XDOF, an emerging artificial intelligence startup focused on generating real-world teleoperation datasets for general-purpose robotic systems, is reportedly finalizing terms for a new Series B funding round. The prospective financing round is expected to assign the company a post-money valuation of roughly $1.2 billion, according to individuals familiar with the negotiations. The fundraising process is being steered by venture capital firm 8VC, marking an exceptionally rapid escalation in market value for an enterprise that stepped out of stealth mode less than three months ago.
The venture round represents a remarkably swift return to capital markets for XDOF, which concluded a notable $70 million Series A investment round in June. That earlier financing garnered backing from prominent Silicon Valley venture entities, including Thrive Capital, Andreessen Horowitz, Lux Capital, and Spark Capital. Although the executive leadership team had not intended to solicit outside equity so soon following the June transaction, aggressive inbound interest from institutional venture investors emerged after the startup demonstrated significant commercial momentum, pushing its annualized revenue run rate toward the $50 million mark.
Academic Roots and Commercial Architecture
Founded in 2024 by University of California, Berkeley researchers Philipp Wu, who serves as chief executive officer, and Fred Shentu, the company's chief technology officer, XDOF emerged directly out of academic bottlenecks encountered while studying robotic learning mechanisms. While pursuing doctoral research on how autonomous machines assimilate diverse behaviors, Wu recognized that the fundamental barrier confronting physical artificial intelligence was the sheer scarcity of high-volume, structured physical datasets required to train advanced machine learning architectures.
To resolve these empirical constraints, Wu and Shentu engineered GELLO, a cost-effective teleoperation platform designed to allow human technicians to remotely guide robotic manipulators. The intuitive hardware-software framework allowed operators to orchestrate mechanical arms with precision, capturing fine-grained kinetic signals and behavioral demonstrations. This breakthrough formed the basis of an influential robotics paper and established the underlying methodology for XDOF, creating a viable commercial mechanism for producing industrial-scale motion and manipulation records for external clients.
Solving the Data Bottleneck in Physical AI
Within venture capital circles, XDOF is frequently characterized as the robotics ecosystem's counterpart to data infrastructure heavyweights such as Scale AI and Mercor. While early foundation models and large language models benefited from mining decades of digital text and media across the public internet, autonomous physical machines lack a comparable corpus of real-world operational demonstrations. Consequently, the lack of curated kinetic interaction data has surfaced as the primary systemic obstacle hindering the deployment of versatile, multi-task robotic systems in commercial settings.
To bridge this critical industry gap, XDOF functions as an integrated, outsourced data-supply architecture for leading artificial intelligence research labs and hardware manufacturers. The company designs proprietary ingestion pipelines, custom collection instrumentation, and precise annotation frameworks that client organizations often lack the logistical capacity to construct independently. By taking on the laborious process of recording physical interactions, the enterprise equips frontier developers with the structured inputs necessary to train next-generation spatial models.
Global Operations and Broadened Data Infrastructure
The operational framework at XDOF relies on a dual-pronged data harvesting methodology that blends robotic teleoperation with wearable human tracking systems. Technicians wearing specialized body sensors record human biomechanics while completing standard domestic and industrial actions, such as folding fabrics or flattening cardboard packaging. Concurrently, remote operators maneuver mechanical limbs across varied physical environments to generate rich control telemetry, building a multi-modal repository of task execution across diverse everyday scenarios.
In a strategic initiative to advance academic benchmarks alongside commercial deliverables, XDOF has aligned with the UC Berkeley Artificial Intelligence Research laboratory to publish a collaborative dataset known as ABC. Designed to stand as the most expansive repository of premium robotic instruction records to date, the initiative reflects the startup's broader ambitions. To maintain this data influx, XDOF is scaling an international labor pool of specialized teleoperators and egocentric data capture personnel worldwide.
Competitive Dynamics in the Robotics Data Race
XDOF has already integrated its data streams into the development roadmaps of approximately 20 enterprise clients, an initial customer base that encompasses several prominent frontier AI developers. However, the commercial landscape for embodied intelligence infrastructure is growing increasingly contested. Competitors such as Mecka AI are actively targeting similar capture methodologies, while established text-and-image data labeling providers including Scale AI and Micro1 are rapidly expanding their operational toolkits to capture physical and embodied data streams.
While discussions surrounding the Series B funding remain ongoing, specific terms including the ultimate gross capital injection and structural clauses have not been finalized and remain subject to market conditions. Neither leadership at XDOF nor representatives from 8VC provided formal public comments regarding the pending transaction. Nevertheless, the valuation trajectory underlines robust investor appetite for the fundamental picks-and-shovels infrastructure powering physical AI and general-purpose automation.
