AI in Military Drones 2026: Anduril, Shield AI, Helsing

Quick read: Military drones in 2026 are a different industry from the consumer drones at Best Buy. Four AI-first companies dominate the conversation: Anduril Industries ($61B valuation, $20B Army contract), Shield AI ($12.7B valuation, V-BAT with Hivemind AI pilot), Germany’s Helsing (~$18B valuation, Centaur AI fighter pilot), and Eric Schmidt’s Swift Beat (formerly White Stork, supplying Ukraine). The US Pentagon’s Replicator initiative transitioned to the Defense Autonomous Warfare Group in 2025 and now has a $70B+ FY27 budget request. Five distinct technologies dominate counter-drone defense: kinetic interceptors, electronic warfare, directed energy, cyber takeover, and AI-detection radar.
The point: AI in military drones is the highest-growth defense-technology category in 2026. The numbers are real, the companies are mostly new, and the technology overlap with civilian drone systems is closer than most coverage acknowledges.
Who needs this: Anyone tracking AI’s real-world deployment, defense-tech investors, civilian drone professionals trying to understand the upstream technology trends, and curious general readers who keep seeing these company names without context.
Skip if: You work in defense procurement professionally. Daily AI fundamentals in our free Beginners in AI newsletter.

This is the fastest-growing corner of the AI industry that most consumer AI coverage doesn’t touch. The valuations are real. The contracts are real. The technology overlap with civilian autonomous-flight systems is closer than most readers realize.

What follows is the factual landscape — companies, capabilities, dollars, and the AI mechanics — with no political framing. The story is interesting on its own.

How modern drone warfare changed between 2022 and 2026

Three shifts redrew the defense-drone industry in four years.

1. Cheap drones at scale changed the cost equation. A first-person-view drone built from off-the-shelf parts costs $400 to $2,000. The targets these drones disable can cost millions. The ratio is the most-discussed economic fact in modern defense procurement.

2. AI-on-board became table stakes for serious operators. GPS-denied flight, computer-vision target ID, and autonomous mission completion are the difference between a drone that works under electronic-warfare conditions and one that doesn’t. Ukraine has been the proving ground. Shield AI’s V-BAT, for example, has logged over 130 sorties in Ukraine since June 2024 specifically because the Hivemind AI pilot keeps the aircraft mission-capable when GPS and communications links go down.

3. The procurement system started catching up. The US Department of Defense launched the Replicator initiative in 2023, transitioned it to the Defense Autonomous Warfare Group (DAWG) in 2025, and requested up to $54.6 billion for FY2027 (against $225.9 million in FY2026). The Trump administration’s combined drone and counter-drone allocation tops $70 billion for the next year — the Pentagon’s largest-ever investment in unmanned systems.

The combination of these three shifts is why a handful of AI-first startups now hold defense-industrial-base positions that took the legacy primes decades to build.

The new defense AI companies

Four companies dominate the AI-defense-drone conversation. Each has a distinctive technical bet and a distinctive founding story. None of them existed in their current form before 2017.

Company	Founded	Valuation (2026)	Headline product	Notable contract / deployment
Anduril Industries	2017	$61B (May 2026)	Lattice OS + Roadrunner	$20B 10-yr Army enterprise contract (Mar 2026)
Shield AI	2015	$12.7B (Mar 2026)	Hivemind AI pilot + V-BAT	130+ Ukraine sorties since Jun 2024; CCA award
Helsing	2021 (Munich)	~$18B (May 2026)	Centaur AI fighter pilot + HX-2 munitions	€269M Bundestag contract; Eurofighter integration
Swift Beat (White Stork)	2023	Private, undisclosed	$400 AI kamikaze drones	Hundreds of thousands of drones shipped to Ukraine

Anduril Industries (Palmer Luckey)

Founded in 2017 by Palmer Luckey, the same person who built Oculus VR and sold it to Facebook in 2014. Luckey was forced out of Facebook in 2017 over political donations; Anduril was his second act. The company is now the largest by valuation in defense-tech: $61 billion as of the May 2026 $5 billion funding round.

The core product is Lattice OS — software that integrates drones, sensors, radar, surveillance towers, and counter-UAS systems into a single command interface. Lattice is what the military buys; Anduril’s drones (Ghost, ALTIUS-700) and counter-drone systems (Roadrunner, Pulsar) are what Lattice operates.

Three contracts define Anduril’s 2026 position:

• A 10-year enterprise contract with the US Army, ceiling $20 billion, signed March 2026. The contract consolidated more than 120 separate procurement actions into a single framework.
• Roadrunner-M counter-drone interceptors: $350+ million in orders to date, including a $250 million package for 500+ units paired with Pulsar electronic warfare systems.
• The company is the leading contender for the $22 billion IVAS (Integrated Visual Augmentation System) program, which the Army has been moving away from Microsoft.

Shield AI (Brandon Tseng)

Founded in 2015 by brothers Brandon and Ryan Tseng plus Andrew Reiter. Brandon Tseng is a former Navy SEAL whose operational experience shaped the company’s technical premise: how do you fly a drone when GPS is jammed, communications are severed, and you can’t see what’s on the other side of a wall?

The answer is Hivemind — an AI pilot that runs on the drone, navigates by camera and inertial sensors rather than external signals, and decides on its own where to look next. Hivemind powers Shield AI’s V-BAT vertical-takeoff reconnaissance drone, which has logged 130+ combat sorties in Ukraine in highly contested electronic-warfare environments since June 2024.

2026 highlights: $1.5 billion Series G funding plus $500 million in fixed-return preferred equity, valuing the company at $12.7 billion (a 140% increase from a year prior). Projected 2026 revenue: at least $540 million, with 80%+ growth over 2025. Shield AI also acquired Aechelon for visual-simulation training infrastructure. The company recently unveiled the X-BAT, a jet-powered autonomous vertical-takeoff fighter aircraft.

One notable internal contradiction: V-BAT now has a public weapons deal, but co-founder Brandon Tseng has publicly argued against armed-drone proliferation. The company says it’s consistent — armed V-BAT requires explicit operator authorization for any kinetic action. The debate over whether that’s sufficient is ongoing.

Helsing (Niklas Köhler, Gundbert Scherf, Torsten Reil)

Europe’s AI-defense champion. Munich-based, founded 2021. Currently raising a $1.2 billion round led by Dragoneer at roughly an $18 billion valuation — which would make Helsing Germany’s most valuable startup.

Two product lines drive the company:

• Centaur, an AI fighter-pilot system. In May/June 2025, Helsing and Airbus conducted “Project Beyond” — a real-world combat trial in which Centaur flew a Gripen E in beyond-visual-range combat against a human-piloted Gripen D. The outcome details aren’t fully public, but the trial happened.
• HX-2 loitering munitions. In February 2026, the German Bundestag’s budget committee approved an initial €269 million contract for HX-2, with a framework that can grow to €1.46 billion over seven years.

Helsing also signed a three-digit-million-Euro contract with Saab Germany in November 2025 to integrate the company’s Cirra AI software into the Arexis electronic-warfare suite on Eurofighter Typhoon EK aircraft over three years.

Swift Beat (formerly White Stork; Eric Schmidt)

Started by Eric Schmidt — the same Eric Schmidt who used to be Google’s CEO — after Russia’s 2022 invasion of Ukraine. Originally named White Stork, then briefly Project Eagle, now operating as Swift Beat. The corporate parent is Tallinn-based Volya Robotics OÜ, of which Schmidt is the sole beneficiary.

The product is mass-produced AI-powered drones for Ukrainian forces — single-use explosive drones with onboard computer vision that can identify targets in GPS-jammed environments. The company is targeting hundreds of thousands of drones per year. Per the Ukrainian government, that number is expected to increase further in 2026.

The Ukraine context made Swift Beat the most-watched non-traditional entrant in defense in 2024–2026. Schmidt himself has publicly framed the venture as a continuation of his work on Pentagon defense advisory boards (he chaired the National Security Commission on Artificial Intelligence from 2018–2021) rather than as a commercial pivot.

The established defense drone players

Context for the four newcomers: the established defense industry isn’t standing still. The legacy and mid-cap players who matter:

• AeroVironment. Publicly traded. Makes the Switchblade 300/600 loitering munitions and the Puma reconnaissance drone. Switchblade has been the workhorse loitering-munition system supplied to Ukraine across the conflict. Has steadily added AI capabilities to its existing platforms.
• RTX (Raytheon), Lockheed Martin, Northrop Grumman, Boeing. The traditional defense primes. All have drone and counter-drone programs, but none of them lead on AI-first design. Most are now either partnering with or competing against the newcomers above.
• L3Harris. Counter-UAS and electronic warfare. Strong in the kinetic and electronic-warfare segments of counter-drone defense.
• General Atomics. Makes the MQ-9 Reaper and the newer Mojave. Traditional larger-airframe drone manufacturer; the AI overlay on top is improving but the airframes are mature.
• Kratos Defense. Mid-cap. Makes the XQ-58 Valkyrie autonomous combat aircraft — the Air Force’s Collaborative Combat Aircraft (CCA) program target.
• Saronic. Sister category — autonomous boats, not drones, but the technology overlaps and the company is in the same defense-tech investor universe.

The honest read: the legacy primes still hold most of the absolute defense-spending dollars by volume. But the AI-first newcomers are taking the categories where AI is the differentiator — autonomous flight, computer-vision targeting, swarm coordination, command-and-control software.

What AI actually does inside a military drone

Strip away the marketing language. Here’s what “AI in military drones” means technically.

• Autonomous navigation under degraded conditions. Visual-inertial odometry (VIO) lets a drone navigate using cameras and accelerometers when GPS is unavailable, jammed, or spoofed. This is the core technical capability that separates drones that work in contested electronic environments from drones that don’t.
• Computer-vision target identification. A model trained to distinguish vehicles, structures, personnel, and decoys from the drone’s camera feed. The model runs on the drone, not on a ground station, because latency matters and communications can’t be assumed.
• Swarm coordination. Multiple drones cooperating, deconflicting their flight paths, dividing a search area, and synchronizing actions without continuous human control. The math here is genuinely hard; the breakthroughs of 2024–2025 made small-scale swarming practical.
• Sensor fusion. Combining inputs from cameras, radar, lidar, RF detectors, and acoustic sensors into a single situational picture. Lattice OS (Anduril) is the most-cited example of this; Helsing and Shield AI have analogous internal systems.
• Loitering and persistent surveillance. A drone that stays on station for hours or days, monitors a defined area, and flags anomalies for human review. The AI is in the “flag anomalies” part — humans can’t watch a 48-hour video feed in real time.
• Human-in-the-loop arbitration. The decision of what to do about identified targets. This is where the ethics architecture lives — covered separately below.

None of these capabilities involve a chatbot-style large language model. Defense drones run computer-vision models, reinforcement-learning policies, and classical control algorithms on small chips. The AI hype cycle gave the entire field a vocabulary makeover, but the underlying techniques are mostly the same ones used in commercial autonomous-flight systems — tuned for harder operational environments.

The five ways to stop a hostile drone

Counter-UAS (the industry term for “counter-unmanned-aircraft-systems”) is its own multi-billion-dollar category. Five distinct technical approaches.

• Kinetic interceptors. Another drone, missile, or projectile that physically destroys the threat. Anduril’s Roadrunner-M is the most-discussed example — a vertical-takeoff jet-powered interceptor that re-lands on its launcher if it doesn’t engage. Costs around $200,000 per unit, which is high for a drone but a tiny fraction of the systems it’s designed to protect.
• Electronic warfare / jamming. Disrupting the radio frequencies the target drone uses for navigation or command. Anduril’s Pulsar, RTX’s SkySafe, and many others. Effective against drones that depend on GPS or RF command links; less effective against truly autonomous AI-piloted drones.
• Directed energy / lasers. Beam weapons that overheat or destroy electronic components. Currently expensive and weather-dependent, but improving. Multiple defense primes have working systems in field trials.
• Cyber takeover / spoofing. Sending the target drone fake GPS or command signals that make it land, return to home, or fly into a controlled area. Effective on consumer-grade drones; rapidly becoming less effective as defense drones harden against this.
• AI-driven detection and tracking. Doesn’t stop the drone directly — it identifies the drone and hands the engagement to one of the four above. Computer-vision plus radar plus RF detection fused into a single track. This is the “sensor” layer that the other four depend on.

The interesting industry observation: every major defense-AI company sells both attack drones and counter-drone systems. The technology overlaps. The companies are essentially competing with themselves at the next level of abstraction.

The Replicator initiative and the Defense Autonomous Warfare Group

The Pentagon’s formal answer to “the US needs more autonomous drones, faster” has gone through two phases.

Phase 1: Replicator (2023–2025). Launched under the Biden administration. Goal: field thousands of attritable autonomous systems within 18–24 months. The August 2025 target date passed with the program having fielded hundreds, not thousands. Critics characterized this as a buying-process failure rather than a technology failure.

Phase 2: Defense Autonomous Warfare Group (DAWG, 2025–). The successor organization. Different reporting structure, different budget authority, broader mandate. FY2026 budget: $225.9 million. FY2027 request: up to $54.6 billion. The Trump administration’s combined drone and counter-drone allocation tops $70 billion for the next fiscal year — the largest unmanned-systems investment in Pentagon history.

The structural shift behind the numbers: defense procurement is being rewritten to fit the AI-startup procurement model. Multi-year enterprise contracts (Anduril’s $20B Army deal is the cleanest example) replace per-program acquisition. Software-update cycles replace decade-long airframe development. The companies that ride that shift well are the new defense unicorns.

The human-in/on/out-of-the-loop architecture

The technical conversation around AI ethics in military drones turns on three specific terms. Worth knowing what they actually mean, separately from the broader policy debate.

• Human in the loop. The drone identifies a candidate target. A human reviews the identification and explicitly authorizes the action (or declines). The drone does not act without that authorization. Most US military drone systems operate in this mode today for lethal action.
• Human on the loop. The drone identifies and engages targets autonomously, but a human supervises and can intervene to halt or override. Used in some defensive scenarios (counter-UAS, missile defense) where reaction time precludes pre-authorization.
• Human out of the loop. The drone identifies and engages targets without human supervision. The Pentagon’s formal policy excludes this mode for lethal action. Some non-lethal systems (loitering above an area to flag activity) operate this way.

The technical detail behind those definitions: every system has multiple decision points (identify, classify, engage, confirm kill). Different decision points can be at different levels of human supervision within the same system. A drone might be “human-in-the-loop” for the engage decision and “human-out-of-the-loop” for the identify decision.

The ethics debate — which is real and ongoing — is about where the lines should be drawn, who decides, and how to verify compliance. This post stays out of the advocacy side of that debate. Anyone reading further on the topic should read the published US Department of Defense Directive 3000.09 (autonomy in weapon systems) as the primary policy document.

The 12-24 month outlook

• Anduril and Shield AI continue to consolidate position. The Anduril/Microsoft IVAS handoff and Shield AI’s Air Force CCA win point at an industry where the AI-first companies take an increasing share of new programs.
• European defense AI scales fast. Helsing’s $18B valuation is the leading indicator; the EU’s collective defense spending trajectory makes this a high-growth segment regardless of US dynamics. Expect more European entrants in 2026–2027.
• Swarm capabilities become operational at scale. Most demonstrations to date have been small (10–100 drones). The technical path to operational swarms in the thousands is now visible; multiple programs are funding it directly.
• Counter-UAS gets the next big procurement wave. Critical infrastructure protection, stadium security, government facility defense, and active-conflict counter-drone all expand. Roadrunner-style kinetic interceptors and Pulsar-style EW systems get the bulk of the spending.
• The civilian-defense overlap grows uncomfortable. Computer-vision target ID, GPS-denied flight, and swarm coordination are dual-use technologies. The same models that make commercial inspection drones useful also make defense drones more capable. Expect more regulatory attention to export controls on AI models specifically.
• The Ukraine learning curve continues to drive the industry. Lessons from active conflict reshape the technology faster than any peacetime R&D program could. The companies that have direct deployments (Shield AI, Swift Beat, Anduril’s ALTIUS) keep a learning-rate advantage.

FAQ

What is the biggest AI defense drone company?

By valuation as of May 2026: Anduril Industries at $61 billion. By drones shipped to active conflict: Eric Schmidt’s Swift Beat (formerly White Stork) and various Ukrainian-domestic builds, supplying hundreds of thousands of units. By legacy procurement footprint: AeroVironment and the traditional defense primes.

Who founded Anduril Industries?

Palmer Luckey, in 2017. Luckey is the same person who founded Oculus VR and sold it to Facebook in 2014. He was forced out of Facebook in 2017; Anduril was his next venture. The company is now the highest-valued private defense-tech company in the United States.

What is Hivemind?

Shield AI’s autonomous-pilot AI software. Hivemind flies aircraft using onboard cameras and inertial sensors rather than GPS or external command links. The company’s V-BAT drone runs Hivemind and has logged 130+ sorties in Ukraine’s contested electronic-warfare environment since June 2024.

Is Eric Schmidt’s drone company called White Stork or Swift Beat?

Both names trace to the same operation. The original name was White Stork. The company briefly attempted to rebrand as Project Eagle (to avoid confusion with a humanitarian charity), then settled on Swift Beat as the current operating name. Corporate parent is Tallinn-based Volya Robotics OÜ, of which Schmidt is the sole beneficiary.

What is the Replicator initiative?

The original Pentagon program (2023–2025) intended to accelerate procurement and deployment of thousands of attritable autonomous drones. After missing its August 2025 target, the program transitioned to the Defense Autonomous Warfare Group (DAWG). DAWG’s FY27 budget request is up to $54.6 billion, part of a $70+ billion total Pentagon allocation for unmanned and counter-drone systems.

Is Helsing publicly traded?

No. Helsing is privately held, currently raising a $1.2 billion round at approximately $18 billion valuation. The lead investors include Dragoneer and Lightspeed Venture Partners. The company has not announced IPO plans.

How does an AI drone find a target without GPS?

Visual-inertial odometry combines onboard camera feeds with motion sensors (accelerometers and gyroscopes) to estimate position over time. Computer-vision models on the drone identify ground features and match them against pre-loaded maps. The combined system can navigate accurately even when GPS is unavailable, jammed, or actively spoofed by an adversary.

What’s the difference between autonomous and remotely-piloted drones?

A remotely-piloted drone is flown by a human operator via radio link in real time. An autonomous drone executes a mission — identifying paths, avoiding obstacles, making routing decisions — without continuous human input. Most modern defense drones are hybrid: human supervisor sets goals, autonomous systems handle execution, human approves any kinetic action.

The bottom line

Military drones are the most capital-rich, fastest-growing, and most operationally consequential AI category in 2026. Anduril’s $61 billion valuation, Shield AI’s 140% year-over-year growth, Helsing’s $18 billion European-champion position, and Swift Beat’s hundred-thousand-unit Ukraine deployments aren’t outliers — they’re the leading edge of an industry restructuring.

The technology that makes it all work — GPS-denied flight, computer-vision target ID, swarm coordination — overlaps closely with civilian autonomous-flight systems. Lessons flow both ways. The same AI capability that makes a commercial inspection drone useful also makes a defense drone more capable, and vice versa. That overlap is the most important thing to understand about this field.

For broader context: AI in Drones: The Complete 2026 Guide (civilian-focused), The Complete Guide to AI in 2026, Every AI Model Worth Knowing in 2026. Daily AI fundamentals in our free Beginners in AI newsletter.

Sources

• TechCrunch, Anduril raises $5B, doubles valuation to $61B (May 13, 2026) — Anduril valuation and funding data.
• U.S. Department of Defense contract announcements at defense.gov/News/Contracts — primary source for the March 2026 Anduril $20B 10-year Army enterprise contract.
• Fortune, Shield AI projecting more than $540M in revenue this year as valuation more than doubles to $12.7B (March 26, 2026).
• TechCrunch, Defense startup Shield AI lands $12.7B valuation, up 140%, after U.S. Air Force deal.
• DroneXL, Helsing Aims For $18 Billion Valuation In $1.2 Billion Dragoneer-Led Round (May 9, 2026).
• Helsing official site, helsing.ai — manufacturer-direct reference for Centaur AI agent and HX-2 strike drone product details. Cross-reference with Helsing’s press releases for Series E ($1.2B at $18B valuation, May 2026) and the Saab Gripen Project Beyond combat trials (May–June 2025).
• Defense Express, Mysteries of Swift Beat: Ukraine’s New Drone Supplier — Swift Beat / White Stork corporate structure.
• Ukrainska Pravda, Billionaire and ex-Google CEO Eric Schmidt is making drones for Ukraine (July 2025) — Ukrainian-press primary reporting on Schmidt’s Swift Beat operation.
• DefenseScoop, DOD moves to make its largest-ever investment in drones and anti-drone weapons (April 21, 2026).
• Breaking Defense, Pentagon officials broadly detail $55 billion drone plan under DAWG.
• Congress.gov, DOD Replicator Initiative: Background and Issues for Congress — Congressional Research Service overview.
• U.S. Department of Defense Directive 3000.09, “Autonomy in Weapon Systems” (revised January 25, 2023) — available via the DoD Issuances portal at esd.whs.mil/DD. The primary policy document on human-in/on/out-of-the-loop architecture.

You May Also Like

• Anduril Industries Explained: Palmer Luckey’s $61B Defense AI Empire
• AI in Drones: The Complete 2026 Guide (civilian companion)
• The Complete Guide to AI in 2026
• Every AI Model Worth Knowing in 2026
• Anthropic company guide (for the civilian-AI parallel)
• May 2026 AI updates cheat sheet