NestAI, founded by Peter Sarlin (ex-Silo/AMD), has released its first production models optimized for autonomous military systems, specifically for drone operations and battlefield orchestration. The models are trained on synthetic and real-world data to enable unmanned systems to operate on the edge, with foundational models for drone autonomy and higher-level orchestration capabilities for coordinating entire missions. The company is currently running pilot programs with the Estonian and Finnish armed forces, allowing them to conduct end-to-end drone missions and plan multi-vehicle operations. NestAI was founded in 2025 and grew to 200 people in one year, having raised €100m in November 2025 from Nokia and Finnish state-owned Tesi.
The release comes weeks after the U.S. government suspended Anthropic model exports, renewing focus on European military independence. Sarlin emphasizes domain specificity over general-purpose frontier models, positioning NestAI as solving the edge deployment and real-time adaptation challenges unique to military operations. The company is partnering with AMD for compute capacity, LUMI AI (Finland's supercomputer) for training, and quantum startup Qutwo (Sarlin is co-founder) for model compression via quantum simulation on GPU clusters.
For practitioners: NestAI signals institutional (state-backed) investment in edge-deployable AI for defense, acknowledging that 4-year-old Ukraine drone data shows Russian attacks increasing from 13,300 (2024) to 56,700 (2025). Architects designing autonomous systems for military or critical infrastructure should expect: (1) domain-specific model requirements to replace off-the-shelf LLMs, (2) real-time adaptation capabilities as primary design constraint, (3) state actor involvement in edge-compute and training infrastructure as a structural shift in vendor selection.