The Pentagon Is Buying a New Defense Stack

The defense industry runs on programs. Not software programs — procurement programs. The F-35 program. The DDG-51 program. A program is a relationship between a prime contractor and the government that spans decades, moves hundreds of billions, and develops institutional inertia that makes displacement nearly impossible. That model is under structural pressure from a different kind of defense company, and the contracts being signed in 2026 make the shift hard to miss.

In the span of three months this spring, three software-native defense startups closed funding rounds that would have seemed implausible five years ago. Anduril closed a $5 billion Series H at a $61 billion valuation. Shield AI is closing a $1.5 billion Series G at $12.7 billion. Saronic Technologies raised $1.75 billion at $9.25 billion. None of them is building a traditional platform. All of them are winning contracts that traditional primes weren't positioned to compete for.

The Architecture Shift

The clearest signal is Anduril's US Army contract — worth up to $20 billion over ten years, signed in March 2026. At its core is Lattice, an AI-enabled software suite that fuses data from drones, radars, satellites, and ground sensors into a unified operational picture. The Army awarded the contract not for a specific weapons system but for a software platform that can integrate future hardware additions without renegotiating procurement each time. The deal consolidates more than 120 separate procurement actions under a single enterprise agreement. That is the structure of a software subscription, not a traditional defense acquisition.

The distinction matters. Traditional defense contracts buy things: airframes, missile bodies, ship hulls. Lattice is buying a platform for decision-making — a continuous feed of sensor fusion, threat prioritization, and command support that compounds as more data flows through it. The Army is, in effect, purchasing a defense operating system.

Air, Sea, and the Autonomy Stack

The hardware platforms feeding into these software stacks are multiplying. Anduril's portfolio — Altius long-endurance UAVs, Ghost counter-drone systems, Menace interceptors, Dive autonomous underwater vehicles — routes telemetry into Lattice and executes decisions it surfaces. The hardware and software are designed as a system, not assembled from separate vendor catalogs.

Shield AI is building the equivalent autonomy stack for air. Hivemind, its autonomous pilot software, was selected in February 2026 for the US Air Force Collaborative Combat Aircraft program — the Pentagon's effort to pair AI-piloted drones with crewed fighters. Shield AI also unveiled X-BAT in October 2025: the first AI-piloted VTOL fighter with a 2,000 nautical mile range, capable of launching from ships, islands, or austere forward sites without ground infrastructure. The CCA program is a multi-decade bet that AI-piloted aircraft will constitute the numerical majority of the US tactical air fleet. If Hivemind is the software stack for that program, the addressable market is structurally very large.

At sea, Saronic launched its first Marauder medium unmanned surface vessel on May 31, 2026 — a 150-foot autonomous ship beginning on-water trials at its Franklin, Louisiana shipyard, with additional hulls already progressing through production. The Navy committed $392 million in an OTA to field Saronic's fleet at scale, targeting production of more than 20 vessels annually by 2027. That is a production pace competitive with crewed shipbuilding, for ships with no crews.

The Ukraine Validation

Before large programs were funded, the technology was validated in live conflict. Auterion's AuterionOS serves as an operating system for coordinated autonomous drone operations at scale. In 2025, Auterion won a $50 million Pentagon contract to deliver 33,000 AI-powered Skynode strike kits to Ukrainian forces — converting manually piloted drones into anti-jam autonomous guided munitions capable of tracking moving targets up to one kilometer away. At 33,000 units, this is production-scale autonomous weapons deployment in a live theater. Auterion subsequently raised a $130 million Series B and established a joint venture with Airlogix to produce autonomous strike drones for Ukraine and NATO allies, with deliveries starting in 2026.

Ukraine has functioned as a proving ground that no synthetic evaluation environment can replicate — live electronic warfare, GPS denial, coordinated countermeasures, contested airspace. Companies whose technology survives those conditions carry a credibility advantage in Pentagon procurement that is difficult to acquire any other way.

What the Primes Can't Match

The constraint for traditional primes isn't software capability — it's business model. A company optimized for 20-year programs, congressional relations management, and cost-plus contracting moves at the speed of government acquisition. It cannot iterate autonomy software weekly, cannot absorb field data from 33,000 deployed units and push updates within a sprint cycle, cannot staff an engineering culture that treats defense contracts as a software business.

What's emerging is a two-layer defense stack. Traditional primes retain ownership of the expensive-to-replace platforms — aircraft, submarines, nuclear systems. Software-native companies are building the intelligence layer that runs on top. Anduril's Lattice inside the US Army, Shield AI's Hivemind inside the Air Force CCA program — these are platform positions. The integration depth, classification barriers, and institutional switching costs that come with defense platform positions make them among the most defensible in any industry.

The Investment Angle

The Pentagon's FY2026 budget carved out a dedicated $13.4 billion line for autonomous and AI systems for the first time — covering aerial, ground, maritime, and underwater platforms. The Defense Autonomous Warfare Group's allocation is slated to reach $55 billion in FY2027. That is not a pilot program. It is a structural reallocation of defense spending toward the exact technology stack these companies are building.

The compounding logic mirrors what happened in enterprise software: the companies that own the platform capture leverage over every application, hardware vendor, and data source that connects to it. Defense platform positions, once embedded through classification requirements and deep integration, are rarely displaced on a short cycle. The investors who understood this in 2022 and 2023 are watching their thesis validate in 2026 at a valuation pace that is difficult to fault with hindsight. The question now is whether the next layer — the software companies building on top of Lattice and Hivemind, not beneath them — generates returns of the same magnitude.