Open vs. Closed AI: How Startups Are Choosing Models at Disrupt 2026
Disrupt 2026 will examine how AI startups choose frontier APIs, open weights, custom models, multi-model stacks and AI-designed chips in San Francisco.
Summary
At Disrupt 2026, four sessions will examine startup choices spanning multi-model apps, frontier APIs, customized open weights, fully owned AI, infrastructure and chips. “The Real Tokenmaxxing: How the Best AI Companies Navigate a Multi-Model World” on the Builder’s Stage brings CapitalG partner Mo Jomaa, Together AI co-founder and CEO Vipul Ved Prakash, and Pathway CEO and co-founder Zuzanna Stamirowska together to weigh cost, performance and flexibility, including when open models beat proprietary alternatives. In “Which AI Should Your Company Actually Deploy: Rent, Customize, or Build” on the Real World AI Stage, Oumi CEO and co-founder Manos Koukoumidis will use audience polls, startup scenarios and a practical framework to deliver three decision principles.
“Building AI Startups Worth Betting On” on the Builders Stage features Nader Khalil, Nvidia’s Director of Developer Tech, and Sydney Sykes, its Global Head of VC Partnerships, examining how frontier APIs and open-weight models affect costs, infrastructure, control, product strategy and durable differentiation. In “When AI Starts Designing Its Own Hardware” on the Disrupt Stage, Ricursive Intelligence CEO Anna Goldie and CTO Azalia Mirhoseini will cover AI-optimized chips, tighter model-hardware links and whether faster chip development accelerates new capabilities. Disrupt runs October 13 to 15 at Moscone West in San Francisco, with 200-plus sessions across six industry stages, 250-plus speakers, 300-plus exhibiting startups and more than 10,000 expected attendees. Passes offer up to $100 off and 50% off a second pass, while a limited $75 Expo+ Pass is available to people affected by layoffs. Startups can preserve flexibility by switching from APIs to customized open models or distributing workloads among models, but ownership requires more time, talent and resources.
Positives
- Multi-model products can assign different workloads to different models, balancing performance, cost and flexibility.
- Oumi’s Manos Koukoumidis will provide three decision principles for renting, customizing or fully owning AI.
- Ricursive Intelligence will examine whether AI-optimized chip design can bring new capabilities to market faster.
- More than 10,000 attendees, 250-plus speakers and 300-plus exhibiting startups are expected at Disrupt 2026.
- People affected by layoffs can access a limited $75 Expo+ Pass, alongside broader pass discounts.
Risks & concerns
- Building more of the AI stack requires greater investments of time, talent and resources.
- Relying on existing models can provide less product control and differentiation than owning more of the stack.
- Frontier APIs and open-weight models carry different cost, infrastructure and performance trade-offs for each workload.
- Rapid advances make preserving architectural flexibility an ongoing requirement rather than a one-time model decision.