AI startups face a strategic fork. TechCrunch Disrupt 2026 will examine whether companies should build around open models, proprietary systems, or both as foundation models keep changing the rules.
The session “The Open vs. Closed AI Debate Is Just Getting Started” will focus on the trade-offs between open and proprietary AI. It will also ask the question that matters to every startup trying to survive the next product cycle: can either approach create a lasting competitive advantage?
The session takes place on the Builders Stage at TechCrunch Disrupt 2026. Its title is not subtle, but the underlying problem is — an open model can expand access and attract developers, while a proprietary model can offer control that competitors cannot copy with a download.
Nvidia sits on both sides of that argument. Its Nemotron 3 Super, launched in March, is an open 120-billion-parameter model designed for agentic workloads, and 145 papers accepted at ICML 2026 cited Nvidia’s Nemotron open models and datasets.
That research footprint gives Nvidia a clear reason to support open AI, even as the company remains involved in proprietary models. Nvidia CEO Jensen Huang framed the strategy in one sentence: “The future is not proprietary versus open, but proprietary and open.”
When the model vendor changes the plan
A second session, “What Happens When OpenAI Ships Your Roadmap,” will address a different threat to startup defensibility. Foundation model companies like OpenAI, Anthropic, and Google are releasing new capabilities that affect startup product strategies, turning a feature launch from a market event into a planning problem.
The session will take place October 13–15 at Moscone West in San Francisco, during TechCrunch Disrupt 2026. A startup can build a product around a capability today, only to see a foundation model company release a similar capability tomorrow — a tidy way to discover that your roadmap belongs to someone else.
This pressure makes model choice only one part of the decision. Startups also need to decide what they own, what they can defend, and what remains valuable after a larger company improves the underlying technology.
Nader Khalil brings a practical infrastructure view to that question. Khalil is Nvidia’s Director of Developer Tech and co-founded Brev.dev, an AI infrastructure company built around simplifying access to GPU infrastructure across different environments.
“Brev.dev was built around simplifying access to GPU infrastructure across different environments,” Khalil said. Nvidia acquired Brev.dev in July 2024. The deal connects the debate over model openness to a more basic startup concern: making powerful computing usable before infrastructure becomes another barrier.
Defensibility beyond the model layer
TechCrunch Disrupt 2026 will also feature “Learn how to build lasting AI defensibility at Disrupt 2026,” another Builders Stage session focused on staying valuable as models evolve. The session features Michel Tricot, Rob Toews, and Linda Tong.
Tricot is CEO and Co-founder of Airbyte, which has more than 7,000 customers, including 18% of the Fortune 500. Toews is a Partner at Radical Ventures, while Tong is CEO of Webflow. Their participation places startup strategy, venture capital, and enterprise software in the same conversation — because durable AI companies need all three perspectives.
The sessions share a central question: where does an AI startup’s advantage live? It may sit in an open model, a proprietary model, infrastructure, customer relationships, or the product layer built around changing models. The answer will not stay fixed just because a pitch deck says “defensible.”
Disrupt 2026 is a flagship startup event with over 10,000 attendees and more than 250 sessions. The event runs October 13–15 in San Francisco, with the open-versus-closed AI session listed for September 18 and the AI defensibility session listed for September 14.
Tickets are currently 25% off, and September 25 at 11:59 p.m. PT is the last day to register with up to $200 in savings. For startups choosing between openness, control, and a third option that borrows from both, the useful answer may arrive before the event does — but the debate is only getting started.
Based on
