I’m going to say something that might make a few VCs choke on their kombucha: the "AI in radiology" space is a bubble. And it’s about to burst.
I’ve seen this movie before. I’ve been building and investing in Silicon Valley for over a decade. I saw the dot-com boom and bust. I saw the mobile gold rush. And now, I see the same patterns in AI, especially in healthcare. Everyone’s chasing the same shiny object, throwing money at anything that smells like a deep learning model for reading X-rays.
I get it. The idea is sexy. An AI that can spot tumors better than a human? It’s a story that sells itself. But having been on the front lines, both as a founder of two acquired companies and as an angel in over 200 startups, including some of the biggest names in AI like Anthropic and OpenAI, I can tell you that a sexy story doesn’t build a sustainable business. Especially not in healthcare.
The Uncomfortable Truth About Healthcare AI
Most founders in this space are making a classic mistake. They think the technology is the product. They spend millions and years of their lives building the most accurate model, publishing papers, and getting FDA clearance. They think that once they have the best algorithm, hospitals will beat a path to their door.
They are painfully wrong.
I remember meeting a team a few years back. Brilliant PhDs from Stanford. They had an algorithm for detecting a specific type of fracture that was, on paper, 10% more accurate than a human radiologist. They had the papers, the patents, the whole nine yards. They asked me for a $2 million seed round. I passed. They were shocked.
Why did I pass? Because they had no answer to the only question that matters: How are you going to get doctors to actually use this?
Their entire plan was "We'll sell it to hospitals." That’s not a plan. That’s a prayer. They had no idea about the brutal reality of hospital procurement cycles, the nightmare of IT integration, or the deeply ingrained workflows of radiologists. The model, as great as it was, was useless in the real world. That company burned through its pre-seed funding and died a quiet death a year later.
It’s Not About the Model
The secret that most AI founders miss is that in healthcare, the model is not the moat. It’s the table stakes. With the open-source tools and pre-trained models available today, a small, smart team can replicate a "state-of-the-art" diagnostic model in a few months. Your accuracy edge is temporary at best.
The real moats in healthcare AI are brutally non-technical:
- Distribution: How do you get into the clinical workflow? Do you have a partnership with an EMR giant like Epic or Cerner? Are you integrated with the PACS (Picture Archiving and Communication System) that radiologists live in all day? Without this, you’re just a fancy research project.
- Data: The quality and quantity of your training data is a far more durable advantage than your algorithm. But getting unique, well-labeled data is a street fight. It requires building relationships, navigating HIPAA, and often, a lot of manual, unglamorous work.
- Regulation: Getting FDA clearance is a long, expensive, and painful process. But it’s also a barrier to entry. The teams that figure out how to navigate this efficiently have a real advantage.
This is why I believe the current wave of "AI for radiology" startups is a bubble. They are focused on the wrong thing. They are competing on model accuracy, which is a race to the bottom. The winners won’t be the ones with the best algorithm. The winners will be the ones who build the best workflow product.
Where the Real Opportunity Lies
So, if I’m so bearish on radiology AI, where am I putting my money? I’m looking for the companies that are using AI to solve problems that are less sexy but far more impactful. The areas that are being overlooked in the gold rush.
1. AI for Mental Health
The mental health crisis is one of the biggest challenges of our time. And it’s a problem that is perfectly suited for AI. We can use AI to:
- Provide personalized therapy: AI-powered chatbots can provide CBT and other forms of therapy at a scale and cost that is impossible with human therapists alone.
- Identify at-risk individuals: By analyzing text and speech patterns, AI can identify individuals who are at risk of depression or suicide and connect them with help.
- Monitor treatment effectiveness: AI can track patient progress and provide real-time feedback to therapists, helping them to adjust treatment plans for better outcomes.
This is a massive, underserved market where the regulatory barriers are lower, and the potential for impact is enormous.
2. AI in Drug Discovery
Developing a new drug costs billions of dollars and can take over a decade. AI has the potential to completely revolutionize this process. Companies like Insitro (which I’m a proud investor in) are using machine learning to predict which drug candidates are most likely to succeed, dramatically reducing the time and cost of drug development.
This is a game of deep science and massive datasets. It’s not for the faint of heart. But the rewards, both financial and societal, are staggering.
3. AI for Diagnostics Beyond the Obvious
Instead of yet another lung nodule detector, I’m excited about companies using AI in less crowded diagnostic areas. Think about:
- AI for pathology: Analyzing tissue samples to detect cancer and other diseases.
- AI for genomics: Identifying genetic markers for disease and predicting patient response to treatment.
- AI for ophthalmology: Detecting diabetic retinopathy and other eye diseases from retinal scans.
These are areas where the data is complex, the need is high, and the competition is less fierce.
Stop Chasing the Bubble
Look, I’m not saying that AI has no place in radiology. It does. But the current approach is a dead end. We’re at the peak of inflated expectations, and the trough of disillusionment is coming.
My advice to founders is this: stop chasing the bubble. Stop trying to build a slightly better mousetrap. Instead, fall in love with a real, painful problem in healthcare. Go deep into the clinical workflow. Understand the user. And then, and only then, figure out how AI can be a part of the solution.
The next generation of great healthcare companies won’t be "AI companies." They will be healthcare companies that use AI to build a product that doctors and patients love. That’s a much harder path. But it’s the only one that leads to a real, lasting business.
Frequently Asked Questions
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.