Scale AI

Scale AI is a San Francisco company founded in 2016 by Alexandr Wang and Lucy Guo, built first on the unglamorous but essential work of labelling the vast datasets that machine-learning models train on. That foundation turned it into one of the most-used data pipelines in commercial AI, and in June 2025 Meta took a roughly 49% non-voting stake valuing the company above $29 billion, with Wang leaving to lead Meta’s artificial-intelligence effort. Jason Droege, formerly the company’s chief strategy officer, took over running Scale.

Its defence arm, marketed as Scale Federal, sells applied AI rather than sensors or hardware. The flagship offering is Donovan, launched in 2023 as what Scale described as the first large-language-model deployment onto a US government classified network, used by the Army’s XVIII Airborne Corps. Donovan ingests thousands of orders, situation reports, and intelligence documents so that planners can query them in plain language and draft courses of action in minutes. In late 2024 the company added Defense Llama, a national-security-tuned version of Meta’s Llama 3 model that runs exclusively inside Donovan’s controlled environments.

Scale has moved steadily deeper into Pentagon work. In 2025 the Defense Innovation Unit selected it to build Thunderforge, an agentic-AI prototype for military planning and wargaming aimed at Indo-Pacific and European commands, alongside Anduril and Microsoft, and it signed a Chief Digital and Artificial Intelligence Office agreement later expanded toward a far larger decision-support contract. Where Palantir has anchored the Pentagon’s Maven Smart System around fusing sensor and video feeds for targeting, Scale’s pitch is text-and-language-centric: turning the written record of a campaign into something a commander can interrogate and act on.