Building the Physical AI Platform for the Industrial Base
ABOUT ARIYAN KABIR
Ariyan Kabir is the co-founder and CEO of GrayMatter Robotics. He started the company in early 2020 alongside two co-founders, building on PhD research conducted at USC’s Center for Advanced Manufacturing, where he and one co-founder studied the intersection of robotics and AI. The third co-founder, chief scientist Satyandra K. Gupta, has spent more than 35 years working at the intersection of AI, robotics, and manufacturing and sits on multiple national boards.
Ariyan’s framing of the problem is shaped by a realization that came from the academic environment: that manufacturing (long assumed to be “solved”) is in fact one of the largest unaddressed application areas for modern AI. The US is currently short half a million skilled manufacturing workers, a gap projected to reach four million in seven years and put $2 trillion of annual GDP at risk.
ABOUT GRAYMATTER ROBOTICS
GrayMatter Robotics builds the autonomy layer for high-mix, high-variability manufacturing, the roughly 90% of factory work that cannot be addressed by traditional pre-programmed robotics. The company’s robots perform tool-manipulation tasks: sanding, grinding, polishing, buffing, coating, blasting, and inspection. Underneath the robots sits the company’s real differentiator: a foundation model for materials and processes, trained on what Ariyan describes as the largest manufacturing process dataset in the world: 7 petabytes of multimodal sensor data spanning 14–18 modalities per cell.
The business is now six years old. This interview was recorded at the company’s 100,000 sq ft Physical AI Experience Center in Carson, California, which is the company’s fifth facility. Today, GrayMatter’s customers include Boeing, Raytheon, Oshkosh, Caterpillar, Riddell, the US Navy, the US Air Force, and Huntington Ingalls Industries, where GrayMatter is a partner on the HYPR (High Yield Production Robotics) program designing the next-generation American shipyard.
The company sells outcomes, not hardware. Robots are deployed on a subscription basis with multi-year commitments, and performance compounds quarterly through software updates, recipe upgrades, and hardware refreshes.
KEY TAKEAWAYS
1. Sequence the market. Commercial first, defense second.
GrayMatter spent its first three-plus years building almost entirely in the commercial world, despite the obvious dual-use relevance of factory automation. Their reasoning was that defense customers tolerate zero failure, and an unproven autonomy stack cannot debug itself inside an aircraft program. By the time the company turned toward national security work, they had already validated the technology, the business model, and the deployment muscle.
2. The business model can be the product.
Performance Composites — GrayMatter’s first customer — had been searching for a robotic sanding solution for five years before the conversation with GrayMatter. IT wasn’t so much a capability they were missing as it was a different risk appetite: manufacturers cannot tie up millions of dollars in Capex on unproven technology. GrayMatter’s response was to invert the deal entirely: zero upfront cost, deployment first, subscription payments only after the system was producing value in production. In exchange, the customer committed to a multi-year contract. The structure became the wedge that opened the entire Fortune 100.
3. Adoption became the geostrategic moat.
Asked how 7 petabytes of US manufacturing data compares to South Korea or Shenzhen, Ariyan reframed the question. The race isn’t about dataset size. The US ranks near the bottom of the top 10 in robot-to-worker adoption ratios; South Korea is #1 and China is climbing fast. Tonnage-wise, China has built more ships in the last 12 months than the US has built since WWII. This reframing has direct strategic implications: a defense industrial base strategy that focuses on technology sovereignty without obsessing over diffusion velocity will lose to one that does the opposite.
4. GrayMatter builds AI into every layer.
GrayMatter identifies eight distinct layers of engineering work that sit between a customer problem and a delivered outcome: industrial engineering, solution engineering, process recipe engineering, cell design, tool/fixture design, implementation, sustainment, and digital-twin operations. Most robotics companies put AI in one layer: the robot. This means the other seven can become bottlenecks. But GrayMatter is now deploying domain agents at every layer, coordinated by an orchestration layer the company calls Factory Super Intelligence (FSI). The focus on under-appreciated elements of the solution set has helped them differentiate.
5. Hire from the intersection. The mixing room doesn’t work.
GrayMatter’s most expensive hiring lesson was that you cannot assemble a great physical AI team by recruiting the best AI specialists, the best manufacturing engineers, and the best roboticists into a room. The translation cost is too high; each discipline optimizes for its own definition of the problem. What does work is hiring people who already live at the intersection of at least two of the three domains, then leveling them up on the third. The company’s head of customers came from Tesla. Their head of aerospace and defense was two steps below Boeing’s CTO. Their head of hardware also came from Tesla. This was a counterintuitive lesson, but one that stuck with me most from the conversation.
Follow Ariyan: https://www.linkedin.com/in/ariyankabir/
For more GrayMatter: https://factory.graymatter-robotics.com/
For more Crossing the Valley: www.valleycrossers.com
This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.valleycrossers.com
