Replicator 3: AI Sustainment for Unmanned Readiness

A fleet that can’t stay ready isn’t a fleet—it’s inventory.

The Pentagon’s Replicator effort has pushed the conversation toward scale: lots of affordable autonomous systems, fielded fast, across the Indo-Pacific. That’s the right instinct. But procurement is the easy part to celebrate. Sustainment is where readiness lives or dies, and it’s also where AI can deliver the most immediate advantage.

Nick Johnson’s argument lands because it’s painfully familiar: the U.S. military already struggles to keep mature, manned platforms available. Government audits found that from 2011–2021, Arleigh Burke-class destroyers averaged 25 days of maintenance delays per ship and 23 severe casualty reports per year (a severe casualty being a complete loss of a mission area). If we can’t consistently keep a crewed surface combatant with established depots and trained maintainers fully mission-capable, betting deterrence on thousands of distributed unmanned systems without a sustainment plan is wishful thinking.

Replicator 3 should exist for one reason: combat power is a function of uptime, not purchase orders. And the fastest route to higher uptime in a distributed, contested theater is AI-enabled sustainment and logistics.

Replicator’s missing pillar: readiness at scale

Answer first: Replicator 3 should focus on sustainment because the Indo-Pacific concept of operations demands operational availability across distance, corrosion, and contested supply lines.

Replicator-1 (mass fielding) and Replicator-2 (defensive layers) are about getting capability into the fight. Replicator-3 should be about keeping it there—week after week—despite failures, software issues, battery degradation, component corrosion, and supply disruptions.

Here’s the trap: when leaders hear “attritable,” they sometimes translate it as “maintenance-light.” That works for some short-life systems in specific contexts. It does not work for long-range maritime autonomy, persistent sensors, and power-hungry platforms that need to sit forward for months, survive weather, and remain secure and controllable.

If Replicator 3 is framed as “just logistics,” it will lose in budget and attention. It should be framed as what it is: the readiness engine for autonomous warfare.

Why this matters more in 2026 than it did in 2016

December 2025 is a useful moment to be blunt. The U.S. and its allies are watching rapid iteration in unmanned systems globally, but the hard lesson from recent conflicts isn’t only “drones matter.” It’s: production without sustainment creates hollow force.

In the Indo-Pacific, distance and environment punish neglect:

• Salt, humidity, and heat accelerate corrosion and battery wear.
• Long lines of communication raise the cost of every spare part.
• Basing and access are political variables, not guaranteed constants.
• “Forward” often means austere sites with limited power and skilled labor.

That’s why Replicator 3 shouldn’t be a nice-to-have. It should be the program that turns unmanned mass into credible deterrence.

The myth: unmanned means a smaller sustainment tail

Answer first: Unmanned systems don’t automatically reduce sustainment—they often shift it from shipboard labor to distributed maintenance, software support, and fragile supply chains.

The popular mental model is: no crew means less life support, fewer people, fewer problems. Real life isn’t that generous.

Unmanned fleets introduce sustainment work that’s easy to underestimate:

• Battery and power management (storage conditions, cycling, replacement, safety).
• Sensor calibration and cleaning (salt spray, dust, biofouling).
• Software patching and configuration control across heterogeneous platforms.
• Secure communications upkeep (keys, waveforms, interference mitigation).
• Swapping modular components quickly in the field, at scale.

Johnson points to a commercial analogy that defense teams should take seriously: even highly automated vehicle fleets still employ large technical staffs. Autonomy doesn’t erase labor; it redistributes it. If anything, autonomy increases the premium on technicians who can do rapid swaps, validate systems, and return them to service under time pressure.

Indo-Pacific isn’t Ukraine, and copying the wrong lesson is risky

Answer first: The Indo-Pacific requires long-endurance, corrosion-resistant, modular unmanned systems supported by forward sustainment nodes—not a high-churn quadcopter replacement model.

Ukraine has taught the world how fast small unmanned systems can be produced, adapted, and consumed. The Pacific is a different problem set:

• It’s an “away game” relying on allies and partners.
• The geography is maritime and immense.
• Many systems will sit idle in peacetime—meaning storage and upkeep become readiness determinants.

A stored unmanned system is not a “saved” unmanned system. It’s a liability unless its batteries, seals, lubricants, software baselines, and corrosion controls are managed deliberately.

Where AI actually helps: the sustainment stack

Answer first: AI in military logistics works best when it’s used to predict failures, optimize spares, orchestrate maintenance workflows, and verify readiness with data—not when it’s treated as a dashboard accessory.

Sustainment is an information problem disguised as a wrench problem. That’s why AI belongs at the center of Replicator 3.

Here’s the sustainment stack that consistently delivers results in complex fleets—military or commercial.

1) Predictive maintenance that’s tied to decisions, not reports

Condition-based maintenance is already a known idea. The difference in an AI-enabled model is decision automation:

• Predict failure probability for specific components based on telemetry and environment.
• Trigger pre-positioning of spares before a failure grounds a platform.
• Recommend maintenance actions that fit the site’s tools and technician skill levels.

The goal isn’t “fewer maintenance events.” The goal is higher operational availability with fewer surprises.

2) Demand forecasting for spares in a contested theater

Traditional supply planning assumes relatively stable transportation and predictable lead times. That assumption collapses quickly in a fight.

AI can help by:

• Modeling consumption rates under different operational tempos.
• Optimizing stockage levels for forward nodes under weight/volume constraints.
• Recommending substitutions when a preferred part is unavailable.

This is where AI can plausibly create the “10x efficiency” effect people talk about—not by magic, but by reducing dead inventory, cutting emergency shipments, and preventing cannibalization.

3) Maintenance workflow orchestration for distributed teams

Unmanned scale creates a coordination problem: thousands of platforms, multiple variants, multiple vendors, multiple software baselines, and dispersed maintenance sites.

An AI-driven sustainment platform should be able to answer, in seconds:

• Which systems are mission-capable right now, and why not?
• Which failures are repeating across the fleet (root cause clustering)?
• Which forward node is best positioned—tools, spares, technicians—to fix what’s broken?
• What’s the fastest path to restore a required number of platforms for a specific operation?

That’s not “nice reporting.” That’s mission planning.

4) Digital threads that preserve government sustainment rights

A quiet but decisive point in Johnson’s piece: if the government can’t maintain what it buys, it doesn’t control readiness.

Replicator 3 should push hard on:

• Open, modular architectures
• Accessible technical data packages where feasible
• Sustainment data requirements that are verifiable and operationally relevant

AI thrives on clean, consistent data. If platform data is locked in vendor silos, “AI sustainment” becomes a PowerPoint concept instead of a readiness function.

What Replicator 3 should fund: nodes, power, people, partners

Answer first: Replicator 3 should resource forward sustainment nodes with resilient power, additive manufacturing where it’s practical, expeditionary maintainer teams, and partner-enabled regional capacity.

Johnson is right to insist that the future sustainment network shouldn’t look like yesterday’s big depots and centralized supply bases. It needs to be adaptive, dispersed, and survivable.

Build forward nodes that can operate under threat

A workable model looks like a mesh of small sites that can store, service, and relaunch systems with minimal resupply. Practically, that means containerized capabilities:

• Battery conditioning and safe storage
• Modular component swap benches
• Secure networking for software updates and mission data
• Environmental controls for corrosion mitigation

These nodes should be designed from day one to operate with intermittent connectivity and degraded logistics.

Treat power as a readiness constraint

Unmanned sustainment is power-hungry: charging, cooling, compute, comms, additive manufacturing, and environmental control.

Replicator 3 should be comfortable investing in resilient power options—especially for austere sites—because power equals sortie generation. Johnson cites options like small modular reactor concepts for larger nodes and off-grid energy for smaller units. Whether those specific technologies mature quickly or not, the underlying requirement is non-negotiable: forward sustainment without reliable power is pretend sustainment.

Create expeditionary maintainer teams for autonomy fleets

Every service should plan for a maintain-and-relaunch capability analogous to what aviation does well: fast turnarounds using trained technicians, standardized checks, and modular swaps.

Replicator 3 should fund:

• New maintainer career pathways for unmanned systems
• Cross-training for software + hardware troubleshooting
• Tooling and diagnostics that work in austere environments

This is also where AI can reduce training time: troubleshooting copilots, guided workflows, and fault isolation that’s standardized across vendors.

Make allies and partners part of the sustainment design

Deterrence in the Indo-Pacific is collective. Sustainment should be too.

A practical approach is partner-led sustainment nodes and shared inventory pools that allow smaller nations to contribute meaningfully without having to build entire supply chains alone. Replicator 3 can bake this in through:

• Common standards for data, parts, and maintenance procedures
• Shared training pipelines
• Regional repair capacity that benefits multiple partners

This isn’t just diplomacy. It’s operational resilience.

A Replicator 3 “scorecard” leaders can use immediately

Answer first: If Replicator 3 can’t measure readiness outcomes, it will drift into activity without impact.

If you’re evaluating an AI sustainment initiative—inside government or as an industry partner—push for metrics that reflect operational truth:

1. Mission-capable rate by platform type and theater location
2. Mean time to restore (MTTR) with forward repair vs. evacuation
3. Parts availability rate at forward nodes (not in a U.S. warehouse)
4. Repeat failure rate (are we fixing symptoms or causes?)
5. Software baseline compliance (how many are patched and configured correctly?)
6. Time-to-train maintainers to proficiency on key tasks

If a vendor can’t talk in these terms, they’re not selling readiness.

One-liner worth remembering: Production buys you inventory. Sustainment buys you deterrence.

Where this fits in AI in Defense & National Security

This is the part of the AI story that rarely gets headlines. People see autonomy on the edge—drones, swarms, decision aids. The less glamorous reality is that AI-enabled logistics is where militaries can bank durable advantage: higher readiness, faster repair cycles, smarter spare positioning, and more resilient operations under disruption.

Replicator 3, done right, becomes the bridge between autonomous systems and operational credibility. Done poorly—or ignored—it creates a paper force: impressive quantities, low availability, and commanders stuck making the same ugly choices auditors have documented for years (wait, cannibalize, or ration).

If you’re building, buying, or advising in this space, the next productive question isn’t “How many unmanned systems can we field?” It’s: How many can we keep mission-capable forward, for months, under threat?

If you want a practical starting point for your organization, I’d focus on a small pilot that proves the whole loop—telemetry to prediction to parts to maintenance to verified readiness—at a forward site. That’s where AI earns trust.

The procurement headlines will keep coming. The force that wins will be the one that can still launch tomorrow.