacent angle: open-weight text-to-speech launching this month at a fraction of incumbent pricing — “pretty good,” in their own words, which isn’t yet the bar a live call center needs but is a disruptive direction if quality holds. And Gladia, French speech-to-text built telephony-native with sub-300ms latency, is reportedly being acquired by OVHcloud specifically to strengthen Europe’s sovereign cloud strategy — voice and sovereignty converging into a single commercial argument.

Notice the pattern in who’s behind this. Kyutai — Gradium’s parent lab — was co-founded with backing from Xavier Niel, the man who built Iliad and Free, who now holds a stake in Vodafone, one of the true telecom entrepreneurs of our era. While the industry debates strategy decks, a telco billionaire is personally funding the voice-AI layer that telecom and CPaaS should consider home turf. I find that genuinely inspirational — proof that people who understand telecom economics deeply are betting real money on this intersection. Those are the kinds of people worth building alongside.

A direct word to my friends in cloud communications — the CCaaS, UCaaS, and contact center world — because this cuts both ways for you. The disruption is real and it’s aimed at your layer: when open-weight voice models cost a hundredth of today’s incumbents, when agents handle conversations that used to mean seats and licenses, the pricing logic underneath a lot of established cloud comms business models starts to wobble. Anyone in that world who watched this floor and felt comfortable wasn’t paying attention. But — and this is the part the RAISE crowd hasn’t figured out — you hold things none of those brilliant young companies have: deep integration into how businesses actually run, customer relationships built over years, the workflows, the compliance, the messy operational reality of enterprise engagement. Cheap voice synthesis is not a product. A trusted system that resolves a customer’s problem end-to-end is. The winners in this layer will be the ones who bolt the new intelligence onto that installed reality fastest — before someone from outside builds around them. Several of our members are already moving exactly this way. The window is open. It won’t stay open indefinitely.

Put it together and a concept starts taking shape that I keep coming back to: intelligent engagement. Not “customer experience” as the web world means it — browsing, searching, recommendation. Engagement that needs all three layers at once: communications infrastructure underneath, real-time voice and interaction in the middle, intelligence woven through, and a business outcome at the end. That’s the triangle expressed as a product category. Almost nobody at RAISE was building it. The handful who were, I made sure to meet.

7. Sovereignty, open source, and the value-on-top playbook

The other big thread of the week — and the one that gives me the most genuine optimism — is that nobody in that room wants a future where a handful of trillion-dollar companies and their founders control the intelligence layer everything else runs on. That instinct surfaced in three increasingly serious forms.

Open-weight models are the near-term answer. Mistral — three researchers out of DeepMind and Meta, Paris, 2023 — is reportedly in talks at a €20 billion valuation, nearly double September’s figure, with ASML as largest shareholder. Europe is pricing “open, sovereign, built here” as a durable category. It’s complicated — open weights alone don’t solve who funds the next generation of training — but here’s why I’m fundamentally optimistic about the open route: we’ve run this experiment before. Open source won the infrastructure wars. Linux, after thirty years of debate, turned out to be more secure than closed systems precisely because anyone can inspect it — and the commercial playbook that grew around it is well understood: the core technology becomes free and communal, and value gets built on top. Red Hat built a great business on a free operating system. The same logic is coming for AI: base model costs trend toward zero, open weights become inspectable and therefore trustable, and even if today’s open-weight suppliers stopped shipping tomorrow, the community would build its own — which brings me to Tapestry.

Project Tapestry, run under the AI Alliance, is the most interesting long-shot of the year: a broad coalition — companies, research institutions, public bodies, many of them otherwise competitors — working toward co-training open foundation models where participants contribute compute and data without ever surrendering the data itself. Yann LeCun, formerly Meta’s chief AI scientist and now the Alliance’s Chief Science Advisor, put the ambition plainly: “no single company or country should determine who can build on it, adapt it, or benefit from it.” It’s early, and federated training at frontier scale is technically unproven. But if it works, think about who it fits: telcos are distributed by nature. Hundreds of regional operators, each with deep local data, none with global scale alone, already accustomed to federating through industry bodies to achieve together what none can alone. Distributed, privacy-preserving, collectively-owned AI is almost a description of how telecom already works. Flower is pursuing the same principle from the open-source tooling side — train where the data lives, share only what’s learned. Watch both.

And the commercial proof arrived live on stage: SambaNova announced a $1 billion raise at an $11 billion valuation during the summit itself, with JPMorgan as a customer specifically for trusted, on-premises inference. A major bank keeping its AI close, controlled, on its own terms — the sovereignty argument, priced by serious capital, in real time.

Which brings me to the harder, more honest part of the sovereignty conversation for telecom — the utility trap. There’s a recurring instinct in our industry that goes: AI can’t run without our networks, so we deserve a revenue share. We tried this argument with Netflix. It’s a little like the water company demanding a percentage of the data center’s business because you can’t cool servers without water. True, and irrelevant: water is a commodity, priced like one, however essential it is. Connectivity, as long as we position it as pure utility — unlimited 5G for next to nothing — gets priced the same way. The way out isn’t demanding rent on the pipe. It’s finding the pockets where telecom holds something genuinely scarce: real-world data nobody else has legitimate access to — voice, location, behavior, the actual texture of human connection at population scale. Regulatory trust earned over decades. Distributed, sovereign, in-country infrastructure at the very moment the world decided it wants its AI distributed, sovereign, and in-country. The trillion-dollar question isn’t “how do telcos use AI internally” — that’s table stakes, and it’s happening. The question is what telcos can offer this new world that it cannot build for itself. Nobody at RAISE was asking it. That’s not a tragedy. That’s an open field.

8. The same week, everywhere else

Zoom out from the summit, and the industry didn’t pause for it. In the same seven days: Meta broke a three-year public silence — Zuckerberg’s first post since launching Threads — to announce its first paid model API, priced 75 to 86% below OpenAI and Anthropic’s flagships, a deliberate bet on commoditizing intelligence itself. Mistral’s €20 billion talks went public via Bloomberg. Apple sued OpenAI in federal court over alleged trade secret theft tied to a rumored first hardware device. Mark Cuban held court on the All-In stage about bubbles, utilization, and why AI models getting cheaper changes everything — a serial entrepreneur who’s seen every cycle, visibly energized by this one. Pricing wars, sovereignty land-grabs, espionage lawsuits, and billion-dollar raises, all inside one working week. This industry is moving fast enough that a two-day summit in Paris couldn’t contain its own news cycle.

9. So how do we contribute?

That’s the question I flew home with, and it’s a better question than the one I arrived with.

I went to Paris asking where telecom fits in the AI world. I left asking something more useful: these worlds — telecom at the bottom of the stack, intelligent engagement at the top, and this enormous, young, brilliantly-funded AI layer growing in between — are going to converge whether anyone plans it or not. The only question is whether that convergence happens by accident, slowly, with value leaking to whoever shows up first — or on purpose, with the people who hold the complementary pieces actually in the same room.

Building that room is what CPaaSAA has been about from the start: connecting telcos and the application layer, back when that meant CPaaS and APIs. The triangle just grew a third corner, and the mission grew with it. It’s what our work with Sandbox is about too — because a point that came up repeatedly in conversations this week, with operators and investors alike, is that the generalist capital filling that Paris room has little appetite for telco-specific complexity, which means the capital for this particular intersection has to be built deliberately, collaboratively, by people who actually understand all three corners.

And it’s what CASA in Amsterdam this September is for. I’ll be honest: RAISE raised my ambitions for it. There’s a thing or two worth learning from how they did this — 2,000 to 9,000 people in three years by staying relentlessly focused on putting builders and capital in the same room. Whether that’s a playbook to study or a bridge to build together, I don’t know yet — but it’s a conversation worth having, and I came home with a list of new connections to have it with: founders building voice AI that actually understands latency, investors who want exposure to exactly this intersection, operators quietly further along than their industry’s reputation suggests.

It was an awesome week. I learned a ton, reconnected with old friends, met a new generation that’s rebuilding the entire stack with an energy our industry should borrow — and came home more convinced than ever that the most interesting place to stand is exactly where these worlds meet.

See some of you in Amsterdam in September. Let’s build the room.