The Meta investment and customer shifts
Scale AI reached a $29 billion valuation in June 2025 after Meta invested $14.3 billion for a 49% stake. In June 2025, Meta’s investment of $14.3 billion for a 49% stake in Scale AI led Alexandr Wang to leave his CEO role to become Meta’s Chief AI Officer. Jason Droege, a former Uber Eats executive, took over as interim CEO. This deal changed how the market views Scale AI’s neutrality. Google planned to end its relationship as the largest customer, while OpenAI began phasing out work with the company. I find the loss of neutrality after the Meta investment to be a massive strategic error for Scale AI’s expansion, especially since revenue reached an annualized run rate of $1.5 billion by the end of 2024 following a jump from $290 million in 2022 to $760 million in 2023. Scale AI currently employs 300 people. Alexandr Wang, a tech lead at Quora, and Lucy Guo, a product designer, founded the company in San Francisco to provide data labeling services. They moved from manual appointment scheduling to providing data for industries such as automotive and defense. Scale AI secured an initial investment from YCombinator and later raised $100 million from Founders Fund in 2019. The company works with customers like PayPal, Pinterest, Toyota, and General Motors. Early clients included Houzz, HigherMe, and RealTalk. This growth was supported by Accel VC, which noted that Alexandr Wang understood that progress in AI would not hinge on technical limitations but on data availability.
Tesla and the Dojo supercomputer
Tesla is shifting focus away from its Dojo supercomputer to prioritize inference chips like AI5 and AI6. Elon Musk ordered the closure of the Dojo team, leading 20 workers to join DensityAI. The Dojo supercomputer uses 25 D1 chips per training tile and 120 tiles to reach 1.1 exaFLOPS of power. Each training tile provides 9 PFLOPS of computing power. This hardware was built to process video from Tesla vehicles to train autonomous software. I would ignore the claim that Tesla can effectively scale its AI without the custom training architecture Dojo provided. The D1 chip provides 362 teraFLOPS of processing power per unit. Morgan Stanley analysts valued the Dojo supercomputer at $500 billion in 2023. Tesla unveiled its first custom AI chip in 2019. Musk announced a $16.5 billion deal to source AI6 chips from Samsung Electronics. These next-generation AI5 chips should enter production by the end of 2026.
Data dominance and government contracts
Scale AI maintains a workforce of over 100,000 contractors to perform data labeling. The company launched its Physical AI data collection platform in 2025 to serve robotics companies like Physical Intelligence and Cobot. This platform collects demonstration data from robotics hardware. Scale AI also handles government contracts, including a $250 million agreement with the U.S. Department of Defense in 2023. In March 2025, the Defense Innovation Unit awarded Scale AI a prototype contract for Thunderforge. In May 2026, the Pentagon’s Chief Digital and Artificial Intelligence Office expanded its agreement with Scale AI from $100 million to $500 million. You already know that training data dictates model performance, so watch how Scale AI navigates this Meta-induced isolation. Scale AI provides the data layer for industries ranging from finance to logistics to government, using its Scale Data Engine to handle images, video, 3D sensor data, text, and audio for a global clientele. Scale Donovan handles intelligence reports, emails, and satellite imagery to support military operations. The GenAI Platform allows organizations to customize models with proprietary data. Scale Evaluation provides tools for assessing AI systems through reinforcement learning from human feedback and safety testing. The company earned $870 million in revenue in 2024. Scale has annotated more than 200,000 miles of data collected by self-driving cars. The Nucleus software allows customers to search through data to find mislabeled examples that degrade algorithm performance. Scale provides specialized labeling like Semantic Segmentation, Image Annotation, and Sensor Fusion. Will Scale AI maintain its independence as Meta’s influence grows?
