Defense autonomy and AI/ML mission systems technology and investment research

Technical bottleneck

• Bottleneck type: Safety certification / Airworthiness / ITAR
• Technical constraint: Autonomous engagement authority (the decision to use lethal force) must operate under human-on-the-loop doctrine — the autonomy stack must recommend, the human must authorize; explainable AI for target identification requires saliency maps and uncertainty quantification — 'why did the system classify this as a T-72?' is an operational requirement, not just a research goal; multi-platform collaborative autonomy (Loyal Wingman, MUM-T) requires distributed PNT (positioning, navigation, timing) and communication under jamming — graceful degradation when connectivity is lost
• Economic constraint: Defense prime moat — Palantir (Maven Smart System, AIP), Anduril (Lattice OS — private), Shield AI (Hivemind — private), and Kratos (XQ-58 Valkyrie autonomy) are the key software players; government ITAR controls and security clearances create high barriers; cost-plus and fixed-price development contracts provide multi-year revenue visibility

Adoption

• Driver: DoD Replicator 2.0 (autonomous systems at scale >1,000 units); Collaborative Combat Aircraft (CCA) program requiring autonomous wingman software for NGAD and F/A-XX; Ukraine conflict accelerating autonomous system deployment and software iteration; AUKUS Pillar 2 (AI and autonomy collaboration)
• Blocker: Autonomous lethal engagement policy and international law (LAWS/CCW debates) constraining deployment; defense acquisition timelines (5-10 years for program of record) mismatched with software development speed (weekly sprints); defense prime incumbents (Lockheed, Raytheon, Northrop) prefer hardware programs with cost-plus margins over software

Public companies exposed

PLTR (Palantir — Maven Smart System

AIP for defense autonomy

TITAN ground station)

BBAI (BigBear.ai — AskSage / ConductorOS secure mission AI)

KTOS (Kratos — XQ-58 Valkyrie + autonomy/EW software)

LMT (Lockheed — F-35 autonomy

loyal wingman

AI for Aegis)

RTX (Raytheon — C-UAS AI

LTAMDS sensor fusion)

LHX (L3Harris — autonomous surface vessels

RASOR software)

LDOS (Leidos — autonomous maritime

Sea Hunter follow-on)

AVAV (AeroVironment — Switchblade autonomy

Blackwing ISR)

Validation signals

Palantir USG revenue growth from defense autonomy programs; CCA Phase 2 software award (autonomy stack prime); DoD autonomy software contract awards shifting from hardware primes to software-first companies (Anduril, Shield AI); autonomous engagement authority policy framework completion; AskSage / ConductorOS deployments in classified and air-gapped environments show accredited mission AI is already monetizing

Invalidation signals

Autonomous lethal engagement policy moratorium constraining deployment; defense prime incumbents maintaining software control through platform ownership; software-first defense companies failing to scale past $1B revenue

Research Update — 2026-07-18

_Source: SEC 10-K / 10-Q / 8-K and stockanalysis.com during BBAI deep-dive_

Technical readiness: deployed / scaling in classified and air-gapped mission environments

Key papers / sources:

• https://www.sec.gov/Archives/edgar/data/1836981/000183698126000018/bbai-20251231.htm — BigBear.ai describes itself as a mission-ready AI provider for national security, supply chain management, and digital identity, and says Ask Sage is FedRAMP-authorized and certified to handle government data up to DoD IL5.
• https://www.sec.gov/Archives/edgar/data/1836981/000183698126000047/earningsrelease-1q26.htm — Q1 2026 release shows the platform is not just a lab demo: backlog increased 14% QoQ to $281.9M, driven by a $53M sole-source prime classified award, while management reaffirmed FY2026 revenue guidance.
• https://stockanalysis.com/stocks/bbai/financials/ — FY2025 annual revenue was $127.67M, gross margin 22.31%, operating margin -167.54%; the company is still loss-making, but the margin profile shows software/platform mix is real.

Bottleneck update: strengthened — the bottleneck is secure, accredited deployment into government environments, not raw model capability. FedRAMP, DoD IL5, and classified procurement create the moat.

Alternative/substitute risk: larger defense software incumbents (PLTR, Booz Allen, Leidos) can compete on procurement relationships; private autonomy stacks (Anduril, Shield AI) can compete on mission software breadth.

Timeline signal: this is a deployment and procurement scaling problem, not a science problem; the 2-5 year window is dominated by accreditation, backlog conversion, and program adoption.

Adoption rate data: Q1 2026 backlog of $281.9M (+14% QoQ), $53M sole-source classified award, and FY2026 revenue guidance of $135M-$165M are the clearest current adoption signals.

Thesis impact: BigBear.ai now fits the defense autonomy / secure mission AI layer directly. The evidence argues for a partial promotion of the technology note from seed to partial and for BBAI to be tracked as a public-company exposure in the defense autonomy note.

Research Update — 2026-07-26

_Source: SEC 10-K / 10-Q / 8-K during KTOS backfill_

Technical readiness: deployed / scaling in classified and air-gapped mission environments; the issue is not whether autonomy works in a demo, but whether the stack can survive procurement, safety, and production constraints.

Key papers / sources:

• https://www.sec.gov/Archives/edgar/data/1069258/000106925826000013/ktos-20251228.htm — 10-K describes Kratos’ unmanned aerial, unmanned ground, unmanned seaborne, and mission-system businesses, and explicitly notes relocation/expansion of small turbojet engine production capacity in Michigan plus a planned small turbofan facility in Oklahoma.
• https://www.sec.gov/Archives/edgar/data/1069258/000106925826000051/ktos202603298kexhibit991.htm — Q1 2026 EX-99.1 says 2H26 revenue should be significantly higher because long-lead items tied to jet engine and propulsion system development/production plus air-defense, missile, radar, and other National Security hardware will begin shipping.
• https://data.sec.gov/api/xbrl/companyfacts/CIK0001069258.json — FY2025 revenue $1.3468B, gross profit $307.9M, operating income $25.6M, enough to show this is already a real production business rather than a concept.

Bottleneck update: strengthened — the bottleneck is qualification + production throughput, not model quality. Defense autonomy is gated by procurement trust, program qualification, and domestic propulsion capacity.

Public parameter table:

• FY2025 revenue: $1.3468B
• FY2025 gross margin: 22.9%
• FY2025 operating margin: 1.9%
• Q1 2026 revenue: $371.0M
• Q1 2026 Unmanned Systems revenue: $82.6M
• Q1 2026 KUS backlog: $375.4M
• FY2026 revenue guidance: $1.700B-$1.760B

Worked calculation:

• Q1 annualized revenue = $1.484B.
• FY2026 midpoint guide = $1.730B.
• Step-up required vs Q1 run-rate = $246M (+16.6%).
• KUS backlog coverage at current run-rate = 4.54 quarters (~13.6 months).

Alternative/substitute risk: software-first defense autonomy stacks can compete on mission software, but they do not automatically solve the propulsion/manufacturing bottleneck. The more serious substitute risk is a prime contractor integrating autonomy internally and capturing the platform economics.

Timeline signal: this is a 2026-2027 execution story. The near-term catalyst is whether long-lead propulsion and National Security hardware actually convert into H2 revenue and larger program-of-record awards.

Adoption rate data: Q1 2026 Unmanned Systems revenue was $82.6M, or 22.3% of total revenue, and that segment carried a 1.2x quarter book-to-bill / 1.1x TTM book-to-bill with $375.4M backlog.

Thesis impact: Kratos remains one of the few public defense-autonomy names with both a flying platform and in-house propulsion. The moat is real, but the math says the moat is still execution-sensitive: if book-to-bill slips below 1.0x and backlog stops covering roughly a year of demand, the thesis becomes timing-dependent rather than structurally compounding.