Open or closed AI? How founders are choosing what to build on at TechCrunch Disrupt 2026

The Great Choice

Walking through a tech conference today feels different than it did three years ago. The air isn't just filled with the usual buzz of venture capital and networking; it is thick with a fundamental anxiety about infrastructure. The conversation at TechCrunch Disrupt this year has shifted from what we are building to what we are building on. For most founders, this boils down to a single, high-stakes binary: open or closed?

On one side, you have the closed-source giants. They offer the equivalent of a luxury high-rise apartment. Everything works. The plumbing is gold-plated. The security is top-tier. But you don't own the building, and the landlord can change the locks or double the rent whenever they feel like it. On the other side is open-source. It is more like buying a plot of land and a stack of lumber. It is harder to build, it might leak at first, but nobody can ever kick you out.

The Convenience Trap

If you are a founder trying to hit a six-month runway target, the closed-source route is incredibly seductive. Companies like OpenAI and Anthropic have created environments that allow a developer to go from an idea to a working prototype in a weekend. The performance is usually better right out of the gate. You aren't worrying about server clusters or fine-tuning weights; you are just calling an API.

But there is a hidden tax. I call it the Sovereignty Tax. When you build on a closed model, you are effectively outsourcing your core intellectual property to a third party. You are also at the mercy of their safety filters, their pricing tiers, and their eventual product pivots. If the model provider decides your startup's core feature is something they want to bake into their own next update, you are effectively dead in the water.

The Case for Open Sovereignty

This is why we are seeing a significant migration toward open-source models among the more battle-hardened founders. Building on something like Llama or Mistral isn't just about saving money on API calls. In fact, hosting these models yourself can often be more expensive in the short term when you factor in the hardware and DevOps talent required.

The real value is in the data and the control. If you control the model, you control the latency. You control the privacy. Most importantly, you control the future. You can prune the model to be exactly what you need, rather than using a general-purpose giant that is trying to be everything to everyone. For builders in highly regulated industries like fintech or healthcare, open-source isn't just a preference—it is becoming a requirement for compliance.

What It Means for Builders

If you are currently deciding where to plant your flag, you need to be honest about your business model. Are you building a wrapper, or are you building a foundation? If your value proposition is just a slightly better UI for a closed model, you aren't a founder; you are a glorified beta tester for Big Tech.

The founders who are winning the long game are taking a hybrid approach. They use closed models for rapid prototyping and testing market fit. It is the cheapest way to find out if anyone actually wants what you are selling. But the moment they find traction, they start the migration path toward an open-source stack. They are building a moat not with their code, but with their ability to operate independently of the giants.

Risk Management in the AI Era

We have to look at this through the lens of platform risk. Anyone who built a business entirely on the Facebook API or the Twitter API a decade ago knows how this story ends. When the platform's interests stop aligning with the developers, the developers lose every time. The AI landscape is no different. The closed-source providers are currently in their growth phase, playing nice with everyone. That won't last forever.

Founders need to ask themselves: if my model provider went offline tomorrow, or tripled their prices, would my business still exist? If the answer is no, you aren't building a company; you are building a feature for someone else's company. Open-source gives you the insurance policy you need to sleep at night.

The Takeaway

The choice between open and closed AI is really a choice between speed and sovereignty. Closed models give you the head start, but open models give you the finish line. The smartest builders I see are the ones who use the closed systems to move fast, while simultaneously building the infrastructure to make themselves obsolete on those very systems. Don't get comfortable in someone else's house. Start gathering your own lumber.

Read the original at TechCrunch Startups →