Scale AI secured a multimillion-dollar contract with the US Department of Defense to build AI tools for Project Thunderforge. The Defense Innovation Unit leads this program, which develops technology for decision-making, simulation, and operational planning. Scale AI works alongside partners Microsoft and Anduril to deliver these capabilities. Scale AI also provides domain-expert labels, GenAI data, and RLHF for frontier models. The company focuses on data engine positioning for global enterprise delivery, including multimodal annotation and autonomous driving work. The Defense AI market is moving toward large-scale deployments, with 60 financings across 53 companies in the last 12 months, up from 50 deals across 42 companies previously. While total capital reached $14.55 billion, 96% of that money went to rounds above $50 million. Autonomous Systems Software captured 88.8% of the capital, even though it only represented 46.7% of deals. The median financing for autonomous software tripled to $45.5 million. I find Scale AI’s ability to provide specialized labels for frontier models to be its strongest advantage in the current market. The median round size for the entire Defense AI sector more than doubled from $12.85 million to $29 million. First financings increased from 17 to 19, and strategic investors participated in 20% of current financings. The largest individual financing in the current period accounted for 34.4% of capital. European companies represented 35% of deals but 20.3% of capital. The median round-to-round step-up increased to 4.08x.
The Anduril Lattice competition
The US Army selected Anduril’s computer vision-based Lattice platform for counter-UAS missions. This platform functions as a command and control AI, complementing systems like Ghost drones and Pulsar electronic warfare. Scale AI and Anduril both participate in the Thunderforge program, which initially focuses on the Indo-Pacific and European Commands. This creates a direct competition for Pentagon influence. The US defense budget reached $886 billion in FY2026, and the Department of Defense aims to field AI systems ten times faster than traditional contractors can. Anduril, founded by Palmer Luckey, holds a $28 billion valuation and builds hardware like the Ghost drone and electronic warfare tools. They hold major contracts with DARPA and SOCOM. I would point out that Anduril’s Lattice OS remains a massive hurdle for other autonomous system providers. The expansion of the market is driven by increasing demand for tools that allow for rapid, secure data transfer and real-time terrain mapping. Does the collaboration between Scale AI and Anduril in Thunderforge mean they have ended their rivalry for Pentagon dominance? Anduril’s systems provide capability for autonomous flight.
| Feature | Scale AI | Anduril |
|---|---|---|
| Primary Focus | Domain-expert labels and RLHF | Autonomous defense systems |
| Key Platform | Data Engine | Lattice OS |
| Target Use | Frontier-model infrastructure | Command and control AI |
| Valuation/Scale | Global enterprise delivery | $28 billion valuation |
Requirements for agentic warfare
Agentic AI in 2026 moves from pilots to production in environments requiring orchestration, guardrails, and monitoring. Systems must be recoverable and auditable to prevent issues like goal hijacking or privilege escalation. The threat of data poisoning in training models makes high-quality, verified data a necessity for mission-critical workflows. Scale AI provides evaluation and safety data needed to build resilient models. I think the move toward domain-specific language models helps mitigate the risk of rogue agents. You must prioritize platforms that offer decision traces and long-horizon reasoning capabilities. The digital battlefield market grows as organizations demand better decision support and real-time decision-making. In the current environment, 70% of organizations cite model risk management as a priority for AI deployment. Organizations also face identity threats from AI-generated replicas of leaders. The browser is becoming the primary autonomous interface for the enterprise. AI agents triage alerts and build financial models. The time to evaluate a tool often exceeds its window of relevance.
