AI · Sep 20, 2026
No one is surprised that Nvidia’s Jensen Huang thinks AI fears are overblownOpen or closed AI? Nvidia’s Nader Khalil and Sydney Sykes take on one of the decisions shaping next-gen startups at TechCrunch Disrupt 2026
Sep 18, 2026, 8:30 AM · TechCrunch
Nvidia’s Nader Khalil and Sydney Sykes headline a Disrupt 2026 Builders Stage session on open vs closed models as a live business decision—not a philosophy seminar.
Why it matters
TechCrunch Events previews “The Open vs. Closed AI Debate Is Just Getting Started” at Disrupt 2026 (October 13–15, San Francisco), led by Nvidia Director of Developer Tech Nader Khalil and Global Head of VC Partnerships Sydney Sykes. The pitch: founders choosing proprietary frontier APIs, open models, fine-tunes, local runs, or hybrids are deciding cost, infrastructure, margins, differentiation, speed, and control.
Open models are advancing quickly—Nvidia said in July that 145 ICML 2026 papers cited its Nemotron open models and datasets, with other Nvidia open families showing up in robotics, AV, and biomedical research—while proprietary labs keep pushing capability. Jensen Huang has argued the future is proprietary and open, not either-or.
Khalil previously co-founded Brev.dev (acquired by Nvidia in July 2024) around multi-environment GPU access; Sykes brings the venture lens. Nvidia’s Nemotron 3 Super, an open 120-billion-parameter model for agentic workloads launched in March, is already being combined with proprietary models in hybrid stacks.
From the desk
We’re with Huang’s “and,” not the tribal either-or—builders should pick the stack that matches data control and margins, then stay ready to swap.
Calling this a business decision is the useful frame. API speed gets you to market; open weights get you data gravity and exit options; hybrids are already how serious agent systems ship. The downside of endless optionality is architecture spaghetti and surprise bills when the “best” model rotates every quarter.
Useful AI spreads faster when open weights are good enough for the boring middle of the stack. I’m watching whether Disrupt surfaces concrete decision trees—workload, latency, compliance—or stays keynote fog, and whether Nemotron-class opens keep closing the gap on agentic tasks without locking teams to Nvidia’s full stack.
Context
TechCrunch published the event preview on September 18, 2026, with ticket discounts cited through September 25, 11:59 p.m. PT.
Who feels it
- Founders
- Treat model choice as margin and lock-in strategy, not brand affiliation.
- Investors
- Probe whether defensibility is data/workflow or a thin layer on someone else’s API.
- Developers
- Design for model swap; Huang’s hybrid future punishes hard-coded single vendors.
What to watch
- Concrete frameworks Khalil and Sykes offer on stage in October.
- Adoption patterns for Nemotron 3 Super in production agent stacks.
- Whether hybrid open+closed becomes the default architecture story for Series A decks.
Companies: NVIDIA