Last Updated on April 30, 2026 by Taya Ziv
In 1987, Nobel laureate Robert Solow looked at the computer revolution sweeping American business and said something that made everyone uncomfortable: “You can see the computer age everywhere but in the productivity statistics.”
Forty years later, we’re doing it again. Only this time, the numbers are bigger and the denial is louder.
A massive NBER study published in February, covering thousands of executives across the US, UK, Germany, and Australia, dropped a finding that should have been front-page news everywhere: 80% of companies using AI report zero productivity gains. Not small gains. Not disappointing gains. Zero.
Let that sink in for a second. Then remember that Q1 2026 just set an all-time record with $300 billion in global venture funding, and 80% of it went to AI companies.
We’re pouring $242 billion into building AI tools that 80% of buyers say aren’t moving the needle. And somehow, the response from Silicon Valley is to pour faster.
The Numbers Nobody Wants to Put Next to Each Other
Here’s what makes this genuinely weird. It’s not that companies aren’t trying. 69% of firms actively use AI. Two-thirds of executives personally use it. The technology is in the building.
But the average executive uses AI 1.5 hours per week. That’s less than they spend in their Monday morning standup. These companies bought the gym membership and then stopped going after the first week.
Meanwhile, the funding world looks at this same market and sees a gold rush. $300 billion in Q1 alone. Four companies (OpenAI, Anthropic, xAI, and Waymo) absorbed $188 billion of that, or roughly 65% of every venture dollar spent on Earth in the first three months of 2026.
I’ve seen this movie before. Actually, I’ve seen it twice. Once with the dot-com bubble, once with crypto. The technology eventually delivers, but not before a lot of people lose a lot of money betting on it too early, in the wrong places, with the wrong assumptions.
Why This Is Actually Great News for Founders
I know what you’re thinking. “Liran, you just described a bubble. Why would I build an AI startup right now?”
Because the gap between “AI works in the lab” and “AI works in my company” is the actual opportunity. And almost nobody is building for it.
Think about it this way. The reason 80% of companies aren’t seeing results isn’t because GPT-4 or Claude can’t write good code or summarize documents. The models are fine. The problem is everything around the models.
Who’s training the sales team to actually use the tools? Who’s redesigning the workflow so AI output plugs into the next step instead of sitting in a chat window? Who’s building the connectors between the AI and the company’s actual systems, their CRM, their ERP, their messy spreadsheets that somehow run the entire operation?
Nobody. Or more accurately, not enough people.
The founders who are going to win this cycle aren’t building the next foundation model. We’ve already established that the vast majority of AI startups are building nothing that Google or OpenAI won’t ship for free next quarter. The winners are building the implementation layer. The boring, ugly, industry-specific work of making AI actually useful inside a company that still runs half its operations on email and Excel.
The Solow Paradox Playbook
Here’s why I’m actually optimistic. The original Solow Paradox, the computer one, eventually resolved itself. Computers did transform productivity. But it took 15 years after Solow’s observation for the gains to show up in the data. Why? Because companies had to completely redesign their processes, retrain their workers, and rebuild their organizations around the new technology.
The companies that helped with that transformation, the Accentures, the Microsofts of enterprise software, the integration shops, became some of the most valuable businesses on the planet.
We’re at the same inflection point right now with AI. The technology works. The market is about to undergo a brutal consolidation that will wipe out the pretenders. And what’s left standing will be companies that solve the adoption problem, not the technology problem.
40% of CEOs are worried about an AI correction. But corrections don’t kill markets. They kill bad companies and create space for good ones.
What Smart Founders Should Actually Build
If I were starting an AI company today, and I’ve told this to founders I work with, I would forget about building another model, another wrapper, or another chatbot. I’d look at the 80% and ask one question: why aren’t they seeing results?
The answers are specific and each one is a startup idea.
The first problem is workflow integration. AI tools sit outside the actual work process. Someone copies a prompt, pastes it into ChatGPT, gets a response, then manually moves that output somewhere else. Build the tools that eliminate that friction. Make AI invisible inside existing workflows.
The second problem is change management. Companies buy AI tools and then expect employees to figure out how to use them. That’s like buying a CNC machine and expecting a carpenter to operate it on day one. There’s a massive opportunity in AI-specific onboarding, training, and organizational design.
The third problem is measurement. With $300 billion flowing into AI this year, companies are spending real money on these tools. But they can’t measure whether they’re working. ROI measurement for AI deployments is practically nonexistent. Whoever builds the analytics layer that proves (or disproves) AI value inside a company will own a very important conversation.
The fourth problem is vertical specificity. A generic AI assistant isn’t helpful to a logistics coordinator, a compliance officer, or an insurance adjuster. They need tools built for their specific job, with their specific data, in their specific regulatory environment. That’s not glamorous work, but it’s defensible work.
The Uncomfortable Truth
I’ll be honest about something that founders don’t always want to hear. The NBER data doesn’t just expose a problem for the market. It exposes a problem for us.
If you’re building an AI product and your customers aren’t measurably better off after using it, you don’t have product-market fit. You have product-market fiction. And no amount of funding, no matter how much “AI” is in your pitch deck, will save you when the music stops.
The 80% number isn’t a condemnation of AI. It’s a condemnation of how we’ve been building and selling it. And for founders willing to do the harder, less glamorous work of making AI actually deliver results inside real companies, it’s the biggest green light you’ll ever get.
Because when 80% of the market is underserved, and $300 billion is chasing the wrong problem, the founder who chases the right one doesn’t need to be lucky. They just need to pay attention.
