The microgrid autopilot: AI takes the wheel of defence power

The diesel generator has powered forward operating bases for a century. Artificial intelligence is about to make it optional.

Military microgrids were sold as the quiet upgrade: solar on a warehouse roof, a battery bank in a container, a controller that holds the line when the grid fails. The promise was resilience. The reality, until recently, was expensive hardware strapped to software that still wanted an engineer on site. A layer of AI orchestration is changing that, turning the microgrid from a backup box into something closer to a self-driving power plant.

The numbers behind the shift

The market is small but moving fast. Global military microgrid revenue reaches roughly $3.2 billion in 2026 and is forecast to pass $13.7 billion by 2035, a 17.2% compound rate, according to Global Market Insights. This is not a consumer trend. It is procurement, and procurement moves on mandates.

Why defence wants autonomy now

Three pressures line up at once. Energy resilience is now an operational mandate, not a sustainability line item. The US Department of Defense widened mobile and tactical microgrid investment under its Operational Energy Strategy in 2024, and the 2026 Microgrids & Energy Resilience Summit points to a target of 14-day energy self-sufficiency at installations. The load is changing as well. Electrified logistics, AI training systems, and directed-energy kit pull far more than a diesel set was sized for. The threat moved too. A cyber hit on a base grid, or on the satellite link feeding it, can black out a camp faster than any shell.

The deployments that already exist show the direction. Fort Hunter Liggett in California installed a $21.8 million renewable-based microgrid in 2024, the first Department of Defense base to do so. Camp Arifjan in Kuwait commissioned a microgrid in September 2024 that integrated solar, battery storage, advanced inverters, and backup generation. These are the static proofs. The autonomous layer is what turns a fixed installation into something deployable.

As we wrote in August, the military, not utilities, is pulling small nuclear to market. Microreactors are the firm-power anchor. Autonomous microgrids are the distribution brain that makes that anchor useful at the edge. The two stories are the same bet on energy independence, told at different voltages.

What is actually falling

Diesel still dominates. Generators held 34.4% of the military microgrid power-source mix in 2025 and are still growing, because long-duration backup has no clean substitute yet. But the logic of the fuel convoy is under pressure. Every truck that carries diesel into a contested zone is a target and a logistics tax. The European Defence Fund's INDY project showed what autonomy buys: energy-independent camps hitting 35–55% self-sufficiency, cutting logistical burden by 45% and ownership cost by 28%.

The new layer: the microgrid autopilot

This is the part that is genuinely new. Companies are bundling the controller, the forecasting, and the dispatch into one autonomous stack. NeutronX, an energy-technology company, markets a microgrid autopilot that manages generation, storage, and distribution in real time, with digital twins and renewable dispatch layered on top. In February 2026 it signed an agreement with NextNRG to deploy energy infrastructure for government and defence projects, and its bidding engine already powers federal contract submissions for NextNRG (NASDAQ: NXXT).

The hardware side is catching up. The US Army Research Lab has folded medium-voltage solid-state transformers (SSTs) into its forward-deployed microgrid work. The three requirements are brutal and specific: rapid deployment, harsh-environment thermal tolerance, and bidirectional power flow. A single mobile chassis with an SST can replace three or four conventional distribution units, which matters when the chassis has to fit a transport aircraft. Eaton brought an SST capability in-house with its August 2025 acquisition of Resilient Power Systems. The US Army already uses MIL-STD-3071 compliant middleware to specify next-generation tactical microgrid requirements, a sign the standard is hardening around autonomous control.

Procurement is real money, not a slide. The Army Corps' microgrid-relevant capital flow runs above $200 million a year, inside a $2 billion Parsons contract and a $2 billion ten-year IDIQ pipeline, per mgrid.org's tracking of SAM.gov awards. The annual contract flow exceeds $200 million in microgrid-relevant procurement, with a multi-year multiplier as forward-deployment tempo varies.

Autopilot versus the conventional controller

The comparison is not about removing the engineer. It is about moving the engineer from the loop to oversight. A conventional controller protects the kit. An autopilot treats the whole base as one optimizable system and acts when conditions change, without waiting for a human to read a dashboard.

The software stack is maturing in step with the hardware. Edge controllers now run machine-learning models locally, so a forward microgrid can rebalance after a hit without a round trip to a data center. Digital twins let operators rehearse a failure before it happens. The standards are following: MIL-STD-3071 gives the US Army a common language for tactical microgrid components, and interoperability, not raw capability, is what lets a base mix vendors without rewiring the control room. That is the quiet unlock. Qualified, interoperable autonomy turns a one-off demo into a procurement category.

The investment map: who owns the stack

The field splits into three groups. Incumbents that already sell into defence power, systems integrators such as Schneider Electric, GE, and Eaton, plus primes like Lockheed Martin and Rolls-Royce, are adding cyber-resilient microgrid control to portfolios they already hold. Their edge is qualification and procurement access, not software novelty. A second group owns the hardware discontinuity: SST and wide-bandgap semiconductor specialists, where Eaton's Resilient Power acquisition signals that the transformer, not the battery, may be the contested component for expeditionary power. The third group is the software layer, the autopilot itself, where NeutronX and a thin bench of energy-AI startups are trying to own the decision engine rather than the steel.

For an investor, the question is which layer compounds. Storage and diesel are commodities with margin pressure. Control software, once qualified and embedded across a base, looks more like an installed base than a product. The US Army Research Lab's forward-deployed microgrid work frames autonomy as the capability that turns a pile of hardware into a self-optimizing network. That is the layer with pricing power, and it is still up for grabs.

Cybersecurity is the choke point

Autonomy and attack surface arrive together. Every node that makes a decision is a node that can be spoofed. Embedded security, verified firmware, and encrypted node-to-node comms are design priorities, not add-ons, and the same AI that balances load can also flag suspicious traffic before a breach lands. The DOE inspector general flagged AI governance and cybersecurity as top management challenges for 2026, a warning that applies directly to autonomous energy infrastructure. The autonomy gap, in practice, is a trust gap: can a commander rely on a system that acts without a human in the loop when the alternative is a dashboard nobody is watching?

This is where the promised-versus-reality tension bites again. Demonstrations show self-healing grids in pilot. Fielded, ruggedized, military-qualified autonomy is rarer, and the gap between an announced capability and revenue service is where several solid-state transformer programs have quietly stalled. Track qualification milestones, not press releases.

Signals to track

The convergence is straightforward to read. Compute is moving to the edge of the grid, and the grid at the edge is moving to autonomy. The companies that own the autopilot layer, not the diesel set, will define how defence power works when the main grid cannot be trusted. The window to own that layer is open now, and it will close as quickly as the first programs of record are signed.