Startup Forum

Across panels, a clear message emerged: in a world where models can one-shot features and code is cheap, moats now live in data, distribution, and domain depth. Founders who align product, platform, and go-to-market around those control points are separating from the pack.

Where defensibility now lives

Speakers agreed that shipping features is not a moat when capable models can replicate functionality. Durable control points vary by layer: proprietary and compounding datasets, embedded integrations and relationships, trusted enterprise deployments, and real distribution at the application layer. Value creation has shifted from code alone to changing user and industry behavior, building community gravity, and owning repeatable paths to customers.

For many teams, distribution strategy, product marketing, and ecosystem partnerships now demand as much founder time as the engineering roadmap. Community engagement and credibility with sophisticated buyers are becoming as critical as technical execution.

How early-stage companies are judged

Revenue thresholds have become less prescriptive, especially for infrastructure startups. Investors are prioritizing whether a team can credibly become the best company in an emerging category, show unusual speed, spark behavior change, and ignite a repeatable motion. Paying customers and early go-to-market flywheels still matter, but the deeper signal is a durable, specific vision for how the company wins.

Underhyped areas the panel highlighted included cost discipline driving adoption of small and fine-tuned models, a potential resurgence in the value of intellectual property, growing opportunity in formal verification, and a widening hardware, chips, and software optimization frontier for inference.

Build versus buy, data ownership, and agent-first products

Founders from security, go-to-market, energy, and AI-native search emphasized that differentiation increasingly stems from owning the data layer and its context. Examples ranged from a domain-specific foundation model for hydrocarbon operations to an AI-native web search API that indexes granular "atoms of information" for agents, an agentic go-to-market operating system, and a statistical approach to anomaly detection.

On platform choices, early teams favored paying a premium for the Databricks Platform to accelerate ingestion, processing, model serving, and governance rather than stitching open source components. That decision shortened enterprise procurement for some, opened doors with customers already on the Databricks Platform, and let small teams focus on their unique value. Talent strategies are evolving too, with business-oriented builders shipping production features alongside specialized engineers and researchers.

Challenge highlights and winners

The Built on Databricks Startup Challenge showcased three finalists built natively on the Databricks Platform: Intello, an agentic workforce for retail merchandising; LinkUp, a production-grade web search API for AI; and VisionHeight, a pre-attack intelligence platform mapping adversary infrastructure. VisionHeight won the grand prize, with LinkUp second and Intello third, illustrating how deep platform integration from ingestion through serving helps small teams operate at enterprise scale.

What to prioritize over the next year

Founders building durable AI companies are focusing on three things: own valuable data and the context that compounds it, design for distribution and trusted deployment, and use platforms that remove undifferentiated heavy lifting. The forum is worth watching for specific tactics on timing a Series A, structuring go-to-market, and translating platform choices into enterprise traction.