Shortly after a business meeting with a new startup, I participated in a panel discussion about AI. Interesting to me, both conversations were relatively the same. It was about startups, time to market, and the future. Sounds like the basis of any number of meetings I might have had over the last 10 years. But this one struck me as slightly different. A word that entered my mind based on how people are flocking to AI, and what that really means for any new business.

The simple word, with extreme psychological weight, was “dependency.”

Don’t get me wrong, despite AI being around for 30+ years already, this iteration is different. It’s more accessible. What the iPhone did to mobile connectivity, ChatGPT is doing for AI. But is it sustainable?

The biggest problem I see is that the technology is forming a dependency that is not sustainable. Also, mainly because it’s too new. We have no idea what will happen to OpenAI. Acquisition, competition, closure… and the problem exacerbates.

If your startup relies on OpenAI, and then OpenAI changes, most likely your company changes or closes as a result.

This formula has all the markings of a market collapse. A bubble, in the context of the dot-com bubble in 2000. If the rush to invest in AI outpaces the adoption and value of AI, we’ll see the same thing happen. And it’s probably closer than people want to accept. Two years, I’ll throw out September 2025, as this is when we begin reaching the end of a quarter, and as preparations for 2026 investments start to form, we’ll have a clear indication if it is sustainable.

But no matter if there is a pop or new AI winter, I don’t think our dependency will stop. We’ve passed that point already.

The Paradox of Progress

There’s something deeply ironic about our current moment with AI. We’re simultaneously experiencing unprecedented innovation and creating unprecedented vulnerability. Every startup pitch deck now seems to include “AI-powered” somewhere in the first three slides, as if those two words alone justify venture funding. But what happens when the magic stops working?

I keep thinking about those early conversations I had with entrepreneurs in the late 1990s. Everyone was going to be the next Amazon. Everyone had figured out how to “leverage the internet” to disrupt some traditional industry. The language was remarkably similar to what I hear today, just with different buzzwords. “Scalable,” “disruptive,” “first-mover advantage,” the same promises, the same breathless excitement, the same fundamental assumption that the technology itself was the moat.

We know how that story ended. Not with the death of the internet, but with the death of companies that confused access to technology with sustainable competitive advantage.

The Infrastructure We Don’t Control

What troubles me most about our current AI gold rush is the foundation it’s built on. When you create a business that fundamentally depends on OpenAI’s API, you’re not just building on someone else’s platform; you’re building on someone else’s entire value proposition.

This isn’t like building on top of AWS, where you’re purchasing infrastructure services that could theoretically be replaced with other cloud providers or your own servers. When your core product offering is essentially a wrapper around someone else’s tech, you’ve positioned yourself for collapse. Packaging problems don’t create sustainable businesses when the underlying product can be repackaged by anyone else, including the original creator.

I’ve watched this pattern before with mobile apps during the early iPhone days. Thousands of developers built businesses around simple utilities and games, only to watch Apple integrate similar functionality directly into iOS or change App Store policies that made their business models obsolete overnight. The difference is that mobile apps were typically small-scale ventures with limited investment. Today’s AI startups are raising millions based on the same fundamental dependency.

The Network Effects We’re Missing

The most successful technology companies of the past two decades built their moats through network effects, data accumulation, and platform lock-in. Google didn’t just build a better search algorithm; they built an advertising ecosystem that becomes more valuable as more people use it. Facebook didn’t just connect people; they created a social graph that becomes harder to abandon as more of your relationships live within it.

But what network effects are most AI startups creating? If your primary value is processing customer data through a language model, what happens when your customers can access that same language model directly? What happens when a competitor offers the same service with a different UI? What happens when the language model provider decides to compete with you directly?

The answer, in most cases, is that you don’t have a sustainable business. You have a temporary arbitrage opportunity that will close as soon as the market matures or the underlying technology becomes more accessible.

Beyond the Bubble

The most concerning aspect of our current situation is that we may be so dependent on AI tools that we can’t easily step back from them, even if the current business models prove unsustainable.

We’ve already integrated AI into our daily workflows, our decision-making processes, our creative work. Students are using it for research and writing, I’m seeing this at an all time high for my classes. Professionals are using it for analysis and communication. Companies are using it for customer service and content generation.

This level of integration happened remarkably quickly, far faster than our adoption of previous transformative technologies. And unlike previous technologies, AI tools often work so seamlessly that users don’t fully understand how dependent they’ve become until the tools aren’t available.

If there is a market collapse in AI, if funding dries up and startups shut down and access becomes more expensive or restricted, we won’t simply return to our pre-AI workflows. We’ll have to rebuild them, and we may find that we’ve lost some of the skills and processes we replaced with AI assistance.

The Path Forward

None of this means AI isn’t valuable or that the current innovations aren’t real. But it does suggest we need to think more carefully about sustainable approaches to AI integration, both as businesses and as individuals.

The companies most likely to survive an AI market correction are those building genuine competitive advantages that happen to use AI, rather than companies whose entire value proposition is AI access. The individuals most likely to benefit from AI in the long term are those who use it to enhance their capabilities rather than replace them.

Sustainability in technology isn’t about avoiding dependency altogether. It’s about understanding what you’re dependent on and having realistic plans for what happens when those dependencies change.

The question isn’t whether AI is sustainable. The question is whether our current approach to AI is sustainable.

I suspect the answer is no, and that we’ll need to learn this lesson the hard way.