Shield AI
Last reviewed
Sources
35 citations
Review status
Source-backed
Revision
v3 ยท 2,862 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
35 citations
Review status
Source-backed
Revision
v3 ยท 2,862 words
Add missing citations, update stale details, or suggest a clearer explanation.
Shield AI is an American defense technology company that builds autonomy software for military aircraft and drones. Founded in 2015 and headquartered in San Diego, California, it is best known for Hivemind, an AI "pilot" that flies aircraft without GPS or a human operator, and for the V-BAT, a vertical-takeoff drone that runs on it. In March 2026 Shield AI raised about $2 billion at a roughly $12.7 billion valuation, making it the second most valuable U.S. defense technology startup after Anduril. [1][2]
Shield AI's pitch is narrower than it first appears. The company does not primarily sell drones. It sells the software brain that flies them, and increasingly puts that brain on hardware built by other people, including fighter jets and cheap one-way attack drones. By 2025 the company employed around 1,000 people and counted backers including Andreessen Horowitz, L3Harris, Hanwha, Advent International, JPMorganChase, and Blackstone. [3][4][30]
Shield AI was started in 2015 by brothers Brandon Tseng and Ryan Tseng, along with Andrew Reiter. Brandon Tseng is a former U.S. Navy SEAL who served combat tours in Afghanistan; the idea for the company came out of his frustration that troops still had to clear buildings without knowing what was inside. By his account the founders started with about $100,000 raised from friends and family. Ryan Tseng, an electrical engineer, had previously founded a wireless-charging startup. Reiter led the early autonomy engineering. [5][6]
The first product was Nova, a small autonomous quadcopter that could fly itself through buildings and tunnels using onboard sensors instead of GPS or a remote pilot. Shield AI has described Nova as the first autonomous robot used in combat, with deployments supporting U.S. forces in the Middle East beginning around 2018. A second-generation Nova 2 followed in 2021. Running underneath Nova was the software that would become the company's real franchise: Hivemind. [5][7]
Two acquisitions in July 2021 reshaped the company. Shield AI bought Heron Systems, a small firm whose AI agent had beaten an experienced human F-16 pilot 5 to 0 in DARPA's 2020 AlphaDogfight Trials, a result that gave Shield AI deep reinforcement learning talent and credibility in air combat. Days later it acquired Martin UAV, a Texas company whose V-BAT drone became Shield AI's flagship hardware platform once Hivemind was integrated into it. [8][9]
Hivemind is Shield AI's aircraft-agnostic autonomy stack, software the company compares to the self-driving systems in cars but built for aircraft operating in contested airspace. It combines several techniques: path planning, mapping, state estimation, and computer vision for navigation and perception, plus reinforcement learning and large-scale simulation to discover and train tactical behaviors. The Heron Systems acquisition fed directly into this, since reinforcement-learned dogfighting was that team's specialty. [3][10]
The central claim is that Hivemind keeps flying when the things drones usually depend on stop working. In an environment where GPS is jammed and the radio link back to a human operator is cut, a Hivemind-equipped aircraft is supposed to navigate, make decisions, and complete its mission on its own, processing everything on the aircraft rather than in a ground station or the cloud. Shield AI calls this "self-driving" for aircraft and describes the behavior as "read and react," meaning the autonomy senses the situation and adapts rather than following a fixed script. [3][10]
Hivemind also handles teaming. The company says a single operator can command multiple aircraft at once, with the software coordinating the group, synchronizing timing, and dividing up sensing or strike tasks among the members. To let other manufacturers build on the stack, Shield AI ships a Hivemind software development kit and an underlying middleware layer it calls EdgeOS, which uses static JSON configuration files and local shared-memory messaging to keep behavior predictable and fast on the aircraft itself. [11][12]
A selling point Shield AI repeats is breadth: Hivemind is not tied to one airframe. By 2024 the company said the same core autonomy had flown on at least eight different types of aircraft, including three quadcopters, the V-BAT, the F-16, several Kratos jet drones, and General Atomics' MQ-20 Avenger. The eighth integration, announced in August 2024 under a contract from the Navy's NAVAIR PMA-281 office, put Hivemind on the Kratos BQM-177A target drone. [13][14]
| Product | What it is | Notes |
|---|---|---|
| Hivemind | The core AI autonomy software ("AI pilot") | Aircraft-agnostic; runs on the edge in GPS- and comms-denied conditions |
| Hivemind Enterprise | A developer platform and SDK for outside manufacturers | Pitched to OEMs and governments to build their own autonomy products |
| Nova / Nova 2 | Small autonomous indoor reconnaissance quadcopter | First product; used by U.S. forces in the Middle East |
| V-BAT (MQ-35A) | Vertical-takeoff-and-landing ISR and strike drone | Group 3 class; acquired via Martin UAV, now Hivemind-powered |
| X-BAT | AI-piloted VTOL fighter drone | Unveiled October 2025; Group 5 class for Collaborative Combat Aircraft |
The V-BAT is the hardware the company is best known for. It is a single-engine VTOL aircraft that takes off and lands vertically in a footprint as small as about 15 by 15 feet, needing no runway or launch catapult, then transitions to efficient horizontal flight. Shield AI cites endurance of more than 12 hours with an electro-optical/infrared payload, a maximum takeoff weight of around 161 pounds (73 kg), payload capacity near 40 pounds (18 kg), and a heavy-fuel (JP-5) engine. The U.S. military designates it the MQ-35A. A block upgrade announced in 2025 added satellite communications and a heavy-fuel engine alongside the autonomy improvements. [15][16]
X-BAT, unveiled in October 2025, is a much more ambitious machine: an AI-piloted fighter drone that takes off and lands vertically, so it can launch from a small clearing, a ship, or a remote island without a runway. Shield AI describes a Group 5 aircraft with an F-16-class afterburning engine and a thrust-vectoring nozzle, a low-signature airframe, and a range it has cited at around 2,000 nautical miles. The company is positioning X-BAT for the Pentagon's Collaborative Combat Aircraft (CCA) effort, where uncrewed jets fly alongside crewed fighters as "loyal wingmen." Shield AI has said first vertical-takeoff flights are targeted for the fall of 2026, with full mission-capable flights around 2028, and that it is working with Pratt & Whitney and General Electric on propulsion using refurbished legacy engines. [4][17]
The demonstration that put Shield AI in front of the public involved a fighter jet rather than a drone. Through DARPA's Air Combat Evolution (ACE) program, a Shield AI autonomy agent was integrated onto the X-62A VISTA, a modified F-16 used as a flying testbed. In 2023 the autonomous jet flew real within-visual-range dogfights against a crewed F-16 over Edwards Air Force Base in California. On May 2, 2024, the Secretary of the Air Force, Frank Kendall, flew in the front seat of the AI-piloted X-62A during a live maneuvering flight against a crewed F-16, with a human safety pilot seated behind him who did not touch the controls. It was an unusually public vote of confidence from a sitting service secretary. Shield AI executive Brett Darcey said the agent was "a direct descendant of one used by Heron Systems" to win the 2020 AlphaDogfight Trials. [18][19][32]
The most consequential proving ground has been Ukraine. Shield AI delivered its first V-BATs there in mid-2024 and opened a local office in Kyiv in January 2025, with engineers working alongside Ukraine's Unmanned Systems Forces. By April 2025 the company said the drones had flown more than 130 sorties under heavy Russian electronic warfare, and it claimed V-BAT was the only long-endurance intelligence, surveillance, and reconnaissance (ISR) and strike platform able to penetrate both sides' GPS and communications jamming. In one mission declassified by Ukrainian forces, a V-BAT flew roughly 60 miles (about 100 km) into Russian-controlled territory while GPS and communications were jammed, located a Buk-M1 air-defense system, and cued a HIMARS strike that destroyed it. The Ukraine results helped drive follow-on interest, including reported adoption by the Netherlands. [20][21]
Earlier U.S. government work built the foundation. Shield AI won an early Defense Innovation Unit autonomy contract in 2016, took Air Force AFWERX/STRATFI awards, and has integrated Hivemind across multiple Navy and Air Force test programs. The breadth of these integrations, rather than any single large procurement, is what the company points to as evidence its software travels across airframes. [13][22]
Shield AI's biggest institutional milestone came through the U.S. Air Force's Collaborative Combat Aircraft program. In February 2026 Hivemind was selected as a mission autonomy provider for CCA, an arrangement in which Shield AI's software is meant to work across the program rather than being tied to a single airframe; reporting noted it would operate alongside competitor Anduril's Lattice system, with Anduril's "Fury" jet among the aircraft, signaling the Air Force's preference for multiple vendors over single-source dependency. [31]
On June 17, 2026, the Air Force followed with a production contract for Shield AI to implement Hivemind mission autonomy for the CCA program. The award reflects what the service calls a "software-first" approach: it separates mission autonomy from the aircraft through an Autonomy Government Reference Architecture (A-GRA), so the software can be upgraded independently of any airframe redesign. "Mission autonomy is a foundational capability for future airpower," said Christian Gutierrez, senior vice president of Hivemind at Shield AI. "The Air Force's approach enables faster innovation, rapid capability deployment, and greater operational advantage for the warfighter." The dollar value of the production award was not disclosed. [33][34]
In May 2026 Shield AI was selected to put Hivemind on the Low-Cost Uncrewed Combat Attack System (LUCAS), a Pentagon program for cheap, one-way attack drones meant to be fielded in large numbers. The selection was made by the Office of the Under Secretary of War for Research and Engineering, under its prototyping and experimentation office. (The Department of Defense was renamed the Department of War in early 2026, which is why these offices now carry "War" in their titles.) Shield AI announced the award on May 19, 2026. [23][24]
The premise of LUCAS is what military planners call "affordable mass": drones cheap enough to lose, fielded in swarms that can saturate an enemy's defenses. The program drew urgency from the spring 2026 conflict with Iran, where U.S. Central Command said it used LUCAS drones in combat for the first time. Hivemind's job is to make the swarm coordinate and adapt in flight from a single operator's intent, rather than requiring one pilot per drone. As Brandon Tseng put it, "LUCAS is about delivering affordable mass, but mass without coordination is limited in value. Hivemind is the AI pilot that makes that mass intelligent." Shield AI said an operational demonstration of an autonomous swarm of ten or more drones is planned for the second half of 2026. Press reporting identified Arizona-based SpektreWorks as the maker of the roughly $35,000 airframe, which is modeled on a captured Iranian design. [23][25][26]
Shield AI grew slowly at first, then raised money at an accelerating pace as defense technology became a hot investment category. The March 2026 round in particular more than doubled the company's valuation in a single year, a 140 percent jump from its $5.3 billion mark twelve months earlier. [1][35]
| Round | Date | Amount | Valuation | Lead / notable investors |
|---|---|---|---|---|
| Series E | Nov 2021 | $210M | ~$1B+ | Andreessen Horowitz, others |
| Series E extension | Jun 2022 | $165M | ~$2.3B | |
| Series F | Oct 2023 | ~$200M | ~$2.7B | U.S. Innovative Technology, Riot Ventures |
| Series F-1 | Mar 2025 | $240M | $5.3B | L3Harris, Hanwha Asset Management |
| Series G + preferred | Mar 2026 | ~$2B | $12.7B | Advent International, JPMorganChase; Blackstone (preferred) |
The March 2025 round, $240 million at a $5.3 billion valuation, was led by strategic investors L3Harris and Hanwha and was tied to commercializing Hivemind Enterprise, the developer platform Shield AI wants other manufacturers to build on. Backers in this and earlier rounds included Andreessen Horowitz, Breyer Capital, and U.S. Innovative Technology. The round made Shield AI the second-largest U.S. defense technology startup by valuation, behind Anduril. [27][28]
The March 2026 raise was the headline event. Advent International and the Strategic Investment Group within JPMorganChase's Security and Resiliency Initiative co-led a $1.5 billion Series G, while funds managed by Blackstone added $500 million in preferred equity (plus a $250 million delayed-draw facility), for roughly $2 billion total at a $12.7 billion post-money valuation. Advent's chairman David Mussafer joined the board and JPMorgan's Todd Combs joined as a board observer. Part of the proceeds was earmarked for acquiring Aechelon Technology, a visual-simulation software company, announced alongside the round. [1][29]
Brandon Tseng has served as president and remained a public face of the company; for years he and his brother ran it together. In March 2025 Shield AI brought in Gary Steele, formerly chief executive of the cybersecurity firm Proofpoint, as CEO, with Ryan Tseng moving to chief strategy officer. The hire signaled a shift toward scaling the business operationally as contracts and headcount grew; the company reported around 1,000 employees by 2025. [30][6]
Not every development has been favorable. In April 2025 Shield AI was added to China's "unreliable entities" list amid escalating trade and technology tensions, a designation that restricts dealings with the company in China; Shield AI responded that it does not rely on Chinese suppliers. The company has also faced safety scrutiny: a 2024 V-BAT ground incident injured a Navy service member, prompting temporary operating restrictions that were later eased after design changes. [31][30]