Back to the Future of War
War returns to industrial fundamentals: the side that manufactures faster wins.
Lessons from a Defense Tech Week conference in Tel Aviv.
The battlefield in five years: Autonomous drone swarms navigate 1,000 kilometers without GPS, running on quantum sensing and hydrogen not batteries. They identify targets through AI models running on the platform itself—no cloud, no communication, no latency and no vulnerability to jamming. High-power lasers intercept dozens of drones per engagement at the cost of electricity. When interceptors run low, 3D printing production lines shift from Arrow missiles to Iron Dome interceptors via software update—same factory, different munition, hours not months. Rockets on demand. A 3D-printed tank’s targeting system fails. The mobile data center on the vehicle retrains the model using battlefield data, air-gapped from external networks.
War returns to industrial fundamentals: the side that manufactures faster wins.
Israel’s defense establishment is rebuilding around this constraint. MAFAT—the Israeli defense R&D directorate equivalent to DARPA—sets technology priorities for the military. Rafael, IAI, and Elbit are Israel’s three defense manufacturing giants, together accounting for roughly 80% of Israeli defense exports and $15 billion in annual revenue. Their CTOs rarely speak publicly about production bottlenecks and technology gaps. At a Defense Tech Week conference at Tel Aviv University in December, they described the same problem from different angles: manufacturing velocity determines strategic outcomes, and current production models can’t match threat tempo.
The stakes are global. China produces a million drones monthly through DJI and builds ships at 200 times the US pace. Russia and Ukraine burned through decades of munition stockpiles in months. Iran’s April saturation attack forced Israel to fire interceptors costing millions against drones costing thousands. The economic asymmetry is unsustainable. The countries that solve manufacturing at scale—high-mix, high-volume production with continuous technology refresh—set engagement terms. The countries that don’t face operational irrelevance.
The Directed Energy Inversion
Rafael’s Iron Beam intercepted dozens of drones and loitering munitions in October—operational deployment against actual enemies, not laboratory tests. The system combines high-power lasers with tracking systems and adaptive optics compensating for atmospheric conditions. Rafael now fields systems from Iron Beam’s strategic power down to Light Beam’s 10-kilowatt tactical units.
The economics invert. Kinetic interceptors—rockets that shoot down rockets—cost millions per engagement. Laser intercepts cost electricity, pennies per shot once fielded. When saturation attacks arrive, defenders with directed energy sustain engagement rates that deplete attacker inventory while preserving their own capacity.
The constraint isn’t weapon design. It’s manufacturing the components: integrated photonics consolidating laser components onto chips, and power electronics using gallium nitride handling electrical loads in compact form factors. Both require fabrication infrastructure that doesn’t exist at scale. Erez Berkovic, Rafael’s CTO, estimates multiple years to build production capacity even with the technology validated.
Edge AI and the Certification Trap
Autonomous systems must operate without continuous communication because adversaries jam frequencies. Edge AI—models running directly on platforms—eliminates dependency on external compute. But deploying AI to weapons creates a regulatory constraint Avital Schrift, IAI’s VP of Core Technologies, calls explainability. High-risk autonomous weapons require certification in Israel, the US, and Europe. Regulators demand proof algorithms do what they’re designed to do.
Most advanced AI algorithms are NP-complete—mathematically impossible to fully explain. A neural network with millions of parameters can’t articulate why it classifies specific objects as hostile. For commercial applications this is acceptable. For autonomous weapons making kill decisions, it’s disqualifying. IAI develops AI architectures trading performance for explainability—models whose decision pathways can be audited and certified.
The gap isn’t model performance. Commercial AI models exceed military accuracy requirements. The gap is achieving certifiable accuracy regulators approve for lethal systems. Companies solving this unlock certification competitors can’t match—a regulatory moat that matters more than algorithmic sophistication.
Missiles on Demand
The endpoint IAI’s Schrift envisions: push a button, receive an Arrow missile in hours. Push again, receive an Iron Dome interceptor from the same production line through software reconfiguration. Additive manufacturing combines composite materials, conductors, and insulators in single print runs. Two gaps remain: materials withstanding extreme conditions—rocket motors require materials surviving 3,000+ degrees Celsius—and interchangeable design where software shifts production between munition types without hardware reconfiguration.
This isn’t theoretical. Elbit implemented predictive maintenance at their ammunition facility using vision sensors, acoustic monitoring, and vibration analysis. The system detects anomalies minutes before failures—enough time to shut down safely and swap components, preventing cascading damage that would scrap hundreds of thousands of munitions. The technology has run operationally for a year, proving the concept outside laboratory conditions.
The Availability Equation
Gal Harari, MAFAT’s CTO, introduces the math that inverts defense procurement: Defense systems traditionally optimize for maximum performance—longest range, highest accuracy, greatest reliability under all conditions. This produces exquisite weapons that spend most of their time awaiting spare parts, stuck in maintenance cycles, or sitting idle because they’re too complex to deploy rapidly.
Harari’s equation exposes the strategic cost: multiply what the weapon can do (capability) by how often it’s actually available to use (availability) to get real-world effectiveness. A theoretically perfect weapon available 10% of the time delivers 10% effectiveness. A good-enough weapon available 80% of the time delivers 64% effectiveness—more than six times better operationally despite being “inferior” on paper.
This changes everything. Instead of designing the absolute best system and accepting low availability as inevitable, design systems good enough to win engagements but simple enough to keep operational and fast enough to manufacture at scale. The US lost seven $50 million MQ-9 drones to Houthi air defenses this year partly because each drone takes months to build—production can’t replace losses fast enough. An 80% solution you can build in weeks and keep flying continuously beats a 100% solution that takes months to produce and spends half its time grounded.
MAFAT’s Blue and White initiative implements this philosophy across the technology stack: photonic chips, infrared sensors, radar components, electronic warfare systems, new materials, production automation. Think of it as a Lego Army: building ecosystems where components from multiple vendors integrate without custom engineering and production scales when demand spikes.
Robotics at Scale
Israel has already proven the manufacturing model works. Colonel Yaron Sarig, Head of AI and Autonomy at MAFAT, presenting two decades of robotics development, demonstrated systems that transitioned from laboratory concepts to tens of thousands of operational units. Ground robotics, aerial drones with networking capabilities, and multi-environment deployment assets now operate as integrated networks across multiple domains.
This matters because it validates the thesis before the US defense industrial base has solved the same problem. Israel manufactures autonomous systems at scale while American defense primes are still debating requirements. The robotics production lines demonstrate that high-mix manufacturing—shifting between different platform types while maintaining volume—works in practice.
Production Tempo as Strategic Outcome
Russia-Ukraine proved production tempo determines outcomes when technologies converge. Both sides field similar capabilities. The differential is who manufactures faster. Ukraine received sophisticated Western systems performing well individually but couldn’t be produced quickly enough to affect strategic outcomes. Russia fielded less sophisticated systems in larger volumes, accepting higher attrition because production replaced losses faster than Ukraine destroyed inventory.
Heven Drones—Israel’s first defense unicorn—addresses this by building infrastructure customers depend on. Their hydrogen refueling stations started as internal necessity, became a separate business serving military customers. Their 300,000 square-foot facility outside Washington combines mass manufacturing with continuous R&D because technology evolves too quickly to separate production from development. Bentzion Levinson, Heven’s founder, describes the timeline: first US Army contract five years ago, unicorn valuation took six years. The customer’s question is always identical: technology validated, when can you deliver 1,000 units?
Military force equals quality times quantity to an exponent. Past the “good enough” threshold, volume overwhelms precision. The defense industrial base that can’t manufacture at scale didn’t lose a production competition. It lost the war before kinetic contact. Stockpiles built with electronics from years earlier arrive obsolete. Production lines optimized for steady-state output can’t surge when conflict demands exponential increases. Supply chains dependent on adversary-controlled ecosystems fail when routes close.
China has manufacturing velocity and ecosystem depth. The US has technological edge and capital. Israel has innovation pressure forcing faster iteration—and a two-decade head start manufacturing autonomous systems at volume. The question isn’t who builds the best weapon. It’s who builds the best weapons factory.