The defence industry spent the last few years proving that autonomous platforms could work.

Now something more consequential is happening.

Governments are starting to build around the software that connects them.

NATO is integrating commercial AI companies directly into its operational digital environment. The U.S. Air Force is separating mission-autonomy software from the aircraft underneath it. Saronic is building a shipyard around autonomous vessels and its own fleet-management architecture. Anduril is considering another major manufacturing footprint for autonomous boats. Ondas is assembling a multi-domain autonomous portfolio through acquisition while simultaneously receiving new U.S. government development funding.

NATO DIANA selected 10 companies for its Decision Superiority challenge, with their AI/ML tools being integrated into a cloud environment alongside NATO's Maven Smart System.

Signal: NATO is moving from buying individual technologies to testing whether they can plug directly into the Alliance's digital operating environment.

The Air Force's Collaborative Combat Aircraft architecture separates mission-autonomy software from the airframes, with six vendors competing in the software marketplace.

Signal: The aircraft is becoming the hardware layer. The autonomy stack is becoming the competitive layer.

Saronic selected Brownsville, Texas for its $3.2B Port Alpha shipyard, with an 835-acre initial site, construction planned for 2026 and production expected to begin in 2028.

Signal: Autonomous-ship companies are no longer trying to fit themselves into the existing shipbuilding base. They're building a new one.

Anduril is in advanced negotiations to establish an unmanned-surface-vessel production site at Baltimore's Sparrows Point, potentially involving hundreds of millions of dollars in investment.

Signal: The neo-primes are becoming industrial landlords, not just defence software companies.

Ondas announced on August 6 that AFRL awarded DZYNE, now part of Ondas, a $6M+ contract to advance Long-Range Grasshopper, an autonomous aerial-delivery system designed for contested logistics.

Signal: The autonomy race is moving beyond weapons into the less glamorous — and potentially more strategically important — problem of keeping distributed forces supplied.

Table

Overall Signal: 93/100 — Integration Is Becoming the Moat

The market is beginning to reward the architecture connecting autonomous systems, not merely the systems themselves.

In July, NATO DIANA selected 10 companies for its Decision Superiority for NATO Warfighters challenge.

The companies are developing AI, machine-learning and software capabilities for modelling, simulation, targeting support and operational planning.

But the interesting part isn't the list of startups.

It's where NATO wants to put their technology.

The systems are being integrated into a cloud-based environment aligned with NATO's operational use cases and existing AI-enabled digital warfighting infrastructure, including Maven Smart System NATO.

Each selected company receives €100,000 to support integration and demonstration. NATO says successful demonstrations may become eligible for follow-on prototyping and potentially production contracts through its Rapid Adoption Service.

This is procurement behaving differently.

NATO isn't simply asking:

"What does your AI do?"

It's increasingly asking:

"Can your AI plug into ours?"

That distinction is enormous.

The platform becomes the distribution mechanism.

The vendor becomes the application.

And once the Alliance's operational users become accustomed to a particular data model, interface or workflow, the integration itself becomes part of the product's competitive advantage.

This is exactly how software platforms become sticky.

Except the customer isn't a marketing department. It's a military alliance.

The next generation of defence startups should be judged partly by integration velocity.

A brilliant model that takes 18 months to certify and integrate may lose to a slightly less impressive system that can operate inside an allied architecture in six weeks.

Direct Beneficiary: Defence software companies built for open architectures.

Watch: Hadean, Onebrief, Picogrid, Matrix Pro Sims and other DIANA-selected innovators.

Next Trigger: NATO's planned December demonstration phase and any follow-on contracts emerging from it.

The U.S. Air Force's Collaborative Combat Aircraft program is doing something strategically unusual.

It's separating the aircraft from the autonomy software.

The service has established a six-vendor mission-autonomy marketplace involving Anduril, General Atomics, Lockheed Martin, Northrop Grumman, RTX Collins Aerospace and Shield AI.

The architecture is designed so mission autonomy can operate across different aircraft platforms rather than being permanently tied to one manufacturer's airframe.

This may ultimately matter more than the aircraft competition itself.

Traditional defence economics says:

Build the platform → integrate the software → sell the platform.

The CCA architecture increasingly suggests:

Build the platform → expose the architecture → compete continuously on the software.

That changes the economics. An airframe can remain in service for decades. The software shouldn't.

If autonomy can be upgraded independently, the military doesn't need to buy an entirely new aircraft every time the algorithm improves.

It can potentially upgrade the brain while keeping the body.

That creates recurring competitive pressure — and potentially recurring software economics — inside a market historically dominated by one-time platform procurement.

The strategic winner may eventually be the company that owns the autonomy standard rather than the company that builds the aircraft.

Direct Beneficiaries: Anduril, Shield AI, RTX Collins and other mission-autonomy suppliers.

Watch: Which autonomy stack becomes the default across future aircraft programs.

Next Trigger: Production software awards and the first evidence that mission autonomy can migrate between multiple operational platforms at meaningful speed.

Saronic Technologies selected Brownsville, Texas for its proposed Port Alpha shipyard.

The project is valued at approximately $3.2 billion.

The initial site covers roughly 835 acres. Saronic says construction is expected to begin in 2026, with the facility expected to be capable of building vessels up to 850 feet in its first phase.

The project could eventually scale much further, and the company has said it could create up to 10,000 jobs over a decade.

The obvious story is shipbuilding.

The more interesting story is industrial architecture.

Saronic is vertically combining:

Its Echelon platform is designed to command individual vessels or fleets through a common interface. That's not simply a boat company.

It's an attempt to build a maritime operating system with a factory attached.

And that combination matters. The company controls the software feedback loop. It controls the production environment. It controls the testing environment.

And every operational deployment potentially produces data that can feed back into the next software and hardware iteration.

That is a very different industrial model from traditional shipbuilding.

The autonomous ship market may not be won by the company with the best vessel.

It may be won by the company with the fastest closed loop between software → factory → fleet → data → software.

Direct Beneficiary: Saronic.

Watch: Whether Port Alpha becomes a true high-rate autonomous manufacturing environment or simply another conventional shipyard with better software.

Next Trigger: Construction milestones and the first evidence of production throughput.

In August, The Wall Street Journal reported that Anduril was in advanced negotiations to establish an unmanned surface vessel production site at Baltimore County's Sparrows Point industrial complex.

The site is a former steel and shipbuilding hub.

The potential investment could reach hundreds of millions of dollars, with manufacturing and testing operations planned if an agreement is completed.

Anduril has already begun low-rate USV production in Seattle and has worked with HD Hyundai on maritime manufacturing.

This is the part most people miss.

The new defence companies aren't merely competing with primes on technology.

They're competing with them on industrial geography.

Sparrows Point isn't interesting because it's a nice place to build boats.

It's interesting because it already has the physical attributes of industrial mobilisation:

The new defence industrial base may therefore emerge less from building entirely new infrastructure and more from reprogramming existing industrial infrastructure around autonomous production.

That is faster.

And potentially much cheaper.

The defence-industrial map of the United States could start changing before the Pentagon formally announces that it is changing.

Watch where companies buy land. Watch where governors offer incentives. Watch which old factories suddenly become interesting again.

The map may tell us where procurement is going before procurement announcements do.

Direct Beneficiary: Anduril and the broader U.S. autonomous maritime ecosystem.

Watch: Maryland incentives, Sparrows Point negotiations and additional autonomous-production site announcements.

Next Trigger: A finalized lease or investment commitment.

On August 6, Ondas announced that the Air Force Research Laboratory awarded DZYNE Technologies, now part of Ondas, a more than $6 million contract to advance Long-Range Grasshopper.

The system is designed as an autonomous, runway-independent aerial delivery platform for extended-range logistics in contested and austere environments.

DZYNE was acquired by Ondas in July in a transaction valued at approximately $875 million. The acquisition brought autonomous logistics, ISR, counter-UAS and autonomous-effects capabilities into Ondas' newly created Sentinel division.

Defence AI conversations still obsess over targeting, but targeting is only half the problem.

You also have to survive long enough to get the ammunition there.

Distributed warfare creates an ugly logistics problem. Runways can be attacked. Roads can be monitored. Satellites can be degraded. Communications can be jammed. Large transport aircraft are expensive and visible.

That creates a growing market for autonomous logistics systems that can operate without conventional infrastructure.

Grasshopper is interesting because it applies autonomy to something militaries desperately need but rarely make headlines with:

resupply.

And the Ondas transaction adds another layer. The company isn't building a single autonomous vehicle, it's assembling a portfolio spanning ISR, counter-UAS, autonomous effects and logistics.

The autonomous warfare market will eventually be measured not just by how many targets a system can find.

It will be measured by how much logistical burden it removes from humans.

That could become one of the biggest — and least glamorous — applications of defence autonomy.

Direct Beneficiary: Ondas / DZYNE.

Watch: Grasshopper's transition from development into larger procurement and fielding.

Next Trigger: Follow-on AFRL or service procurement beyond the current development award.

The market is starting to reveal a hierarchy.

At the bottom:

Platforms.

Above them:

Autonomy.

Above that:

Integration.

And above integration:

The data and command architecture that determines how the entire force behaves.

That hierarchy matters because the higher you move, the harder the layer becomes to replace.

A drone can be shot down. A vehicle can be upgraded. A sensor can be swapped.

But replacing the digital architecture connecting thousands of systems across an alliance?

That's a different procurement decision entirely. And potentially a much bigger moat.

Do the 10 DIANA companies actually become part of operational workflows, or do they remain impressive demonstrations?

Which autonomy vendor becomes indispensable across multiple airframes?

Does Saronic demonstrate production economics fundamentally different from traditional shipbuilding?

Where do Anduril, Saronic and other neo-primes build next?

Does autonomous resupply become a formal procurement category rather than an R&D experiment?

If the integration layer becomes the moat, expect companies to start buying the missing pieces rather than building everything internally.

The first phase of defence AI was about proving that machines could do things humans normally do.

The second phase was about manufacturing enough of them. The third is becoming something more interesting.

Making them behave like one force.

That means the next great defence platform may not look like a tank, aircraft or ship. It may look like software.

And the company sitting between the sensor, the drone, the command post, the cloud and the human operator may ultimately control more of the value chain than the company that built any one of those things.

The weapon is becoming the node. The network is becoming the weapon.

And the integration layer? That's where we're looking next.

01 — NATO DIANA / Decision Superiority

NATO DIANA announcement — Selection of 10 companies and integration into NATO's digital warfighting environment.
NATO DIANA challenge documentation — TRL requirements, €100K funding, integration process and follow-on procurement pathway.

02 — USAF / Collaborative Combat Aircraft

USAF CCA software architecture and contracts — Six-vendor mission-autonomy marketplace and production framework.
USAF open architecture announcement — Government-owned Autonomy Government Reference Architecture and separation of software from airframe.

03 — Saronic / Port Alpha

Saronic official site — Port Alpha and Echelon architecture.
USNI reporting on Port Alpha — $3.2B project, 835-acre site, construction and production timeline.

04 — Anduril / Baltimore

Wall Street Journal reporting — Advanced negotiations for a Sparrows Point autonomous-vessel manufacturing facility.

05 — Ondas / DZYNE / Grasshopper

Ondas DZYNE acquisition announcement — $875M acquisition and autonomous-defense portfolio.
Ondas AFRL Grasshopper announcement — August 6 AFRL contract exceeding $6M.

Defence AI is no longer just a race to build smarter machines — it's becoming a race to own the software layer that makes thousands of machines operate as one.