The $100M Bet on an AI Brain That Commands No Hardware

The largest defense-tech Series A in US history, $100 million, went to a company that builds no drones, no trucks, and no ships. Scout AI sells the intelligence layer that commands other manufacturers' robots. Investors paid more for that layer alone than most hardware startups in the sector have raised in their entire existence.

Embodied AI is the shorthand for models that perceive the physical world, understand instructions, and issue commands to machines. Fury belongs to a subclass called Vision-Language-Action (VLA) models. A VLA model takes in camera data and language, then produces motor actions, which is a different job from the text generation most people associate with AI. Founded in 2024 by Colby Adcock and Collin Otis, the startup applies that architecture to defense.

Why the biggest defense-tech Series A went to a company with no hardware

The round was announced in April 2026 and was oversubscribed. Total capital now sits above $115 million including the earlier $15 million seed round led by Align Ventures and Booz Allen Ventures. The company runs a 34-person team out of Sunnyvale, California, with backgrounds drawn from autonomous vehicles, robotics, and national security. Otis previously built autonomy systems at Kodiak Robotics and Uber ATG. Adcock sits on the board of the humanoid robotics company Figure AI.

The size matters because of what it signals. Defense-tech has attracted record capital in 2026. Investors poured more than $17.4 billion into defense startups in the first half of the year, according to Dealroom data cited by CNBC, already surpassing the $11.2 billion recorded for all of 2025. Anduril closed a $5 billion round in May. Quantum Systems raised $1.2 billion in July. Most of that money went to hardware makers. The $100 million that went to software alone was the outlier.

A foundation model that turns intent into action

Fury is a camera-only autonomy system. It runs on a single commercial off-the-shelf camera and a low-power inference chip, which keeps cost and power draw low while removing dependence on lidar or radar. The model was trained to interpret a commander's objective and break it into coordinated actions across heterogeneous vehicles. It is designed to keep working in environments where communications and GPS are denied, which is the hard case for any autonomous system.

The military has been promised true, one-to-many autonomy for years. Fury finally delivers it.— Collin Otis, CTO and Co-Founder, Scout AI

In February 2026, the company demonstrated the Fury Autonomous Vehicle Orchestrator live in Central California. A single operator issued a mission in natural language. The system generated the plan, submitted it for approval, then tasked an unmanned ground vehicle and multiple aerial systems, monitored their progress, and adjusted the plan as conditions changed. The demonstration ran on real hardware without scripted control.

Underpinning the demo is a simple operational problem. Legacy autonomy requires one operator per vehicle, or pre-programmed missions with limited adaptability. Neither scales. A unit that fields dozens of platforms needs a layer that translates high-level intent into per-vehicle tasks, handles timing and priorities, and re-plans when something goes wrong. That layer is the product.

The one-to-many math that broke the operator model

The economic argument is straightforward. One operator commanding a single teleoperated vehicle costs the same labor as one operator commanding a fleet. The entire cost curve of unmanned systems shifts when the ratio stops being one to one. Every platform becomes cheaper to run, and the value of the coordination layer grows with every vehicle added.

The evidence on the demand side is already on the books. In its first year, the company booked $11 million in Department of War contracts, won the Army's xTechOverwatch competition, and demonstrated an end-to-end autonomous mission executed by AI agents. In August 2025, it was awarded an Army Uncrewed Systems (UxS) autonomy contract, collaborating exclusively with Textron Systems for vehicle integration and with Edge Case Research for independent safety validation.

The open questions

Three risks deserve weight. First, integration. Fury adds intelligence to platforms built by others, which means the company depends on partners to deliver value, a model with inherent friction. Second, validation. Safety cases for autonomous operation are still being written, and Edge Case Research's involvement is acknowledgment that fielding autonomy in contested conditions carries certification hurdles. Third, competition. Hardware primes are building their own software stacks, and established defense contractors are acquiring or investing in autonomy startups directly.

The dual-use angle softens the downside. Camera-only, low-power inference is the same stack needed for industrial inspection, warehouse logistics, and infrastructure monitoring. If defense procurement slows, the technology transfers to commercial robotics without a redesign. That optionality is part of why the round attracted crossover and institutional participation.

Who owns the orchestration layer by 2027?

Development scenarios