SMAC Technologies raises $32M to build deep reinforcement learning models for military decision-making
Mar 2, 2026 with Andy Markoff
Key Points
- SMAC Technologies, founded by former Marine Raiders, raises $32 million including a $26 million Series A to build deep reinforcement learning models for military decision-making rather than relying on general-purpose LLMs.
- The startup deploys physics-grounded simulations trained by domain experts to encode tacit military knowledge directly into models, arguing LLMs fail for roughly 80% of military decisions where no labeled dataset exists.
- SMAC is already contracted with the Air Force, Marine Corps, and Navy, with plans to expand across all six services and deploy edge-deployable versions to frontline units by year-end.
Summary
SMAC Technologies is a 14-month-old defense AI startup founded by two former Marine Raiders who spent their careers running the kill chain and wanted to build the decision-making tools they never had. Andy Markoff and Clint Elenise started the company in January 2024. SMAC has raised $32 million to date, including a $26 million Series A co-led by Geodess and Costanoa Ventures, with early backing from 72. The team is 18 people split between El Segundo and Texas, where SMAC runs a physics lab.
The technical approach
SMAC is not building on top of general-purpose LLMs. Markoff argues that LLMs are the wrong tool for roughly 80% of military decision-making because they are trained on labeled datasets, and no labeled dataset exists for peer-level conflict. Instead, SMAC builds deep reinforcement learning models trained inside physics-grounded simulation environments built by physicists and domain experts. Most military expertise lives in people's heads, not in documents, so SMAC encodes that tacit knowledge directly into the training environment rather than trying to scrape it from text.
The product operates across multiple time horizons: strategic planning over one to six months, operational planning over one to four days, and real-time tactical decisions. It converts petabytes of multimodal sensor data into actionable outputs. SMAC is also building edge-deployable versions for frontline units with bandwidth constraints, alongside heavier versions for command centers.
Human control
Markoff is explicit that fully automating the kill chain is not the goal. The frame is "intelligent autonomy" — removing humans from low-value decision points while keeping them in the loop for decisions that require ethical or tactical judgment. The practical challenge is orchestrating potentially hundreds of thousands of autonomous systems, manned platforms, and unmanned platforms across a theater that could span 100 million square miles.
Current deployment and 2025 priorities
SMAC is already deployed under contract with the Air Force, Marine Corps, and Navy. This year's priority is expanding across all six services at the command level and deploying edge versions to frontline units by year-end.
The company's name comes from a tactical radio term: "smack" is the call used to strike a target. The initial go-to-market was in the fires functional area, which was Elenise's specialty, before expanding into the broader range of military decision-making.