The company that taught drones to survive Ukraine's electronic warfare just unveiled a fighter jet that launches straight up and lands on its tail.
Shield AI revealed the X-Bat today. The tail-sitting combat aircraft was designed from scratch around AI autonomy rather than after the fact. While the Air Force's Collaborative Combat Aircraft need a manned "quarterback" for command and control, X-Bat carries enough sensors and computing power to operate completely standalone. When communications cut out, it keeps hunting.
The technical approach: instead of bolting AI onto existing fighters, Shield AI designed the entire aircraft around what autonomous systems can do. Shield AI took thrust vectoring nozzle technology from 1990s F-15 experiments, paired it with an F100/F110 afterburning engine (the same family powering F-16s), and built guidance systems that let the aircraft transition from 90-degree vertical takeoff to horizontal cruise flight without human intervention. Then they wrapped it in low-observable shaping with a cranked-kite wing design optimized for range over dogfighting.
The specs sound implausible until you look at the math: 26 feet long, 39-foot wingspan, powered by a big engine in a small airframe. That gives it 2,000 nautical miles of range at high subsonic cruise speeds and a 50,000-foot ceiling, all while carrying internal weapons. It takes off using full afterburner thrust (thrust-to-weight ratio greater than one), then returns light enough to land without, avoiding the heat that would melt its recovery trailer.
The real story is what Shield AI's Hivemind autonomy software enables. The system already flies their V-Bat drones in GPS-denied environments across Ukraine, withstanding electronic warfare that downs other platforms. X-Bat scales that capability to combat jets. The aircraft navigates without GPS using a combination of technologies working in concert. It can accept tasking, report back what it finds, and execute missions autonomously when communication links fail.
Armor Harris, Shield AI's SVP who previously led SpaceX's Falcon 9 vertical landing development and Starlink engineering, framed the autonomy philosophy clearly: humans should be on-the-loop for offensive kill decisions, but the aircraft needs to operate independently when networks aren't available. Think defensive autonomous systems like Navy Phalanx CIWS, but airborne and hunting across 1,000-nautical-mile radius.
Why this matters for autonomy: X-Bat represents a fundamental shift in how defense systems are designed. Rather than adding autonomy to existing platforms, Shield AI built the aircraft around what their AI can do. That changes everything from inlet design (engineered for 90-degree angle of attack transitions) to mission systems architecture (sized for standalone sensor processing).
The vertical takeoff capability solves a problem China already exploited in wargames. In simulations, the U.S. loses more aircraft on the ground than in the air. Runway cratering and tanker denial keep conventional fighters from ever reaching the fight. X-Bat launches from a truck-mounted trailer, lands on moving ships, and operates from confined spaces. No runway, no catapult, no recovery net required.
Naval operations showcase the design's flexibility. An LHA amphibious assault ship could carry 60 X-Bats, turning it into an autonomous air wing carrier projecting power 1,000 nautical miles in any direction. Littoral Combat Ships get strike fighter capability. Even unmanned surface vessels could pre-position launchers deep inside contested zones for pop-up threats.
Shield AI has completed wind tunnel testing, radar cross-section pole model testing, and engine ground tests. Vertical takeoff and landing demonstrations are planned for late 2026, with full flights in 2028. Whether 18 months of development translates to actual flight hardware remains to be seen.
Harris's SpaceX background shows in the philosophy. Falcon 9 used 1960s gas generator cycle engines (not the world's most advanced rocket, but reliable, simple, and quick to develop). X-Bat follows the same pattern, bringing together mature technologies rather than pushing cutting-edge everything.
Cost targets the CCA range at $20-30 million per aircraft (Shield AI says around $27 million for X-Bat), roughly one-tenth the life-cycle cost of an F-35. Operating costs run 10x lower than fifth-generation platforms, making it affordable to actually fly rather than just warehouse.
The software update capability might matter most. Harris argues future conflicts will be won by whoever updates software fastest, citing Shield AI's experience in Ukraine where assumptions proved wrong on day one across multiple systems including Starlink and V-Bat. X-Bat's architecture prioritizes rapid iteration over making changes difficult.
Reality check: Competing against Lockheed Martin, Northrop Grumman, and Anduril (valued at $30.5 billion as of June 2025) won't be easy. But Shield AI's $198 million Coast Guard contract for V-Bat drones and deployments across Navy ships and Marine Expeditionary Units prove they can deliver operational systems, not just PowerPoint.
The open question: can a defense startup that revolutionized small drone autonomy scale that expertise to jet fighters? Harris makes the case that modern computer-aided design and AI development tools let teams of 2,000 engineers accomplish what once required 250,000. X-Bat will test that thesis.
The takeaway: Autonomy works best when you design the platform around the AI's capabilities from day one, not retrofit intelligence onto legacy systems. Shield AI is betting everything on that principle.
By The Numbers
2,000 nautical miles: X-Bat's range with full weapons load
50,000 feet: Service ceiling for high-altitude cruise
60 aircraft: Number of X-Bats that fit on one LHA amphibious assault ship
26 feet: Aircraft length (one-third the size of an F/A-18)
10x lower: Operating cost versus fifth-generation fighters
2028 Expected year for full flight testing
$20-30M Target unit cost, matching other CCA programs
Design Thinking: The Vertical Advantage
Three autonomy breakthroughs converge in X-Bat's vertical takeoff capability.
First, the guidance problem. Tail-sitting aircraft date to the 1950s X-13, but human pilots struggled landing backwards. Shield AI solved this by scaling up control algorithms from their V-Bat drone, which already performs autonomous vertical landings on moving ships in 25-knot winds. The software translates directly: same physics, different scale.
Second, the inlet challenge. Most fighters can't sustain airflow at 90-degree angle of attack without suffocating the engine. X-Bat's inlet uses heavy CFD modeling validated in wind tunnels to work in both horizontal cruise and vertical hover, with flight trajectories shaped to never blank the inlet during the cobra-maneuver transition from forward flight to tail-sitting descent.
Third, the perception problem. Autonomous vertical landing requires real-time sensor fusion to position precisely on the recovery trailer, all without GPS when jammed. Shield AI's Hivemind already does this routinely on ship decks. X-Bat inherits that capability.
The lesson: Solving autonomy problems at small scale creates reusable building blocks. Shield AI didn't invent vertical autonomous landing for X-Bat. They proved it worked on V-Bat in combat conditions, then engineered the fighter jet around that validated capability.
What I'm Watching
Can Shield AI actually build this at scale? Moving from 18-month development to factory production while competing with defense primes will test whether modern CAD tools really let small teams move this fast. Harris claims they've assembled aerospace's best talent, but talk is cheap until demonstrators fly in 2026.
How does DoD handle fully autonomous combat aircraft? Current policy requires human-in-the-loop for kinetic decisions, but X-Bat is designed to operate independently when communications fail. The gap between what the technology enables and what policy permits will force uncomfortable conversations about defensive autonomous weapons.
Will the vertical launch/recovery system actually work? Wind tunnel data looks good, but the real test comes when a jet-powered aircraft has to nail a precision landing on a truck-mounted trailer after a combat mission. This is the highest-risk piece of the entire program.
What happens to CCA pricing if X-Bat succeeds? If Shield AI delivers fifth-gen capabilities at one-tenth the lifecycle cost, it resets expectations for what autonomous combat aircraft should cost. That pressures everyone else to match or explain why their platforms cost more.
Forward this to someone building autonomous systems who needs to understand where defense autonomy is actually heading, not just what's getting hyped.
The Autonomy Report: Where physical autonomy meets real deployment.