What Is an AI Defense Company?
A defense AI company is a firm that builds artificial intelligence and autonomy for military and national-security use — and sells it as a finished, software-defined product rather than a multi-decade custom program. Their work falls into three layers: the software layer that fuses sensor and intelligence data into decisions (Palantir's AIP, Anduril's Lattice, Scale's Donovan); the autonomy layer that lets uncrewed systems operate without GPS, communications, or a human operator (Shield AI's Hivemind, Helsing's Centaur); and the hardware layer of autonomous drones, aircraft, and ships (Anduril's Fury and Ghost, Helsing's HX-2, Saronic's Corsair and Mirage).
What sets this generation apart from traditional prime contractors is the model: they raise private capital, build products speculatively, iterate software continuously, and aim to manufacture at scale and low unit cost. This guide covers companies with fielded systems or live government contracts and a distinct technical position. Several also appear in our AI robotics companies and LLM companies guides, since defense autonomy draws directly on robotics and foundation models.
AI Defense Companies — Detailed Reviews
Ordered roughly by centrality to the defense-AI landscape: the two anchor companies first,
then the European leader, the autonomy and data specialists, and the naval-autonomy challenger.
1. Anduril Industries
Costa Mesa, USA · Founded 2017 · Autonomy + hardware
Lattice OS
Private
$2.2B
2025 revenue (2x YoY)
Arsenal-1
Ohio megafactory
Anduril is the company that proved venture-backed defense could work. Founded in 2017 by
Oculus creator Palmer Luckey with Trae Stephens, Brian Schimpf, Matt Grimm, and Joe Chen, it
rejected the cost-plus contracting model and instead builds finished products with private
capital. Its core is Lattice, an AI command-and-control
operating system that fuses thousands of sensor feeds into one real-time picture and recommends
responses — then tasks Anduril's own hardware: the Ghost and Anvil drones, the Fury / YFQ-44A
Collaborative Combat Aircraft, the Roadrunner reusable interceptor, Barracuda air vehicles,
Dive-LD and Ghost Shark underwater vehicles, and Sentry surveillance towers.
Revenue doubled in 2025 to roughly $2.2 billion, and Anduril is scaling output through
Arsenal-1, a hyperscale autonomous-systems factory near Columbus, Ohio. In 2024 it took over the
U.S. Army's IVAS soldier headset program from Microsoft. A May 2026 Series H led by Thrive
Capital and Andreessen Horowitz set a $61 billion valuation — roughly double the level just
eleven months earlier. Anduril is the default anchor of any defense-AI shortlist for buyers who
want autonomous hardware and command software from one vendor.
2. Palantir Technologies
Denver, USA · Founded 2003 · Data & AI software · NASDAQ: PLTR
AIP
Public
~$7.2B
2026 revenue guidance
+104%
U.S. revenue growth YoY
£240M
UK MoD contract (Dec 2025)
Gotham
Defense & intelligence platform
Palantir is the software backbone of Western defense and intelligence. Founded in 2003, it built
Gotham for defense and intelligence operations and
Foundry for enterprise data integration, then added the
Artificial Intelligence Platform (AIP), which connects
large language models to an organisation's own operational data so analysts and commanders can
query, plan, and act on it safely. The platforms turn fragmented sensor, logistics, and
intelligence data into a single decision-ready picture — in one example, AIP cut a submarine
scheduling task from 160 hours to ten minutes.
Palantir is the only publicly traded company on this list, with 2026 revenue guidance above $7
billion — more than double 2025 — driven by triple-digit U.S. commercial growth and expanding
government work including the U.S. Army's Vantage program and a £240 million UK Ministry of
Defence contract. Where Anduril leads autonomous hardware, Palantir leads the data-and-decision
layer; the two increasingly partner (both are in the Pentagon's Thunderforge program). Choose
Palantir when the problem is integrating and reasoning over data at scale rather than fielding
physical systems.
3. Helsing
Munich, Germany · Founded 2021 · Defense AI software + drones
Centaur
Europe
~€12B
Valuation (Jun 2025)
€269M
Initial HX-2 contract (2026)
100 km
HX-2 strike-drone range
CA-1
Europa UCAV (first flight 2027)
Helsing is Europe's flagship defense-AI company, founded in Munich in 2021 by Torsten Reil,
Gundbert Scherf, and Niklas Köhler. It began as a software company — building AI that turns raw
sensor data into a real-time battlefield picture for electronic warfare and ISR — and has
expanded into hardware. Its HX-2 is a roughly
12-kilogram strike drone with a ~100 km range that uses onboard AI to operate in GPS-denied
environments, and its Centaur AI agent piloted a Gripen
E fighter in a beyond-visual-range trial — among the first publicly known cases of AI flying an
operational fighter jet.
Helsing reached an estimated €12 billion valuation in a June 2025 round led by Spotify founder
Daniel Ek, and was reported in May 2026 to be raising about $1.2 billion at roughly an $18
billion valuation — which would make it one of Europe's most valuable startups. In February 2026
Germany's Bundestag approved an initial €269 million HX-2 contract within a framework worth up to
€1.46 billion. Its unmanned combat aircraft, the CA-1 Europa, targets a 2027 first flight. Choose
Helsing when European sovereignty, NATO interoperability, and support for Ukraine and allied
forces are priorities.
4. Shield AI
San Diego, USA · Founded 2015 · AI pilot / autonomy
Hivemind
Private
$12.7B
Valuation (Mar 2026)
$540M+
Projected 2026 revenue
F-16
Flown by Hivemind (VENOM)
Shield AI builds the AI pilot. Founded in 2015 by
former Navy SEAL Brandon Tseng with his brother Ryan Tseng and Andrew Reiter, its mission is to
fly aircraft and operate them in teams without GPS, without communications, and without a human
in the cockpit. Its Hivemind autonomy stack reads
sensors and flies the aircraft itself — it has flown the F-16 in the Air Force's VENOM program,
quadcopters in combat, and Shield AI's own V-BAT, a
vertical-takeoff drone that needs no runway and is used for ISR by U.S. and allied forces. The
company sells Hivemind Enterprise so other manufacturers can add autonomy to their own platforms.
In March 2026 Shield AI raised a $1.5 billion Series G at a $12.7 billion valuation — led by
Advent International and JPMorganChase's Security and Resiliency Initiative, with $500 million of
preferred equity from Blackstone — more than doubling its valuation in a year, and it acquired
the simulation company Aechelon to accelerate its Hivemind Foundation Model for Defense. It
unveiled X-BAT, a jet-powered VTOL collaborative combat aircraft, in 2025, and projected over
$540 million in 2026 revenue. Choose Shield AI when autonomous flight in contested,
GPS-denied conditions is the core requirement.
5. Scale AI
San Francisco, USA · Founded 2016 · Defense data & models
Defense Llama
Meta-backed
$500M
Pentagon contract (2026)
49%
Meta equity stake (2025)
Donovan
Government AI platform
Scale AI built its business on the data layer of AI — labeling and curating the training data
behind frontier models — and has turned that expertise toward national security. Through
Scale Donovan, its government platform, it runs
Defense Llama (a national-security language model built
on Meta's Llama framework, available only in controlled U.S. government environments) and
participates in the Defense Innovation Unit's Thunderforge
program — the DoD's first effort to embed AI agents across military planning workflows — alongside
Microsoft and Anduril.
Founded in 2016 by Alexandr Wang and Lucy Guo, Scale became Meta-backed in 2025 when Meta took a
roughly 49% equity stake, deepening its compute and model access. In 2026 the Pentagon awarded
Scale a $500 million contract to process data and support military decision-making — five times an
earlier $100 million deal — marking its shift from data vendor to strategic player inside the
national-security AI stack. Choose Scale AI when the need is data pipelines, model
customisation, and government-grade LLM deployment rather than hardware.
6. Saronic Technologies
Austin, USA · Founded 2022 · Autonomous surface vessels
Maritime autonomy
Private
$9.25B
Valuation (Mar 2026)
$1.75B
Series D (Mar 2026)
20+/yr
Ships targeted by 2027
Port Alpha
Next-gen shipyard
Saronic is doing for ships what Anduril did for drones: building
autonomous surface vessels designed to be
mass-produced cheaply, so the U.S. Navy can field large numbers of uncrewed boats and close the
shipbuilding-capacity gap with China. Founded in 2022 by former Navy SEAL Dino Mavrookas with
Vibhav Altekar, Doug Lambert, and Robert Lehman, its vessel line runs from the small Spyglass and
Cutlass to the mid-size Corsair and the larger
Mirage and Marauder,
each running Saronic's own autonomy software for navigation and coordinated multi-vessel
operations in contested, GPS-denied waters.
In March 2026 Saronic closed a $1.75 billion Series D led by Kleiner Perkins at a $9.25 billion
valuation — more than double its early-2025 level — to fund Port Alpha, a next-generation
shipyard, and scale production toward more than 20 ships a year by 2027 across facilities in
Louisiana and Texas. It is the clearest example of software-defined, mass-manufacturable naval
autonomy, and the natural maritime complement to the air- and land-focused companies on this
list.
The Defense-AI Stack: Software, Autonomy, and Hardware
The most useful way to compare these companies is by which layer of the stack they own. At the
software layer, Palantir and Scale AI turn fragmented
data — sensors, intelligence, logistics — into decisions and government-grade models. At the
autonomy layer, Shield AI's Hivemind and Helsing's Centaur
provide the "brain" that flies or operates a platform without GPS, communications, or a human at
the controls. At the hardware layer, Anduril, Helsing, and
Saronic build the physical drones, aircraft, and ships that act in the world.
The strategic frontier is convergence. Anduril spans hardware and command software; Helsing spans
software and drones; Shield AI sells its autonomy as a platform others can build on. For a buyer or
partner, the question is whether you want a single full-stack supplier (Anduril) or best-of-breed
layers integrated together — for example Palantir's data backbone feeding Shield AI's autonomy onto
Anduril or Saronic hardware. Increasingly these companies interoperate inside shared programs such
as the Pentagon's Thunderforge, rather than competing end to end.
Reality Check: What Defense AI Still Has to Prove
Valuations have run ahead of fielded revenue. These are some of the best-funded private companies
in the world, but defense procurement is slow, and the gap between a successful
demonstration and a multi-year
program of record is wide. Autonomy that works in a
controlled test can behave unpredictably against jamming, spoofing, and adversaries who adapt — and
the same machine-learning systems that detect threats can be deceived by them.
The policy and ethics layer is unsettled: the rules for
lethal autonomy, meaningful human control, and accountability are still being written
internationally, and export controls (ITAR and allied regimes) constrain who can buy what. Scaling
manufacturing to the volumes these companies promise — thousands of drones or dozens of ships a
year — is itself unproven at the price points claimed. None of this means the shift is hype; it
means treating defense AI as a fast-maturing but still-emerging sector where fielded contracts,
not funding rounds, are the real signal.