Physics Gets a Vote: Nominal Cofounders on Hardware Development in an AI World

Market Trends and Strategic Shifts

• Nominal is currently used by four of the top five US defense primes and major companies including Anduril and Corvette Racing.
• The industry is experiencing a macro "re-industrialization" trend, with a significant compression in hardware development timelines.
• There is a strategic "pendulum swing" back toward rigorous hardware testing as the sector recognizes the limitations of simulation-only approaches ("physics gets a vote").
• A generational wave of founders trained at SpaceX, Tesla, and Anduril is reinvigorating the hardware startup ecosystem.
• Hardware is increasingly viewed as a defensible investment moat due to its capital intensity and complexity.

The Core Problem: Data Fragmentation

• Historically, hardware engineering data was managed locally, fragmented across disconnected systems (simulation, prototyping, manufacturing), and often archived as PDFs or screenshots.
• The traditional workflow involves engineers downloading data to local machines for individual analysis rather than utilizing centralized, cloud-based platforms.
• Most defense and aerospace organizations lack a "system of record" for testing, leaving insights trapped in silos across test stands, labs, and flight ops.
• Unlike software, where GitHub solved version control and DevOps, hardware engineering lacks a unified infrastructure for test data management.

Product Strategy and Capabilities

• Nominal has chosen testing as its "wedge" strategy because it is the most iterative phase of the hardware lifecycle where ROI is most immediate and visible.
• The platform acts as a central hub to author, version control, and deploy "validation logic" (digital checklists) across design, manufacturing, and field operations.
• Nominal integrates real-world sensor telemetry with simulation outputs to create a continuous feedback loop for model training and validation.
• The company is currently deploying "verification agent" capabilities to automatically detect anomalies in physical data, such as sensor calibration issues or mechanical malfunctions.
• Users can now utilize natural language prompts (e.g., "plot the kinematics") to automate complex data analysis and reporting tasks previously done manually.

AI Integration and "Physical AI"

• The Defense Department is shifting from skepticism to aggressive experimentation with AI, particularly in collaborative combat and autonomous aircraft programs.
• Nominal is collaborating with DARPA and the US Air Force on the "CIPHER" project to use AI agents for real-time test optimization rather than static, sequential test matrices.
• The platform is moving toward "physical AI" by creating a semantic layer that catalogs hardware data with metadata, enabling AI to ingest and learn from diverse data sources.
• Forward-looking strategy includes using AI to generate "design agents" that are validated by "test agents," creating a closed-loop development environment.
• Jason and Cameron anticipate a future where "one human" using agentic tools can manage the testing equivalent of 50 systems, though full "vibe coding" of hardware remains distant.

Future Outlook and Vision

• The speakers predict a "race to collect data" as companies realize current models lack sufficient real-world training data for physical systems.
• The ultimate vision is a "GitHub for hardware" that allows for seamless iteration from design to production, minimizing the need for dangerous real-world testing through superior simulation.
• There is a long-term thesis that successful hardware companies will eventually become "Physical AI companies," where AI unlocks greater flexibility and versatility in physical product design (e.g., removing inefficient F-18 components).
• Nominal is exploring the possibility of building hardware itself to deepen understanding of the data supply chain, though they remain primarily focused on the software platform.
• The "nirvana" state involves AI agents continuously optimizing test vectors in real-time, dynamically adjusting digital twins to maximize learning efficiency.