The economic argument for directed energy is not complicated. A high-energy laser can neutralize a drone swarm target for approximately $3.50 per shot. The same engagement with a PAC-3 MSE missile costs roughly $3.7 million. When the threat calculus shifts toward massed attritable systems — the drone swarm, the cruise missile salvo, the saturation attack — the cost-exchange mathematics of kinetic defense become strategically unsustainable. In March 2026, Assistant Secretary of Defense for Critical Technologies Michael Dodd framed the DoD's 36-month directed energy scale-up commitment explicitly around this cost logic, citing the expenditure of hundreds of Patriot missiles in recent operations as the forcing function. The Pentagon's FY2027 budget request includes more than $2 billion for directed energy research, development, testing, and evaluation — a figure that represents not just hardware investment but institutional commitment to treating high-energy lasers and high-powered microwave systems as first-tier defense infrastructure, not experimental capability.
The Army's concurrent decision to discontinue its most advanced laser program tells the other half of that story. In March 2026, the service confirmed it would not transition the 300-kilowatt IFPC-HEL system — developed under the Indirect Fire Protection Capability program and known informally as "Valkyrie" — to a program of record. Future funding has been eliminated from Army budget plans starting in FY2026, with the single prototype continuing through final lab testing before being divested as a development reference for the Joint Laser Warfighting System. The 300kW power class is more than sufficient to engage the cruise missiles and drone swarms the system was designed to defeat. What the IFPC-HEL experience surfaced is the hard gap between generating the beam and directing it effectively under the conditions where it needs to work: against maneuvering targets with hardened casings, at forward-deployed positions that face the highest threat density, in the atmospheric and electromagnetic conditions of contested operations. The physics of the kill mechanism are proven. The fire control intelligence required to apply that mechanism with tactical precision is where the engineering challenge actually lives.
The Three AI Problems Inside Every Laser Engagement
Effective directed energy employment against agile threats requires three distinct AI inference functions running concurrently and interdependently. The first is threat classification and prioritization: in a swarm scenario, determining which target represents the highest-consequence intercept — and sequencing engagements accordingly — demands processing sensor feeds faster than any human operator can manage. The second is aimpoint selection: identifying the structurally most vulnerable location on a target based on aspect angle, material estimate, and closing geometry, then handing that geometry to the tracking loop. The third and most latency-sensitive is aimpoint maintenance: continuously updating beam direction as the target maneuvers, compensating in real time for atmospheric turbulence, thermal blooming, and platform motion that would otherwise deflect or disperse the beam before it reaches effective fluence on target. Naval Postgraduate School research on AI-automated aimpoint selection and maintenance has demonstrated measurable lethality improvements against UAS targets — and has confirmed that these models produce valid outputs only when running on compute hardware fast enough to close the control loop at sub-100-millisecond latency. That compute requirement does not fit a network architecture where targeting inference is routed through an enterprise cloud.
Why DDIL Changes the Deployment Equation
The forward-deployed positions where directed energy matters most are precisely where network connectivity is most constrained. A DE fire control system operating in a denied or degraded communications environment cannot offload inference to a datacenter. The targeting AI has to run on edge hardware co-located with the weapon — hardened to survive the same shock, vibration, thermal range, and electromagnetic environment as the laser aperture and beam director it manages. AeroVironment's LOCUST X3, released in March 2026 as the third generation of its high-energy laser system, integrates AI directly into the targeting pipeline at the edge for exactly this reason: the engagement timelines for drone swarm defense do not permit any architecture that introduces network latency into the fire control loop. This is not a product differentiator — it is a basic operational requirement that every DE system deployed to contested forward positions will need to satisfy. The Army's E-HEL program, carrying a $994 million FY2027 budget request with intent to field 24 systems, will face the same architectural question as it moves from lab testing to forward deployment.
The manufacturing constraints on directed energy hardware — rare earth dependencies that China currently dominates, limited domestic production capacity for advanced photonics — are real and documented. But the 36-month fielding timeline the Pentagon has committed to will not be set back by optics production rate. It will be set back if the AI fire control layer is treated as an integration problem to be resolved after hardware delivery rather than a verified, tested capability exercised against realistic threat scenarios before the first platform ships. The gap between a laser that works in a lab against static targets and one that works in forward-deployed operations against maneuvering swarms is not a power gap. It is a fire control gap. Closing it requires the same adversarial, environment-aware testing — edge hardware in loop, degraded conditions, realistic target sets including multi-axis swarms and hardened cruise missile profiles — that every other AI system destined for the operational kill chain must pass through. Programs that treat fire control AI as first-class critical infrastructure from program inception will meet the 36-month mark. The ones that treat it as software-to-be-integrated will not.
