I’ve just decoded my Amsterdam notes, and the five conversations that mattered most didn’t happen on a stand. They happened in aisles, over coffee, and at a networking event I nearly skipped. Here’s what that tells us about where broadcast is actually heading.

By Ben Anchor — Wednesday, 16 September 2026 · Listen to the podcast episode

I nearly didn’t go to the after-show drinks. A WhatsApp message the day before IBC, from a company I’d been circling for a year — polite emails, good intentions, nothing landed. But I went. And by the end of the evening, we’d both said out loud what twelve months of video calls hadn’t quite closed: we want to work together on the right projects.

That sentence sounds small. It isn’t. Because the broadcast technology world is tiny — the same few thousand people rotating between vendors, broadcasters, consultancies, integrators — and the mechanism by which projects actually get commissioned in this business is still trust. Trust is built in person. The 14 halls and 1,300 exhibitors are the excuse. The networking drinks around the RAI are the point.

I’ve just got back from IBC 2026 with a phone full of threads and a list of things that stuck with me, and not one of them was on the floor plan. The trade press will tell you this was the year AI stopped being a demo and started being a line item — 41,225 people, 170 countries, the usual superlatives — and that’s broadly true. But it’s not what I remember. What I remember are five conversations, and what each one says about where the interesting layer of this industry has moved.

The CTO Who Asks Where It Breaks

I had a drink with a chief technology officer I’ve worked for twice, both times in businesses under real pressure. Ninety seconds in, it stopped being a networking drink and turned into two engineers at a whiteboard that doesn’t exist. We went deep on the nowcasting model I’ve been building — the thing that tries to tell a broadcaster what’s happening with the audience now, rather than what happened last night.

A CTO who has run real infrastructure asks different questions. Not can it work, but where does it break? Who owns the data? What happens when the feed drops? How do you explain a number to a scheduler who has 30 seconds to make a decision? I came away with a list of things to change. That’s the best possible outcome of a drink.

What his questions told me is that the evaluation has moved from the demo room to the operations room. Nobody senior is impressed by a model anymore. They’re impressed by a model that survives contact with a live broadcast day. If you’re selling into broadcast in 2026, the question you must be ready for is not what does it do — it’s what happens at 3 a.m. when it’s wrong.

The Startup in the Next Aisle

I bumped into someone I’d met at a different show back in May. He’s building a company at roughly the same stage as mine, in an adjacent corner of the same problem. We’re not competitors, we’re neighbours. In about ten minutes, standing in an aisle, we identified two things: we’ve each already solved problems the other is about to hit, and there’s work that comes to each of us that would be better done by the other. We agreed to talk properly.

The number of first-time exhibitors was up 36% this year, and a lot of those are small companies. That tells you the layer below the big vendors is getting crowded. In a crowded layer, the winners tend to be the ones who collaborate rather than the ones who guard.

Is that naive? Sometimes. But broadcast isn’t a winner-takes-all market. It’s a hundred medium-sized deals a year, each with its own integrator, its own legacy stack, and its own politics. There’s more work than there are credible small vendors to do it. The risk for a startup in this industry isn’t that a peer steals a deal — it’s that you spend two years building something a broadcaster won’t buy because nobody who’d been inside one told you it wouldn’t fit. Peers are how you find that out early.

Sovereignty Stopped Being a Think Piece

I was getting a coffee, started chatting to the person next to me, and it turned out to be a fairly senior leader at one of the very large technology companies. Not a broadcast person, a platform person. We started talking about sovereign AI — the idea that a broadcaster, especially a public service broadcaster, shouldn’t have its editorial and operational intelligence dependent on a frontier model owned by a company in another jurisdiction.

Then the startup contact from May raised it independently. Then I walked past the EBU stand and there it was in the programme: ORF Innovation leading a partnership with the ARD cloud unit, EBU technology and innovation, and Schwartz Digits on a sovereign AI platform for public service media with locally hosted language models, speech-to-text, text-to-speech, and generative services. NVIDIA and the EBU announcing hybrid cloud architectures built to European public media standards. ATME, Scaleway, and Qvest showing a full streaming workflow in just-to-play-back on European infrastructure. There was a whole conference session literally titled From Open Models to Owned Intelligence.

Sovereign sounds noble, but is it actually a technology decision or a political one? It’s both. And the honest answer on budget is not yet, not entirely. Nobody I spoke to was proposing to rip out their cloud provider. What they were proposing was a hierarchy: keep the frontier models for the tasks where they’re clearly better; run open-weight or locally hosted models for anything touching archive, editorial judgement, audience data, or anything a regulator might one day ask about.

That’s where the IBC keynote line came back to me. Eileen van der Velden, creator of the AI actor Tilly Norwood, said Europe can’t win the AI arms race — that China and the United States will fight that one out — and so Europe wins by being more creative with the tools other people build. I wrote that down because it came back to me three times over the following days, in conversations that had nothing to do with acting.

If Europe can’t win the model race, the question for a broadcaster isn’t which model is smartest — it’s which model you’re allowed to depend on. That’s a procurement question, a legal question, and a resilience question. And IBC 2026 was the first show where I heard it asked by people who sign the purchase orders, rather than people who write the think pieces.

Audience Measurement Became Urgent

Number five is audience measurement, which is the least glamorous topic at any trade show, and for me, the most important. Because whatever intelligence you build for a broadcaster, it’s only as good as its knowledge of who is watching right now.

I had a set of conversations this year with people who live in that world: measurement companies, data people inside broadcasters, a couple of platform operators. And the mood has changed. A year ago, the response to real-time audience data was polite interest. This year it was, when can we meet?

One of the previews going into the show put it well: the same data set now has to improve ad pricing, power recommendations, and predict churn, and getting it to work in real time is still the hard part for most platforms. That’s exactly the gap. Everybody has data about last night. Very few people have a trustworthy picture of the last five minutes, and the ones who do are suddenly very interesting to advertisers.

The Action Has Moved to the Seams

So tie the five together. Networking. A CTO. A startup peer. Sovereignty. Measurement. What’s the single story?

The single story is that the interesting layer of this industry has moved. For 20 years, the action at IBC was in the big halls with the big vendors showing the big boxes. This year, the action was in the seams between them. Small companies solving specific problems. Broadcasters asking operational questions instead of capability questions. A trade body — the International Association of Media Tech, formerly IABM — renaming itself to admit that the boxes are gone.

And underneath all of it, a quiet, serious conversation about ownership. Who owns the model? Who owns the data? Who owns the relationship with the audience? Every one of my five conversations was, in the end, about one of those three.

Someone listening runs technology at a mid-sized broadcaster. They didn’t go. What should they take from IBC 2026?

Three things. One: when a vendor shows you AI, ask what happens when it’s wrong at three in the morning. If they can’t answer, walk on. Two: write down which of your workflows you’d be uncomfortable running on a model you don’t control. Because within 18 months, someone on your board will ask you that question, and sovereign options now exist. Three: find out how quickly you actually know what your audience is doing. Not overnight — now. Because that number is about to become the thing your commercial team argues about.

Every year, I ask myself on the flight out whether it’s worth the sore feet. And every year, the answer is the same. The show floor is the excuse. The conversations are the point. Five of them changed what I’m going to do in the next six months. And not one of them was scheduled.

That’s IBC.