Which defense startups are building software that looks more like distributed systems than traditional weapons platforms?

Yes. If by “looks like distributed systems” you mean software that coordinates many heterogeneous nodes, survives intermittent connectivity, runs at the edge, exposes APIs, and treats hardware as replaceable endpoints, there’s a surprisingly clear group emerging.

The strongest fits

The conceptual split

I'd actually divide the market into three software architectures:

1. “Kubernetes for the battlefield”
Defense Unicorns, Varada, parts of Applied Intuition.

The core problem is: How do you deploy, update, observe and coordinate software across thousands of unreliable, heterogeneous edge machines?

That's a very recognizable distributed-systems problem: orchestration, service discovery, data synchronization, intermittent connectivity, security boundaries, fleet management, rollback, telemetry, etc.

2. “Operating system for autonomous fleets”
Anduril and Shield AI are the strongest examples.

Here the primitives aren't containers and services; they're vehicles, sensors, effectors, autonomy behaviors and human operators. Anduril describes Lattice as integrating thousands of sensors/effectors and coordinating systems across distributed geographies and contested communications.

Shield AI is particularly interesting because the software is increasingly being treated as portable infrastructure rather than aircraft-specific code. Its CCA contract explicitly allows autonomy software to evolve independently of the aircraft.

3. “Distributed database + AI for the battlespace”
Helsing is a good example.

The interesting abstraction isn't necessarily “control this drone.” It's aggregate huge amounts of heterogeneous sensor data, perform inference locally, maintain a common world model, and distribute useful state to other nodes/operators.

---

The really interesting thesis

The common architectural shift is:

Platform-centric defense → network-centric defense → software-defined distributed system.

Traditional defense procurement tends to think:

radar → vehicle → missile → command center

The new architecture increasingly looks like:

sensors ↔ edge compute ↔ network ↔ shared state ↔ AI agents ↔ effectors

And the individual drone, boat, radar, satellite, etc. becomes almost analogous to a node in a distributed computing system.

That's why Anduril's Lattice is especially significant: the company itself describes it as an extensible network connecting sensors and effectors, with mesh networking and an SDK for integrating third-party systems.

If you're looking at this as an engineering/investment landscape, I'd put Anduril, Shield AI, Defense Unicorns, Applied Intuition, Helsing, and Varada at the top of the list—but for somewhat different layers of the stack.

The particularly interesting underappreciated category is “infrastructure for disconnected distributed computing at the tactical edge.” That's where the defense problem starts looking remarkably similar to distributed systems research: partitions are normal, bandwidth is scarce, nodes are intermittently available, trust boundaries are hostile, and you still need the system to converge on a useful state.

The defense tech ecosystem is undergoing a major shift away from standalone "exquisite" hardware (like multi-billion-dollar jets or single-purpose missiles) toward software-defined, distributed architectures . Instead of building just a weapon, these startups treat the entire battlefield as a distributed, degraded-network computing problem—focusing on mesh networking, edge data-fusing, decentralized state management, and real-time collaborative consensus.

Key defense startups building software that functions more like a distributed systems architecture than a traditional weapons platform include:

• Chaos Industries – Builds sensor, radar, and effector hardware powered by what they call Coherent Distributed Networks (CDN™) . Rather than treating a radar or a sensor as an isolated data collector, their software orchestrates multi-node, decentralized sensor webs that allow tactical and border units to share tracking state and achieve domain dominance dynamically.
• Onebrief – Tackles the operational military planning and command-and-control (C2) bottleneck like a real-time, cloud-native collaborative document editor (think Figma or Google Docs for joint-force command staffs). It uses a distributed database of modular, interconnected "cards" (representing units, logistical facts, or target sets) where a change to one data point instantly propagates and recomputes across sync matrices, automated operations orders, and collaborative maps.
• Volantyx Aerospace (via its defense division, Neura Defense Systems) – Focuses on distributed edge-AI architectures designed to run locally on decentralized compute nodes (like NVIDIA Jetson Orin arrays) without relying on a centralized cloud or constant backhaul connectivity. The system continuously fuses raw multi-modal data streams—acoustic, radar, RF, electro-optical, and infrared—across local nodes to form real-time aerial tracks, treating threat detection as a localized consensus problem.
• Helsing – Operates as an AI-and-software layer meant to "connect existing hardware into a new AI-enabled network". Rather than manufacturing traditional prime-contractor platforms, they build the underlying software architectures (such as algorithmic electronic warfare and reconnaissance-strike coordination tools like Cirra and Altra ) designed to coordinate autonomous software mass across disparate defense platforms.

Would you like to explore how these distributed software models handle DDIL (Disrupted, Intermittent, Limited) network constraints, or are you interested in how they integrate with legacy military hardware like TDL (Tactical Data Links)?

Instead of building "bigger iron" (traditional kinetic platforms like heavy tanks, monolithic radar stations, or single-purpose jets), a new wave of defense tech startups approaches modern warfare through the lens of distributed systems engineering, mesh networking, consensus algorithms, and edge state synchronization.

Operating under the reality of DDIL (Denied, Degraded, Intermittent, and Limited ) environments, these companies treat the entire battlefield like a giant, hostile, highly latent distributed cluster.

Several standout defense startups are explicitly building software and network architectures that mirror distributed systems:

• 1. Chaos Industries and their Coherent Distributed Networks (CDN)
  • The Approach: Rather than relying on a massive, expensive, centralized early-warning radar or sensor tower, Chaos builds omniscient sensing and detection portfolios powered by distributed network fabrics.
  • The Distributed Analogy: It functions like a microservice architecture for physical space. Nodes dynamically discover each other, pool their sensor data locally, and route tracking information across a mesh. If a single node is jammed or destroyed, the remaining nodes re-form consensus and maintain tracking without needing a central "master node" or cloud reach-back.
• 1. Onebrief
  • The Approach: Onebrief replaces static, manually updated military planning slides and fragmented legacy command-and-control (C2) binders with a real-time, collaborative workspace for distributed staff.
  • The Distributed Analogy: It behaves like a real-time collaborative database (think Google Docs or Kubernetes state synchronization combined with Git branching) for military operations. Changes made by a brigade in one location instantly propagate, resolve conflicts across echelons, and maintain a synchronized operational state even when network connectivity drops and resyncs asynchronously.
• 1. Anduril Industries (specifically their Lattice OS platform)
  • The Approach: While Anduril builds hardware (interceptors, submersibles, and counter-UAS), their true product is Lattice—an open, AI-powered software operating system that turns any collection of dumb hardware into a smart, decentralized swarm.
  • The Distributed Analogy: Lattice acts as an orchestration layer (similar to a tactical Kubernetes for hardware). It abstracts away underlying device specifics, enabling edge-to-edge coordination where autonomous assets dynamically assign tasks, share sensor feeds locally, and balance workloads peer-to-peer without routing everything back to a human-attended headquarters.
• 1. Shield AI (with their Hivemind autonomy stack)
  • The Approach: Hivemind is an AI pilot designed to run local autonomy on aircraft without GPS or remote human piloting.
  • The Distributed Analogy: Multi-agent swarming requires edge consensus. Hivemind allows a cluster of drones to coordinate maneuvers, allocate targets, and execute missions as a distributed computing cluster in the sky, where individual units act as compute nodes sharing local state over a localized ad-hoc mesh.

Would you like to dive deeper into how these companies handle edge consensus and state-sync over lossy tactical radios , or are you looking at a specific layer of the stack (like networking vs. orchestration)?

If by “looks more like distributed systems” you mean software-first defense companies building networks of sensors, agents, data pipelines, autonomous nodes, and decision systems rather than a single exquisite platform (tank, jet, missile), the interesting companies are mostly in the defense autonomy, C2, and AI infrastructure layer.

A few stand out:

Anduril Industries — battlefield operating system / sensor-effectors network

Anduril is probably the closest analogy to a distributed systems company. Its core product is Anduril Lattice, a software layer that connects sensors, autonomous systems, and human operators into a common operational picture. The architecture resembles a distributed control plane:

• many heterogeneous nodes (cameras, drones, radars, towers, vehicles)
• real-time data fusion
• automated detection and tasking
• APIs and integration with third-party systems
• autonomous agents acting at the edge

The hardware (drones, towers, vehicles) is almost like the “servers”; the networked intelligence layer is the product.

Palantir Technologies — data infrastructure and decision graphs

Palantir is less of a weapons company and more like a military-grade data operating system.

Its stack resembles:

• a distributed data model across fragmented sources
• knowledge graphs
• event streams
• operational workflows
• AI agents that query and act on organizational data

For defense users, the interesting shift is from “one platform carries one weapon” toward “many platforms contribute data into a shared decision fabric.” Palantir’s AI and operational platforms are designed around that model.

Shield AI — autonomous agents and multi-agent coordination

Shield AI is building something closer to a distributed robotics system. Its Shield AI Hivemind stack is essentially an AI pilot layer:

• autonomous perception
• planning and decision-making
• operation in GPS- or communications-denied environments
• multi-agent coordination

The interesting abstraction is not “a better drone”; it is “a reusable autonomy runtime that can inhabit many vehicles.”

Helsing — battlefield data mesh

Helsing is building software that looks like a defense-oriented intelligence mesh:

• ingesting many battlefield data sources
• creating shared situational awareness
• helping humans and machines make decisions faster

The company has described its focus around AI-enabled defense systems and has expanded into autonomous platforms, but the software layer is the core differentiator.

Saronic Technologies — fleets as distributed systems

Saronic is more hardware-forward, but its conceptual model is distributed:

• many inexpensive autonomous vessels
• networked coordination
• persistent sensing
• software-defined missions

It resembles cloud architecture applied to naval operations: scale horizontally rather than rely only on a few extremely expensive platforms.

Scale AI — data pipelines for military AI

Scale is not primarily a weapons company. Its defense relevance is closer to the “data infrastructure” layer:

• training data
• model development workflows
• AI evaluation
• operational AI systems

In a distributed-systems analogy, Scale is closer to the data engineering layer that makes the rest of the stack possible.

---

The common pattern

Traditional defense architecture:

platform → operator → mission New defense-tech architecture:

sensors → data fabric → AI agents → autonomous systems → human oversight The companies that most resemble distributed systems companies are:

1. Anduril — distributed battlefield operating system
2. Palantir — defense data/decision infrastructure
3. Shield AI — distributed autonomous agents
4. Helsing — battlefield intelligence mesh
5. Saronic — autonomous fleet networking

The deeper shift is that the “weapon” is increasingly becoming the network and software layer that coordinates thousands of cheaper, connected assets rather than a single physical platform.

Traditional defense architecture:

platform → operator → mission New defense-tech architecture:

A major shift is happening in defense technology: rather than building monolithic, multi-billion-dollar hardware platforms (like a single jet or radar tower), a new wave of startups is treating the battlefield like an unreliable, highly distributed cluster of microservices. These companies rely heavily on edge compute, mesh networking, containerized tactical orchestration, and event-driven architectures.

The defense tech startups and platforms building software that maps cleanly to distributed systems principles include:

• Armada — Operating much like a distributed cloud provider for the tactical edge, Armada builds ruggedized, portable micro-data centers (Galleons) integrated with satellite constellations like Starlink. Their software platform orchestrates compute, data storage, and AI processing locally where it's generated, removing total reliance on centralized continental cloud data centers.
• Anduril Industries (Lattice OS & Lattice Mesh) — While Anduril builds physical hardware (drones, counter-UAS, autonomous subsea vessels), the actual core of their enterprise is Lattice . Lattice treats every sensor, effector, human operator, and autonomous drone as a stateless node in a massive IoT mesh network. The software dynamically re-routes data, handles intermittent connectivity, and builds a common operating picture via decentralized consensus rather than a top-down rigid command structure.
• Defense Unicorns — Specializing in software delivery for the tactical edge, they build tooling (like UDS Fleet) designed to let operators deploy, update, and manage complex mission software across air-gapped, fractured, and distributed military hardware. They use an open-source, Kubernetes-centric approach to make remote military nodes look like a manageable cluster.
• Hadean — Known for its spatial computing and distributed simulation engine (Aether Engine), Hadean abstracts the complexities of distributed processing nodes. In defense, they provide the spatial-temporal orchestration layer for massive, multi-domain synthetic wargames and real-time data fusion that traditional cloud setups choke on.
• Havoc AI — Building software for collaborative autonomy, Havoc focuses on how large swarms of low-cost, uncrewed systems coordinate intent and execute missions. Their command-and-control stack acts as an orchestrator for distributed, asynchronous agents where individual nodes can drop offline without breaking the broader mission objective.
• Distributed Spectrum — This electronic warfare startup builds decentralized, AI-powered radio frequency sensors. Instead of relying on a massive central intelligence listening post, they deploy tiny, cooperative nodes that talk to each other to map and triangulate enemy emissions across a contested spectrum.

If you're looking at the engineering side, would you like to dive deeper into how they handle consensus and intermittent network partitions (CAP theorem challenges) in tactical environments, or are you more interested in the hardware-software integration layer?

If you mean “defense startups whose core product feels like a distributed systems problem — nodes, networks, orchestration, sensor fusion, autonomy, edge compute, resilient communications — rather than a single tank/aircraft/missile,” the interesting companies are mostly building software layers that coordinate many heterogeneous systems.

A shortlist:

A useful mental model:

• Traditional defense contractor: “Build the aircraft, ship, missile, or radar.”
• Distributed-systems defense startup: “Build the operating layer that lets thousands of aircraft, sensors, robots, humans, and data sources behave like one system.”

The closest analogies from software engineering:

The companies most explicitly pursuing the “software-defined battlefield” thesis are probably Anduril, Palantir, Shield AI, and Helsing. They are trying to own the coordination layer, because future conflicts may involve thousands of cheap autonomous systems rather than a small number of exquisite platforms.

If you’re looking specifically for startups founded by ex-Silicon Valley distributed-systems engineers (rather than defense companies with software), the list gets narrower and includes companies like Picogrid, Helsing, and some newer autonomy infrastructure startups.

The modern defense tech ecosystem has experienced a massive shift: instead of focusing exclusively on heavy, standalone hardware (like building a better tank or a single jet), a wave of venture-backed companies and defense software firms are treating the entire theater of operations as a distributed system.

These platforms rely on edge computing, containerized microservices, resilient mesh topologies, and decentralized state synchronization—essentially treating military units, sensors, and autonomous swarms as nodes in a massive, fault-tolerant network.

Key defense startups and platforms building software that mirrors distributed systems include:

• Anduril Industries — Best known for Lattice OS , an AI-enabled command-and-control platform that functions like a distributed IoT network for the battlefield. Lattice ingests data from disparate nodes (stationary towers, drones, ground sensors, and counter-UAS platforms) and decentralizes the common operating picture, allowing autonomous effects to coordinate without needing a heavy, centralized cloud connection. Their modular field hardware (like the Menace-T portable edge data center) explicitly brings distributed cloud and routing architecture to off-grid environments.
• Shield AI — Creators of Hivemind , an AI pilot software platform designed for autonomous aircraft. Rather than relying on constant, vulnerable teleoperation from a ground station, Hivemind uses edge-computing architecture to let drones and aircraft run decentralized, collaborative autonomy in GPS-denied or communications-constrained environments. The aircraft act as intelligent, independent nodes that can dynamically share tasks in a swarm.
• Helsing — A European defense AI startup building software layers that process operational and sensor data across multiple military domains. Their platform is designed to run synchronously across distributed assets, providing real-time data fusion and electronic warfare coordination that looks much more like a distributed enterprise software mesh than a legacy weapons fire-control computer.
• Firestorm Labs — Focused on modular, open-architecture, 3D-printed uncrewed systems managed by an adaptive software layer. Firestorm treats hardware as commodity endpoints attached to a decentralized software orchestration engine, optimizing for rapid instantiation and distributed manufacturing/deployment logic.
• Onebrief — Tackles military planning and joint operations by turning the notoriously sluggish, document-heavy process of staff planning into a collaborative, real-time distributed application. It acts less like a legacy defense tool and more like a modern, cloud-native collaborative workspace tailored for high-tempo synchronization across dispersed command nodes.
• Kinnami Software (AmiShare) — Specializes in resilient, decentralized data integrity and synchronization for the tactical edge. Moving away from traditional centralized data repositories, Kinnami’s software treats data as a distributed mesh where information self-replicates, secures, and updates peer-to-peer across unreliable, intermittently connected military nodes.

Would you like to dive deeper into how these companies handle consensus and state synchronization in GPS/comm-denied environments , or are you looking at a specific layer of the stack (like tactical edge mesh routing vs. C2 application layers)?

Yes. If by “looks more like distributed systems” you mean the core engineering problem is coordination, networking, state, deployment, fault tolerance, and computation across unreliable nodes—rather than designing a new missile/airframe—there’s a surprisingly interesting cluster.

The ones I’d watch most closely

A useful way to divide the category

I’d actually split these into four layers:

1. Distributed infrastructure

This is the most “software engineer” end of the spectrum:

Defense Unicorns → Gentle Weapons → HardpointAI

Their fundamental problems are deployment, orchestration, identity, networking, updates, observability, offline operation and synchronization. Defense Unicorns is particularly explicit about making software delivery work across air-gapped tactical systems.

2. Distributed autonomy

Shield AI → Lagrange → Fractal Computing

Here the distributed system isn't just infrastructure—it is the intelligence. Nodes have partial information, communications can disappear, and the system has to continue making decisions.

Shield AI is the more mature example: Hivemind is explicitly platform-agnostic and supports coordinated multi-agent autonomy in GPS- and communications-denied environments.

Lagrange is particularly interesting because its pitch essentially starts with the classic distributed-systems problem: how do multiple agents stay coordinated when the network breaks?

3. Distributed command-and-control

Anduril → Palantir

This is where the abstraction becomes a distributed operating environment for the force.

Anduril's Lattice is probably the strongest embodiment: it describes itself as connecting thousands of first- and third-party systems, while Lattice Mesh is a decentralized network that dynamically distributes data across nodes.

Palantir is the more data-centric version—data integration, model deployment, edge processing and distributed multi-domain operations rather than primarily controlling autonomous vehicles.

4. Distributed sensing

Arlo is a neat example of taking the distributed-systems idea all the way down to the sensor layer. Instead of “make the radar better,” the architecture is “make the network of cheap sensors smarter.” Its thesis is explicitly that a mesh architecture changes the economics of wide-area tracking.

The really interesting pattern

The architectural shift is roughly:

Traditional defense: platform → subsystem → operator
Software-defined defense: node ↔ node ↔ node → shared state → autonomous behavior

That makes a bunch of classic distributed-systems concepts suddenly strategically important:

• eventual consistency
• partition tolerance
• gossip / peer-to-peer networking
• service discovery
• zero-trust identity
• edge caching
• offline-first software
• state synchronization
• graceful degradation
• dynamic task allocation
• observability across unreliable nodes
• continuous deployment to disconnected systems

The particularly interesting companies are therefore not necessarily the ones making the most sophisticated weapon. They're the ones making the network of weapons behave like one system.

If I were building a watchlist specifically for distributed-systems engineers, I'd rank the current crop roughly:

Anduril ≈ Shield AI > Defense Unicorns > Lagrange > Fractal Computing > Gentle Weapons > HardpointAI > Arlo

—with the caveat that the bottom half are much earlier and have considerably less publicly demonstrated scale.

The defense tech ecosystem is undergoing a massive architectural shift. Rather than building monolithic, proprietary hardware platforms (the "exquisite" jet or tank approach), a new wave of software-defined defense startups treats the entire battlespace as a distributed, edge-computed system of systems.

These companies focus on mesh routing, distributed state management, decentralized node orchestration, and autonomous consensus under degraded network conditions.

Key defense startups and platforms building software that mirrors distributed systems architecture include:

• Anduril Industries — Known for Lattice OS , which functions as an AI-enabled IoT operating system. Lattice operates as an edge data mesh, ingesting telemetry, video, and sensor tracks from heterogeneous hardware (towers, drones, ground sensors) and synchronizing state across a decentralized network without relying on a permanent cloud connection.
• Shield AI — Creator of the Hivemind autonomy stack. Hivemind acts as a decentralized "AI pilot" that treats fleets of uncrewed aircraft like a distributed compute cluster, allowing assets to share local processing, coordinate maneuvers, and execute mission logic even when completely disconnected from GPS or command-and-control links.
• Defense Unicorns — Specializes in airgap-native, portable software delivery. Their UDS (Unified Data Service) Tactical Edge treats tactical military hardware (tanks, ships, forward operating bases) as remote Kubernetes-style deployment targets, allowing continuous delivery, package management, and zero-trust updates at the furthest edge of the network.
• Forterra — Builds Vektor , an edge-deployed, software-defined data-brokering and communications layer. It acts as an adaptive middleware designed specifically to manage state and route data for autonomous vehicle swarms operating in Disrupted, Degraded, Intermittent, and Limited-bandwidth (DDIL) environments.
• Swarm Aero / Swarm Defense Technologies — Focuses heavily on collaborative autonomy and path deconfliction engines (Legion / Swarm Studio ). Their software handles distributed consensus, collision avoidance, and dynamic task allocation so that a single human operator can orchestrate thousands of decentralized nodes simultaneously.
• Meshmerize / Rajant (Tactical Edge Mesh) — While Rajant provides the physical kinetic mesh hardware layer, software-focused innovators like Meshmerize provide software-defined routing that treats mobile units as dynamic, self-healing network nodes that continuously route packets around interference or destroyed infrastructure.

If you're exploring this architectural shift, would you like to dive deeper into how these startups handle consensus algorithms in GPS-denied environments , or are you more interested in the Modular Open Systems Approach (MOSA) standards driving procurement?

Yes. If by “looks like distributed systems” you mean the hard product is the software layer that coordinates many heterogeneous nodes under unreliable connectivity, rather than a single exquisite platform, there’s a pretty clear group emerging.

The strongest examples

1. Anduril — probably the clearest example

   Lattice is essentially a military distributed-computing/control plane: it ingests sensor data, maintains a shared operational picture, runs autonomy, and connects sensors and effectors across different platforms. Its Lattice Mesh is explicitly a decentralized mesh that distributes data among nodes at the tactical edge, including degraded/low-bandwidth environments.

   The especially interesting bit is that the Army's 2026 NGC2 effort describes the architecture in almost pure distributed-systems language: edge processing, mesh networking, no central point of failure, common data foundations, continuously integrable applications.

   Software analogy: Kubernetes/service mesh + real-time event bus + edge compute + autonomous agents.

2. Shield AI — the autonomy-runtime version

   Hivemind is less about C2 dashboards and more about the runtime/decision-making layer for autonomous agents. It is designed to operate when GPS and communications are degraded, execute missions autonomously, and coordinate multiple vehicles. Shield AI explicitly positions it as platform-agnostic software that can be integrated into different unmanned systems.

   That's a very distributed-systems-shaped problem: each vehicle has partial state and compute, communications are intermittent, and the system has to make locally robust decisions while coordinating globally.

   A particularly strong signal: Hivemind is now being used alongside Anduril's software on the Air Force's Collaborative Combat Aircraft effort, rather than requiring Shield AI to own the aircraft itself.

   Software analogy: distributed robotics runtime + multi-agent systems + edge AI.

3. Helsing — software-defined mass

   Helsing is interesting because it is pushing toward large numbers of relatively inexpensive autonomous nodes rather than a few exquisite platforms. Its Lura system, for example, is software/AI for underwater surveillance and is designed to coordinate large constellations of autonomous sensors.

   That's much closer to designing a distributed sensor network than designing a conventional weapons platform.

   Software analogy: massively distributed sensor network + autonomous edge agents.

4. Saronic — distributed systems applied to fleets

   Saronic's Echelon is particularly revealing. It is a C2/software layer intended to command an entire fleet of autonomous vessels, with multi-asset learning, edge AI and collaborative behaviors. Their vessels themselves are networked and designed for multi-asset teaming.

   So although Saronic absolutely builds hardware, the strategic abstraction is “fleet as a software system,” not “boat as a product.”

   Software analogy: fleet scheduler/orchestrator + robotics middleware + distributed control plane.

5. Epirus — software-defined effects

   Epirus is somewhat different because its core product is a directed-energy system. But Leonidas is explicitly software-defined: waveform optimization, AI-enabled targeting, sensor integration, and software-controlled effects. Its systems can be integrated into larger sensor/C2 architectures.

   The interesting software architecture is therefore separating the “effect” from the platform that discovers and manages targets. Its integration with Anduril Lattice is a good concrete example.

   Software analogy: programmable accelerator/device + API rather than monolithic weapon.

The next tier

I'd also watch True Anomaly, Aalyria, Onebrief, Govini, and Scale AI's defense business, though they're different flavors.

• True Anomaly is pushing toward autonomous space systems where spacecraft, sensors and software operate as a network rather than as isolated satellites.
• Aalyria is particularly interesting if you're thinking actual networking infrastructure: its technology comes from advanced optical/communications networking rather than traditional weapons.
• Onebrief is closer to collaborative software/workflow infrastructure than autonomy: think military planning software as a modern SaaS product.
• Govini is essentially defense data infrastructure/analytics.
• Scale AI sits at the data/model layer rather than the distributed tactical-runtime layer.

And Palantir deserves mention even though I wouldn't call it a startup anymore. Its defense business explicitly focuses on software at the tactical edge and distributed multi-domain operations.

My mental model

I'd divide the emerging stack like this:

The really interesting architectural shift is that the weapon/vehicle is increasingly becoming a node in a distributed system. Anduril's own description of Lattice as a network connecting sensors and effectors—and its explicit move away from hub-and-spoke architectures—is probably the clearest manifestation of that idea.

If you're asking this from a software-engineering career/company-building perspective, I'd put Anduril, Shield AI, Helsing, and Saronic at the top of the list. Those are the ones where concepts like consistency under partition, edge computation, service discovery, message passing, swarm coordination, simulation, fault tolerance, and real-time systems are becoming unusually central to the actual product.