For AI startups, choosing a model is no longer necessarily a one-time architectural decision nowadays.
On TechCrunch Disrupt 2026, four discussions will approach these decisions from different aspects of the AI technology stack, covering multi-model applications, custom models, infrastructure, and their underlying chips.
When one AI model is not sufficient...
When making a choice between open models and proprietary models, it is often assumed that a company must choose one of them first. However, an increasing number of AI products are now invoking different models for various tasks.
These discussions will explore why companies use multiple models, how to strike a balance between cost, performance, and flexibility, and when the performance of these models may surpass that of proprietary alternatives.
For founders, this flexibility affects not only model performance but also operational costs, product decisions, and how quickly the company can take advantage of better models when they become available.
Purchase Disrupt tickets to learn how companies decide which model is suitable for which task. Buying tickets now can save up to $100, and you can also get a 50% discount on the second ticket.
How much should your AI technology stack consist of?
Model selection also raises a larger question: what does your company need to have?
Oumi, the Chief Executive Officer and Co-Founder, will appear on Real World AI Stage to participate in a discussion titled " Which AI Should Your Company Actually Deploy : Rent, Customize, or, Build ". He will discuss whether startups should build their own models, when customization will become a competitive advantage, and how to make a choice between open AI and proprietary AI.
Through on-site voting, startup scenario analysis, and practical frameworks, Koukoumidis will help participants evaluate cutting-edge API, customized open-weight models, as well as fully proprietary AI solutions. Participants will take away three decision-making principles that can be used in their next architecture discussions.
The choice of founders simultaneously shapes both the product and the business model behind it. Building more AI technology stacks can bring greater control and differentiation, but it may also require more time, talent, and resources than relying on existing models.
Add this discussion with Oumi to your Disrupt agenda to obtain a practical framework for determining exactly how much AI technology stack a company needs. Purchase tickets now to be among the first to participate in this discussion.
What scenarios are AI and the proprietary AI suitable for respectively?
For some startups, the decision will still depend on the choice between open models and proprietary models.
Nvidia, the Technical Director of Developers, Nader Khalil, and Nvidia, the Global Cooperation Leader for VC, along with Sydney Sykes, will take part in Builders Stage to engage in a discussion titled “Building AI Startups Worth Betting On”. They will examine today's choices made by the founders, the trade-offs between cutting-edge API and open-weight models, and how these decisions affect product strategy and long-term differentiation.
These choices will have commercial consequences. The model selected will affect costs, infrastructure requirements, and the degree of control a company has over its product. It will also determine where a startup can create differentiation that is difficult for competitors to replicate, and how easily such differentiation can be achieved.
Purchase Disrupt tickets to learn how founders and venture capitalists influence model decisions. Buying tickets now can save up to $100, and you can also get a 50% discount on the second ticket.
When the architecture moves towards the hardware level
The model architecture is just a part of it. The performance of AI ultimately depends on its underlying hardware, and the advancements in AI have also begun to change the way these hardware components are designed.
Ricursive Intelligence Founder and CEO Anna Goldie, as well as Founder and Chief Technology Officer Azalia Mirhoseini, will discuss "When AI begins to design its own hardware" on Disrupt Stage. They will explore how AI can optimize chips and hardware, why model architecture is becoming more closely linked to hardware, and what an increasingly open AI ecosystem means for its underlying infrastructure.
For founders, the connection between AI and hardware may affect the speed at which new capabilities enter the market. Faster chip development could also change the infrastructure available to startups developing products for the next generation of AI.
Buy Disrupt tickets to find out what will happen when AI starts helping to design the hardware that supports its operation.
Learn how to maintain the openness chosen by AI on Disrupt 2026.
These discussions are just a part of the more than 200 sessions of Disrupt. The event also includes six industry stages, roundtables, and group discussions. The conference will be held at the Moscone West in San Francisco from October 13th to 15th. It is expected that over 10,000 founders, investors, operators, and technology leaders will attend, along with more than 250 speakers and over 300 startups exhibiting their products.
In addition, pairing sessions, transaction matching, and informal social interactions will also provide participants with opportunities to establish connections with founders, investors, potential partners, and other builders who face many of the same technical and business decisions.
Buy now TechCrunch Disrupt 2026 tickets to learn how the leaders of AI will address the choices of "what to build" and "how to build". Save up to $100 now, with a 50% discount on the second ticket. Unemployed individuals can purchase Expo + Pass for only $75.
