EraDrive, a Stanford spinoff with 23 years of heritage and NASA TRL 9 flight verification via the Starling swarm, is partnering with Northrop Grumman to provide AI autonomy as a service for satellites, enabling spacecraft to sense nearby objects, understand orbital context, and…
The satellite servicing business runs on a human bottleneck. Two Mission Extension Vehicles are operating in geostationary orbit right now, keeping Intelsat satellites alive past their designed lifetimes. They work. They are also each controlled by a team of ground operators who run every maneuver by hand. Northrop Grumman's Mission Robotic Vehicle, armed with a U.S. Naval Research Laboratory robotic arm and scheduled for its first flight later in 2026, will be able to do more: inspect, repair, install mission extension pods. But it faces the same constraint. One vehicle, one ground crew. To make that economics work, you need to fly a lot of MRVs. And that requires letting the spacecraft think for itself.
EraDrive is the Stanford answer to that problem. The company, a 2025 spinoff of the Stanford University Space Rendezvous Laboratory, announced a teaming agreement with Northrop Grumman at the 2026 Space Symposium in Colorado Springs this week. The pitch: an autonomy-as-a-service module that gives individual satellites the ability to sense nearby objects, understand their orbital context, and execute high-level goals with minimal ground supervision. The software-hardware kit draws from onboard cameras and edge compute. Customers can plug it into individual spacecraft or fleet architectures. Northrop Grumman gets an AI autonomy layer for the MRV pipeline. EraDrive gets prime defense contractor credibility and an on-orbit testbed.
EraDrive builds on 23 years of Stanford research, five distributed space missions, 13 satellites, and flight heritage on NASA's Starling cubesat swarm that the company says brought its autonomy stack to Technology Readiness Level 9. That is not a startup claim. That is NASA verifying something works in the actual space environment. EraDrive also holds a $1 million NASA sole-source contract to develop orbit tracking software using star tracker data from the Starling spacecraft. The company raised $5.3 million in an oversubscribed seed round in December 2025, led by Hackstack Ventures with participation from Point Nine, Brave Capital, and Entropy Industrial Capital.
The founding team has direct lineage to the problem they are solving. Chief Science Officer Simone D'Amico is the Stanford associate professor and Space Rendezvous Laboratory director who ran the Starling program. CEO Sumant Sharma is a D'Amico PhD alumnus who spent three years as autonomy lead at Wisk, Boeing's urban air mobility subsidiary. CTO Justin Kruger was the Starling postdoctoral fellow. Together they have operated formation-flying spacecraft in orbit. They are not explaining to satellites what autonomous rendezvous means. They have already done it.
The near-term focus of the Northrop collaboration is specific: AI-enabled pose estimation, safe integration with guidance navigation and control systems, and fleet-wide operationalization of autonomous servicing and inspection. Over the next year the companies will work toward demonstrating these capabilities in a mission-relevant context. The MRV, once operational, would be the platform. EraDrive's module would be the brain.
The market problem is real and worsening. The space industry expects 30,000 to 50,000 satellites in orbit by 2030, up from roughly 10,000 today. Current spacecraft autonomy is narrow. Satellites run maneuvers computed on the ground and replayed on board. They cannot react to an unexpected object, adjust formation without human input, or make a collision avoidance decision without an operator in the loop. When you have ten satellites, one ground operator per satellite is expensive but manageable. When you have a hundred, or a thousand, the model breaks. Every megaconstellation operator knows this. None of them has solved it.
EraDrive's approach is to add a module rather than redesign the spacecraft from scratch. The kit pairs optical elements with compute, timing, communications, and flight software in a compact payload. Satellites with the module can share vision data with other on-orbit modules and the ground, building what the company calls a space traffic intelligence system. Over the longer term, the roadmap includes cognitive spacecraft that operators could direct using natural language instructions. A satellite operator tells a spacecraft to de-tumble an unfamiliar object; the spacecraft decides how to do it safely.
The autonomy-as-a-service framing is deliberate. EraDrive is not selling a satellite. It is selling the ability to make any satellite autonomous, at scale, without each customer having to rebuild the autonomy stack from scratch. If that model holds, the economics of in-orbit servicing, debris removal, and persistent inspection become fundamentally different. You are no longer paying for bespoke human operation of each vehicle. You are paying for the software that lets the vehicle operate itself.
EraDrive is pre-revenue with one named commercial customer and a modest seed round for a hardware company. The MRV has been scheduled for first flight later in 2026; it has also been described as targeting a 2026 launch in previous reporting. The autonomy-as-a-service model has not yet been demonstrated at commercial scale in orbit. Northrop Grumman's interest suggests a supplier relationship more than a firm procurement commitment.
But the TRL9 heritage is the part that changes the calculation. This is not a pitch deck with a roadmap. This is software and hardware that has already flown on 13 satellites and done formation-flying in orbit under NASA's Starling program. The gap between "we have tested this on orbit" and "we are now integrating with the MRV" is real, but it is shorter than the gap between "we plan to test this" and "we have a prime contractor agreement." EraDrive is not asking anyone to believe the physics. The physics already worked.
What the Northrop teaming agreement signals is that the fleet era is close enough that a major prime defense contractor is willing to put a Stanford spinoff inside its robotic servicing vehicle. The question is whether the industry can build autonomous servicing vehicles cheaply enough to make the whole model pencil out. EraDrive argues the bottleneck is not physics. It is the absence of a standard module. Northrop Grumman is apparently betting the same.