Scale AI maintains a $29 billion valuation following Meta’s $14.3 billion investment in June 2025. This deal involved Meta purchasing 49 percent of the company. Alexandr Wang left Scale AI to lead Meta’s Superintelligence Labs. Scale AI lost Google as a customer after this investment. The company now prioritizes government and defense contracts through Scale Federal. Scale AI generated $870 million in revenue last year and expects sales to reach $2 billion in 2025. High-quality labeled data remains the bottleneck for AI development. Without accurate training data, AI systems fail to identify objects or understand language. You already know that training data quality dictates model performance. The company organizes workflows to tag images, text, audio, and video for automotive firms and defense agencies. Scale AI also provides AI evaluation and model testing to help organizations measure how well large language models perform. The company competes with Surge, Turing, and Invisible in the evaluation market.
The AI strategy and algorithmic warfare
The Department of War issued an AI strategy in January 2026 to create an AI-first warfighting force. This strategy emphasizes speed and scale through seven Pace-Setting Projects. One project aims to turn intelligence into weapons in hours, not years. JADC2 provides the digital nervous system to fuse sensors and shooters into a coherent warfighting organism. In January 2026, Replicator 2 acquired the DroneHunter F700 counter-drone system. The Drone Dominance Program held Gauntlet I evaluations in early 2026. The Pentagon wants to equip units with 300,000 to 340,000 drones by 2027. The Gauntlet I results in March 2026 placed Skycutter at the top with a score of 99.3. Neros followed with 87.5, while Napatree Technology earned 80.3.
| Label Type | Application | Geometry |
|---|---|---|
| Bounding Box | Autonomous driving | Rectangular |
| Cuboid | Robotics/3D planning | 3D (X, Y, Z) |
| Polygons | Irregular objects | Vertices |
| Points | Pose estimation | Spatial locations |
| Lines | Roadway markers | Vertices |
Data quality requires accuracy, completeness, consistency, relevance, timeliness, and representativeness. The Department of Defense uses the Advana platform to pull data from 3,000 different business systems. Booz Allen Hamilton won a $647 million contract to grow this program. Advana helps the military track readiness data during exercises. The Department’s May 2021 memorandum mandates that components establish federated data catalogs with a 30 day time window for delivery. Labelers use polygons for irregular objects like buildings or trees. Bounding boxes are faster but leave gaps around objects.
Subsea autonomy and the data bottleneck
The US Navy requires autonomous underwater vehicles to provide cheap undersea mass and maintain maritime dominance against an adversary that continues to expand its own uncrewed undersea fleet through massive, parallel design pipelines. China launched ten nuclear-powered submarines between 2021 and 2025 and unveiled two classes of XLUUVs in September 2025. AUKUS Pillar II started in May 2026 to develop AUV payloads. The Director of Submarine Programs announced interest in advanced payload launchers for future attack submarines on July 29.
The US military needs expert networks to supply high-difficulty human data. These networks include PhDs, doctors, and military specialists. Effective data labeling requires users to capture corners and edges accurately. The Senate Armed Services Committee advanced legislation in 2026 to codify ultimate human responsibility in the AI-powered kill chain. The US Coast Guard also seeks technology to detect and defeat uncrewed underwater vehicles. How will the Pentagon manage the tension between human responsibility and autonomous machine execution in contested waters? The US Coast Guard continues to seek technology to defend domestic littoral waters against uncrewed underwater vehicles.
