Interview, Fireside Chat
Physics Gets a Vote: Nominal Cofounders on Hardware Development in an AI World
- Organizations anticipate a shift toward significantly increased hardware testing over the next two years to address data gaps, with a longer-term projection of a 10-to-20-year timeline for the full maturation of the physical AI ecosystem.
- The industry aims to evolve into "physical AI" entities where AI agents optimize the entire development lifecycle, linking design and test spaces to eventually minimize real-world testing, though 100% satisfaction via a single run remains distant.
- Companies plan to merge simulation outputs with real-world telemetry to create continuous data threads, moving away from local data storage to central, metadata-rich repositories to enable root cause analysis of field anomalies.
- The workforce strategy involves a disproportionate increase in software engineering roles, specifically doubling the count to build AI capabilities, with a focus on automating tedious data review and scaling human productivity from 50-person teams to single-operator multi-system control.
- Future testing methodologies intend to utilize AI agents and digital twins for faster-than-real-time validation, targeting a "nirvana" state of continuous learning loops, although current infrastructure for unit testing hardware is not yet established.
- The defense sector is accelerating AI integration in autonomous aircraft and cyber-physical systems, shifting from AI disqualifiers to active experimentation, including efforts like the "Cipher" initiative to maximize test knowledge in real time.
- Hardware companies face a race to collect physical data assets, with some firms expecting to build hardware internally to control the data flow, while others leverage AI to strip inefficiencies from existing designs.
- Capital-intensive hardware development is expected to remain a defensible market sector, attracting investment despite the distinct differences between physical and digital worlds that necessitate rigorous validation for AI-designed physical objects.
- There is a strategic push to incorporate unstructured operator audio data into training platforms and to version validation logic for deployment at the edge to perform on-the-line quality testing.
- Market conditions may lead to a reversal of the SaaSification trend as the ambition to build physical goods increases, requiring new data engineering practices to support the growing fusion of model outputs and real-world test data.