Every few weeks, someone asks me: Are we in an AI bubble?
Honestly—we might be.
A bubble forms when valuations and expectations disconnect from real value creation. AI is showing classic signs: multibillion-dollar rounds with minimal revenue, startups valued on potential rather than traction and a sense that “AI-powered” is a business model in itself.
The most striking dynamic is the circular relationship between OpenAI and Nvidia. Nvidia invests in OpenAI. OpenAI buys billions of dollars in Nvidia GPUs. Nvidia’s stock surges because of the demand OpenAI generates. OpenAI’s valuation rises because investors see Nvidia betting on them. It’s not a Ponzi scheme in a legal sense. However, the self-reinforcing loop feels Ponzi-scheme-adjacent: new capital flowing in to sustain exponential expectations before sustainable business value is proven.
But here’s where eye care diverges from general tech hype.
Our problems are real, structural and worsening. And AI is not a novelty here—it’s increasingly a necessity.
Bill Gates said: “Most people overestimate what they can do in two years and underestimate what they can do in ten.”
AI fits this perfectly. The next two years will be full of friction. The next ten will reshape our field.
But we need to stay grounded about the barriers.
The Barriers
- Payment remains the biggest brake. In eye care, we already see this. AI diabetic retinopathy screening is FDA-cleared, accurate and scalable—but reimbursement is patchy, inconsistent and often not worth the operational lift. Autonomous systems for glaucoma or AMD? Not reimbursed yet. Even AI-enhanced OCT interpretation doesn’t have a payment pathway, despite saving clinicians time and improving consistency. Tele-refraction is another example. The technology is excellent. The patient experience is strong. But payment varies wildly across states and payers, slowing widespread adoption.
- Regulation moves slowly—and eye care is early in autonomy. The FDA is still figuring out how to regulate adaptive algorithms. In OCT and visual field interpretation, data drift is a real issue. Your model might work beautifully on Heidelberg data—but what about Zeiss? Topcon? Older platforms? Robotic cataract systems, like Horizon’s Polaris, are proving that AI-assisted surgery is possible—but regulatory scrutiny is high, and autonomy will advance in gradual steps, not leaps.
- Provider and patient trust must be earned Eye care is uniquely visual. Patients want to see what you see. Clinicians want explainability. When AI flags a subtle RNFL defect on OCT or predicts progression risk, doctors still want the “why.” And when AI recommends an IOL or triages a retinal image, patients need reassurance that their doctor—not an algorithm—is still guiding care.
So yes, the broader AI economy may be frothy. The Nvidia–OpenAI loop may feel circular enough to raise eyebrows. But in eye care, our problems—access, burnout, diagnostic overload, surgical precision—are too big to ignore.
The next two years will be messy. The next ten will be transformative.
AI may be experiencing a bubble.
Eye care, on the other hand, is experiencing inevitability.
Author
