Nvidia's Nader Khalil and Sydney Sykes to Debate Open vs. Closed AI at TechCrunch Disrupt 2026

Nvidia's Nader Khalil and Sydney Sykes will appear on the Builders Stage at TechCrunch Disrupt 2026 to work through the open-versus-closed debate that founders now face before they write a line of product code. The session puts two people from inside the chipmaker at the center of a question that shapes which models a young company can build on, what it pays to run them, and how much control it keeps over its own stack. TechCrunch confirmed the pairing as part of the conference's builder-focused programming.

The split matters because the two paths carry different economics. Open-weight models can be downloaded, fine-tuned on private data and served on a company's own hardware, which trims per-token API bills and keeps sensitive data in-house. Closed models from the large labs tend to arrive with stronger out-of-the-box performance, managed infrastructure and vendor support, but they lock a startup into someone else's pricing and roadmap. Nvidia sits on both sides of that line: its GPUs train the frontier systems and its software stack, including NIM microservices and the NeMo framework, is pitched at teams running open models in production.

Khalil and Sykes bring different vantage points to the stage. Khalil has worked on developer-facing efforts tied to Nvidia's AI platform, while Sykes has focused on the company's startup and venture relationships, giving her a read on what early-stage teams actually deploy. Putting an engineering-side view next to an ecosystem-side one should surface the gap between what the tooling allows and what founders choose under time and budget pressure. Neither has publicly staked out a fixed position ahead of the panel.

Disrupt 2026 lands in a crowded field of AI conferences, but the Builders Stage format is narrower than a keynote: sessions are aimed at people shipping products rather than at researchers presenting papers. That framing pushes the conversation toward practical trade-offs, such as whether a two-person team should fine-tune a small open model or rent capacity from a closed provider, and how that call changes as the company scales. For Nvidia, the subtext is commercial, since more open-model inference on its hardware means more demand for its chips and software.

What to watch after the session is whether the pair offers a clear rule of thumb or leaves the choice to circumstance. Founders will want specifics on cost per token, latency, fine-tuning effort and licensing terms, and those numbers shift with every model release. The panel's real value may be in naming the cases where open wins outright and the cases where closed still does, rather than declaring one side the default for the next generation of startups.