Davos has always been a place where different worlds collide. This year, they barely made contact.
For those who’ve never been, a quick primer: Davos is a ski town in the Swiss Alps that has hosted the World Economic Forum (WEF) since 1971. To “go to Davos” means going during the week the WEF is in session.
The WEF has its own Congress Center in the middle of town. Security is tight. Invitations are reserved for heads of state, Fortune 500 CEOs, and recognized global thought leaders—plus their entourages. What happens inside the Congress gets covered extensively by international media.
But outside the Congress Center, there’s a parallel Davos that’s harder to see from afar. Hundreds of events run throughout the week, most of them not officially part of the WEF program. These are organized around “Houses”—branded spaces that corporations, nonprofits, and coalitions sponsor along the main drag, known as the Promenade. This year, Palantir had a House. Anthropic had a House. There was a USA House, a Ukraine House, an AI House sponsored by a consortium of tech companies and nonprofits. And beyond the Houses, a giant dome hosts speakers who draw crowds as large as anything on the official program.
If you are not a billionaire, a world leader, or traveling on a lavish corporate tab, your Davos experience is unglamorous. You split a small room with six people in a town a few kilometers away, or a hike up the mountain. You email relentlessly to secure meetings, land on VIP lists, and get into events. You’re cold. You’re hungry. And every warm space and restaurant has been converted into a private reception.
I offer this context because while the Congress proceedings are widely reported, the real texture of Davos is in what happens at the Houses. This is one of the biggest stages for international conversation outside of formal diplomacy. Sponsoring a House on the Promenade costs a fortune. People come with purpose.
This year, that purpose was unmistakable. It was written on the billboards: “We Help Determine Your ROI on AI.” “The Future is AI.” “Your Home for AI Solutions.”
AI wasn’t just a topic at Davos 2026. It was the topic. And the conversations about it revealed three disconnects that anyone thinking seriously about AI governance should be watching.
Disconnect #1: The ROI Question No One Could Answer
The most revealing moment I witnessed wasn’t a keynote or a fireside chat. It was a question from the audience.
During a panel on AI and enterprise adoption, a somewhat befuddled attendee asked the head of a major consulting firm to walk through one strong use case where a business had integrated AI and seen clear returns. The panelist—someone whose firm is betting heavily on AI consulting—didn’t have an answer. He hesitated, then pivoted. The room noticed.
This wasn’t an isolated moment. Across the Houses and stages, I kept encountering the same pattern: tremendous confidence that AI would transform everything, paired with striking vagueness about how, exactly, that transformation would generate profit. The phrase I heard more than once was “capabilities overhang”—the idea that AI can already do remarkable things, but businesses haven’t figured out how to translate capability into value.
This matters beyond Davos. If the gap between investment and realized value persists, we face one of two outcomes: a painful market correction when patience runs out, or pressure to deploy AI systems before they’re ready—rushing adoption to justify the spend. Neither is good for the people who will live with the consequences.
Disconnect #2: Are LLMs Overhyped or Underhyped?
If you walked the Promenade long enough, you’d encounter two completely different narratives about where Large Language Models (LLMs) are headed—sometimes in the same hour.
One camp argued that LLMs are overhyped. The real action, they claimed, is elsewhere: physical AI, robotics, world models that interact with real environments rather than generating text. LLMs are impressive, but they’ve hit a ceiling, or soon will. Scaling laws don’t go on infinitely. The enterprise applications are narrow. The AGI talk is overblown.
The other camp described a technology on the verge of flipping society upside down. The models that will be released in the next one to two years, they insisted, would be qualitatively different—capable of sustained reasoning, genuine autonomy, economic agency. We’re not in a hype cycle; we’re in the early innings of the most significant technological transformation in human history. God AI is not only possible; it’s almost here.
At most industry gatherings, expert disagreement is normal and even healthy. But the AI discourse at Davos felt different—less like a debate between informed perspectives and more like people describing different technologies entirely.
For those of us working on governance, this presents a real problem. How do you create governance frameworks for a technology whose near-term trajectory is genuinely contested? How do you build public trust when the people building the systems can’t agree on what they’re building?
Disconnect #3: The Conversation in the Other Room
The third disconnect was spatial. You could literally see it on the map.
A few Houses—Human Change in particular—were focused on navigating the AI transition with real solutions. They hosted programming on children, on education, on the ethical obligations of developers, on what a good AI future might actually look like. By the accounts of those I spoke to, and the sessions I attended, the conversation was substantive. Thoughtful. Even hopeful.
What was striking was that participants in those rooms openly acknowledged that their conversations bore little resemblance to what was happening on the rest of the Promenade. The humanist track and the commercial track weren’t in tension. They were barely in contact.
This is, in some ways, the oldest story at Davos. The business of the world gets done in the main rooms; the questions about whether the business is good for people happen in the side rooms. But with AI, the stakes of that separation feel different. We’re talking about a technology that very well may fundamentally restructure the economy and the nature of labor itself. The people asking “how do we make sure this goes well for humans?” shouldn’t be in a side conversation.
What This Means
These disconnects aren’t matters of Davos gossip. They’re governance challenges, and they’re connected.
If the ROI question remains unanswered, pressure will build: either to correct the market or to force deployment before systems are mature. If the people building AI can’t converge on what it can and can’t do, regulators will struggle to set meaningful guardrails—and the public will struggle to know what to believe, weakening trust further. And if the questions about human benefit and risks remain sidelined, the technology’s trajectory will be shaped by commercial incentives alone.
What we need is governance infrastructure that can operate under uncertainty: frameworks that don’t require us to know exactly where AI is heading before we can act—like a marketplace of independent verification, where third-party experts regularly update standards, criteria, and tooling, and other trust mechanisms.
The billboard version of Davos—AI as an inevitability you simply buy into, and profit from—isn’t adequate. Neither is the side-room situation, where the humanist questions get asked but never reach the main stage. The work is to bridge that gap: to bring the public into the conversation about where AI is heading, understand what trade-offs they’re willing to accept, and make publicly mandated questions about human benefit central to how this technology is developed, deployed, and governed.
Resolving the Disconnect
Davos is a strange place to look for clarity. It’s a gathering defined by competing agendas, strategic messaging, and the performance of confidence. But sometimes the performance reveals more than it intends.
What I saw this year was an industry at an inflection point: publicly triumphant, privately uncertain. The money is real. The capabilities are real. But the path from here to a future that’s actually good for people is far from obvious, and the people asking the hardest questions are not always in the rooms where decisions get made.
But inflection points are also openings. Whether we resolve these disconnects will determine whether we are successful in shaping the AI future—not just for the industry, but for all of us.
Onward,
Andrew