Scale AI

Scale AI builds the data, evaluation and application layers used to train, test and deploy artificial intelligence systems. Its original business organized large human workforces and software pipelines to create high-quality training data. It has since moved up the stack into model evaluation, post-training, agentic applications and government mission software. For national-security customers, Scale combines data operations with Donovan and other application tools that can operate in classified environments. The strategic shift matters because Scale is no longer competing only for labeling work: it is now seeking production AI workloads inside the Department of Defense, including planning, decision support and computer-vision programs.

Data Engine & Physical AI

Scale Data Engine supports data collection, curation, annotation, training and evaluation across text, images, video and sensor data. The company also markets Physical AI Data Engine for robotics and autonomy. Scale said it delivered more than 150,000 hours of robotics data in 2025 and added 10 robotics customers.

Scale Labs, SEAL & Model Evaluation

Scale Labs combines the company’s evaluation, post-training and research work. Its Safety, Evaluations and Alignment Lab (SEAL) publishes model benchmarks and expert evaluations. Scale said SEAL introduced 15 new frontier leaderboards and published more than 450 evaluations across more than 50 model releases in 2025.

Donovan & Government Applications

Donovan is Scale’s government-focused AI application layer for model-agnostic agents, search, analysis and decision support. Scale says its national-security stack can be deployed from unclassified networks through Top Secret environments. The company’s government work now includes the Pentagon-wide CDAO production OTA, Thunderforge operational planning, Army AI research and National Geospatial-Intelligence Agency data services.

Scale does not publish audited revenue, margins, cash or backlog. In January 2026, however, the company said 2025 was its strongest financial year and generated well over $1 billion in new business across data and applications, with nearly half of new bookings arriving in the fourth quarter. Scale also said its applications business more than doubled revenue in the second half of 2025 and that it expected the applications business to roughly double again in 2026. Its international public-sector business doubled revenue in 2025 and was projected by the company to double again in 2026. Defense growth is increasingly tied to vehicle conversion rather than headline ceilings: the CDAO production OTA can support up to $500 million of work, but the ceiling is not funded backlog and Scale does not disclose cumulative obligations. The ICG Solutions acquisition adds real-time streaming-data analytics to the national-security stack. No acquisition price or revenue contribution has been disclosed.

The most important 2026-2027 indicators are how quickly DoD components issue project agreements under the CDAO production OTA, whether Thunderforge expands beyond its initial combatant-command deployments and whether Scale turns the ICG acquisition into an integrated intelligence workflow rather than a stand-alone capability. Scale’s own operating targets point to continued expansion: it expects its applications business to roughly double in 2026 and projects its international public-sector business to double. Those are company forecasts, not guidance backed by public financial statements. A second watch item is customer mix. Government and enterprise growth can diversify Scale away from a small number of frontier-model customers, but the company does not disclose revenue concentration. Defense Briefing found no company-announced initial public offering timetable or public securities filing for Scale AI as of September 5, 2026. The 2025 Meta transaction provided substantial shareholder liquidity, reducing the need to infer a near-term public-market path.

Scale’s most defensible advantage is the combination of a large data-production network, model-evaluation capability and government deployment software inside one vendor. That combination lowers handoff costs between training data, evaluation and production applications. In defense, the CDAO production OTA and prior classified deployment work create procurement access and switching friction that a new entrant must reproduce. The moat is not absolute. Model labs can bring data work in-house, federal systems integrators can bundle AI services around other models and software vendors such as Palantir already have entrenched government footprints. The strongest test of durability will be whether Scale converts its contracting vehicles into recurring, funded mission workloads while preserving customer trust across competing model providers.

Customer neutrality is the clearest commercial risk. Meta became a large minority shareholder in June 2025 while Scale’s founder moved to Meta, creating concern among rival AI labs over sharing sensitive training priorities with Scale. Scale says it remains independent and that Meta has no access to customer confidential information. Reuters reported that Google, previously Scale’s largest customer, planned to cut ties and that other model developers were reevaluating work. Defense contracting creates a different risk: ceiling values can look like backlog even when funds are not obligated. Scale does not publish funded backlog, so the $500 million CDAO vehicle should not be modeled as booked revenue. Execution risk also sits in Thunderforge expansion, classified accreditation, integration of ICG Solutions and the shift from project services toward repeatable application software. Private-company disclosure remains limited, leaving investors, customers and procurement analysts without audited revenue, margin, cash, ownership percentages or a complete board roster.

Scale AI was founded in 2016 and first built infrastructure for producing labeled training data for machine-learning systems, with early demand from autonomous-vehicle developers. The company expanded from annotation into higher-complexity human feedback, model evaluation and generative-AI data work as foundation models grew more capable. Scale entered U.S. defense contracting at material scale in 2020, when the Army awarded a $90.87 million research-and-development contract for annotated AI and machine-learning datasets, with $16.47 million obligated at award. It subsequently expanded into government application software and operational planning, including Thunderforge and the CDAO production OTA. The company closed a $1 billion Series F financing in May 2024 at a $13.8 billion valuation. In June 2025, Meta made a large strategic minority investment that valued Scale at more than $29 billion. Founder Alexandr Wang moved to Meta while remaining on Scale’s board, and Scale appointed Jason Droege interim CEO. The transaction provided shareholder liquidity and materially changed Scale’s governance and customer-risk profile. In 2026, Scale added ICG Solutions to its national-security portfolio and appointed Francis deSouza CEO. The company now presents itself as a two-business platform: data and evaluation infrastructure for advanced AI models, plus applications for enterprises and governments.