Web Summit 2025 in Lisbon set new records: 71,386 attendees from 157 countries, 2,725 startups and 1,857 investors, a 74 percent jump in investor attendance year over year. It was the largest edition in the conference’s history. Almost every session, booth, and pitch had AI in it somewhere. By the third day, the label had stopped functioning as a signal at all. That is the more interesting story.
The novelty phase of generative AI is ending. The fastest-scaling software category in history is also the category enterprise buyers trust the least. An MIT NANDA study released over the summer, “The GenAI Divide: State of AI in Business 2025,” found that 95 percent of enterprise generative AI pilots have failed to deliver any measurable impact on profit and loss, despite $30 to $40 billion in enterprise spending. The report rattled markets through the summer and into the fall. By the time founders arrived in Lisbon in November, the question was no longer whether AI could ship a demo. It was whether anyone in procurement still believed the demo.
That gap, between what AI can do on stage and what it does in production, is now the defining tension of the market. Lisbon made it visible.
The saturation problem
The first phase of generative AI was theatrical. A chatbot passed an exam. A weekend prototype shipped. Anton Osika, the former particle physicist who runs the Swedish startup Lovable, told TechCrunch’s Connie Loizos on the Web Summit stage that 100,000 new products are being built on Lovable every day. British dictionary publisher Collins made vibe coding its word of the year for 2025.
That phase was useful because it expanded what people thought was possible. It also produced a lot of noise. Almost every company in Lisbon could now demo something that looked like magic. Almost none of them could explain why a customer should sign a multi-year contract for it.
If AI is no longer enough to make a company interesting, something else has to do the work. The product has to be clearer, the use case sharper, the team’s judgment visible on a problem the next ten startups using the same models do not understand as well.
The demo still matters. It is no longer enough.
What’s getting funded now
The capital markets have already started filtering for this. AI startups captured 52.7 percent of global venture funding in 2025, $270.2 billion of the $512.6 billion total, according to PitchBook and CB Insights data. Deal volume, meanwhile, kept falling. Investors wrote fewer cheques, larger ones, into companies they thought had a chance of surviving the next round of model releases.
Jake Flomenberg of Wing Venture Capital described the filter in TechCrunch’s December VC outlook: “I’m skeptical of moats built purely on model performance or prompting. Those advantages erode in months. The question I ask: if OpenAI or Anthropic launches a model tomorrow and is 10x better, does this company still have a reason to exist?”
That question is becoming a useful test for the whole AI cycle. The companies that survive the next two years will be the ones that can answer it. The companies that wrap a frontier model and call it a product probably will not.
The European angle most coverage skips
European tech gets reduced to regulation by reflex: slower, more cautious, more bureaucratic. Some of that is fair. But Atomico’s State of European Tech 2025 report, released the same week as Web Summit, makes a different case. European tech is now worth nearly $4 trillion, 15 percent of EU GDP, up from 4 percent in 2016. There are roughly 40,000 funded European tech companies, up from 13,000 a decade ago. Europe generates 17 percent of new global enterprise value.
It captures only 10 percent of exit value. That is the European problem, and Atomico locates the cause in capital flows, not regulation.
Tom Wehmeier, the report’s lead author and Atomico’s head of intelligence, framed the answer this way: “Sovereignty in technology isn’t about protectionism, it’s about agency and choice, building the capability, confidence and capital to shape the future.” If the next phase of AI is about trust earned through deployment in regulated, high-stakes contexts (healthcare, finance, public services, legal), Europe’s boring strengths in governance and domain expertise stop looking like a handicap. They become the surface where serious AI products will be judged.
That is also where 95 percent of enterprise pilots are failing. The MIT authors located the failure in integration, not in model capability. The companies that can close that gap, in markets that take governance seriously, may turn out to be Europe’s most exportable product.
What to watch in 2026
The temptation is to assume the AI race will be won by whoever ships the most, raises the most, and talks the loudest. Some companies will win that way. SoftBank’s $40 billion investment in OpenAI in March was the largest private tech round in history. Distribution and capital still matter.
Another kind of company is also worth watching. The one that knows exactly what problem it solves, does not pretend its model is magic, and can survive if a frontier lab ships something better next month. Boston Dynamics CEO Robert Playter, on the Web Summit center stage, was unusually direct about his own field: humanoid robots will not be entering private homes for at least a decade. That kind of honest scoping is rare in this market, and it is probably underrated as a competitive signal.
Web Summit 2025 was full of AI. That was not the signal. The signal was that AI is becoming too common to be the story by itself. What replaces it is the question for 2026, and going by the data out of Lisbon, record investor turnout, record deal concentration, record pilot failure rates. The answer looks like whoever can prove they deserve to be trusted with the next layer of business and work.
