In Silicon Valley, the creed that “software eats the world” has been echoed for two decades. We’ve grown accustomed to the lightness of SaaS and to code automatically racing through CI/CD pipelines. Yet, beyond the spotlight, the physical world of steel and circuits that underpins our survival has long been stuck in the mud of digital transformation.Author and source: GD Gradient Descent
Recently, Sequoia Capital spoke with Nominal’s two founders, Cameron and Jason. This startup, whose background includes SpaceX and Anduril, is trying to address a seemingly unbelievable reality: in 2026, the most advanced defense and aerospace engineers are still using PDFs and screenshots to report test results.
Physics holds the final authority.
In recent years, a prevailing belief in the hardware engineering community has been “simulation omnipotence”: if we can simulate everything in software, do we still need expensive real-world testing?
The founder of Nominal gave a calm response: "Physics always has the final say."
Even the most perfect simulation cannot fully capture the complexity of the real world. When a drone flies in strong winds or a robotic arm grasps an unknown object, the "spikes" of software-defined hardware must meet reality.
Current industry pain points:
- Software Development: Code Commits, Automated Testing, Automated Deployment (GitHub workflow, already closed-loop).
- Hardware development: Sensor data collection, stored locally; engineers manually download and run analysis programs, then take screenshots for PowerPoint presentations.
This extremely inefficient process is precisely the chain Nominal aims to break.
Build the GitHub of hardware
Nominal aims to become the System of Record for hardware engineering. By building a unified platform, it ensures that hardware data is no longer siloed:
- Data structuring: Automatically catalog sensor telemetry data consisting of millions of data points per second to establish a "semantic layer."
- Verification logic reuse: Test logic developed during the R&D phase can be seamlessly reused in production and on-site operations.
- Full lifecycle traceability: Ensure transparent traceability for data from every screw to every ignition.
Just as GitHub has made code collaboration transparent, Nominal is making the process of "bending metal" as iterative as writing code.
When AI Meets Defense
The most exciting part of the interview was Nominal’s collaboration with DARPA (the U.S. Defense Advanced Research Projects Agency) on the CIPHER project.
Traditional testing is like a rigid spreadsheet: run under condition A, record data, deploy, analyze, then test condition B. This often takes years. Nominal is revolutionizing the game by introducing AI agents:
Real-time recommendation for the next optimal test point.Using digital twin technology, AI can predict faster than real-time which test point will yield the greatest knowledge gain. Testing is no longer a static matrix but a dynamically optimized gradient curve. This "autonomous testing" will significantly shorten the development cycle for cutting-edge hardware.
The moat of physical AI
Everyone is talking about AGI, but the risk levels of AGI in the virtual world and the physical world are entirely different:
- Virtual AGI: It can write a game for your child—even if it has bugs, it’s just a broken game.
- Physical AGI: If it creates a physical toy, you must ensure it does not cause physical harm.
Hardware test data is the only fuel for training physical AI. Nominal found that many AI training failures are not due to flawed algorithms, but because of poor data quality (damaged or uncalibrated sensors). Nominal’s automated anomaly detection identifies truly “high-fidelity” data, providing the real fuel needed for robotics and autonomous driving models.
At the end of the interview, the founder revealed an ambitious perspective: hardware is the only true moat.
As AI tools make software development increasingly inexpensive, the real barriers will return to the physical world. Nominal isn’t just selling software—it’s managing the “data supply chain” of the physical world.
The founder even hinted at the end that Nominal may very well enter the hardware manufacturing space itself—because only by controlling the entire chain, from sensors to final reporting, can one truly define "physical AI."
“All systems nominal,” once the most beautiful phrase in space launches, is now becoming the efficiency mantra of a new generation of hardware engineers.
As an AI market observer, which consumer industry do you think will be the first to experience an AI-driven hardware revolution?