The $29 Billion Data Infrastructure Pivot and Federal Safety Shifts

Meta leadership changes and industry friction

Meta invested $14.3 billion for a 49 percent stake in Scale AI in June 2025. This deal set the company’s valuation at $29 billion. The $14.3 billion investment from Meta granted the company a 49 percent stake and placed the valuation of Scale AI at approximately $29 billion, which caused major customers like Google and OpenAI to reconsider their long-term partnerships. This massive deal prompted founder Alexandr Wang to leave the CEO position to work on Meta’s superintelligence efforts. Jason Droege, formerly the chief strategy officer and an executive at Uber Eats, took over as interim CEO. You should monitor if these competitor departures cripple Scale’s neutral position in the market. The Meta investment complicates Scale’s role as a neutral data partner. Scale AI previously raised $1 billion in a series F round led by Accel with participation from Nvidia, AMD, Amazon, Cisco, Intel, and ServiceNow. This funding brings the total capital raised to $1.6 billion and supports the company’s goal to build a "data foundry for AI." Scale AI worked to improve the performance of OpenAI’s GPT-3.5 and helped Meta during the development of the Llama 2 model. The company also expanded its operations into Europe by opening an office in London.

Alexandr Wang and Lucy Guo met at Quora before starting a mobile app in 2015 to book doctor appointments. They identified a need for a human API to handle manual tasks like appointment scheduling and content moderation. Scale AI launched in 2016 to provide developers with an API for on-demand task completion. The company provides the data layer behind AI, including data labeling, expert human feedback, and model evaluation. The company compensates its workforce, known as Scalers, with an hourly rate to encourage thoroughness. This model differs from crowdsourced platforms like Amazon’s Mechanical Turk, which often suffer from low quality and spam. Scale AI’s success also relies on its ability to notice market shifts early and move its product line accordingly.

Defense contracts and the competition for geospatial intelligence

The Pentagon’s Chief Digital and Artificial Intelligence Office expanded its enterprise agreement with Scale AI from $100 million to $500 million in May 2026. Scale AI also holds separate agreements with the U.S. Army and the Pentagon worth up to $199 million combined. However, a startup called Enabled Intelligence secured a $708 million single-award agreement from the National Geospatial-Intelligence Agency in November. Scale AI lost this Sequoia contract despite its history as a primary vendor for Project Maven. Enabled Intelligence uses a neurodiverse workforce where 60 percent of employees are on the autism spectrum to perform pattern recognition. This startup partners with BAE Systems, Vantor, and Whitehead Federal to support flagship programs applying AI to intelligence, surveillance, and reconnaissance sensors. The contract covers automated object detection, object tracking, and pattern detection in satellite imagery of aircraft, armored vehicles, and personnel. Scale AI maintains its presence in the defense sector through other programs, such as the Defense Innovation Unit’s Thunderforge prototype contract awarded in March 2025.

Red-teaming performance and federal safety oversight

Scale AI dismissed 12 contractors from its "Red Team" because of performance issues and caused a disruption in its safety testing workflows. These removals occurred alongside reports that workloads for some affected workers slowed down. Federal oversight of AI models moves away from voluntary participation. The Center for AI Standards and Innovation signed formal pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI to assess the security and capabilities of new models before they become available to the public. These agreements follow a trend where specific capability assessments, like those involving Anthropic’s Mythos model, trigger broader government oversight. The Department of Commerce intends to assess the security and capabilities of these models before they become publicly accessible. This regulatory interest grew after Anthropic’s CEO met with Trump administration officials regarding concerns about the Mythos model’s power. Will the government successfully move from voluntary memoranda to mandatory enforcement? Scale AI continues to build its "data foundry for AI" using funds from a $1 billion series F round that included investors like Nvidia, AMD, and Amazon. Alexandr Wang suggests that AI models will soon produce all the code he has ever written and emphasizes a shift toward "vibe coding" where users type instructions to generate software.