By Marina Temkin | TechCrunch
Less than three months after emerging from stealth with a high-profile splash, XDOF—a specialized startup dedicated to capturing real-world teleoperation data for training general-purpose robots—is already barreling toward unicorn status. According to multiple people with direct knowledge of the negotiations, the company is locked in late-stage talks to raise a Series B funding round at a staggering valuation of approximately $1.2 billion.
The round is reportedly being led by prominent venture capital firm 8VC.
The breathtaking speed of XDOF’s ascent highlights a profound shift in the artificial intelligence landscape. While software-based large language models (LLMs) feasted on the collective text, code, and media of the internet to achieve general intelligence, physical robots face a severe supply-chain bottleneck. They cannot browse the web to learn how to manipulate the messy, unpredictable physical world. They need ground-truth physical data—and they need it at scale.
XDOF has positioned itself as the definitive outsourced data pipeline for this emerging industrial revolution, earning comparisons among venture capitalists to data-labeling titans like Scale AI and Mercor.
Main Facts
- The Valuation Surge: XDOF is in late-stage talks for a Series B round valuing the startup at roughly $1.2 billion, mere months after introducing itself to the market.
- Lead Investor: The round is being spearheaded by venture capital firm 8VC.
- Staggering Revenue Growth: Driven by an annualized revenue run-rate approaching $50 million, eager VCs heavily pressured the company to accept a new round much sooner than initially planned.
- The Core Mission: XDOF builds data pipelines, teleoperation collection tools, and annotation systems that act as an outsourced data-supply chain for frontier AI labs and robotics companies.
- Academic Roots: Co-founded in 2024 by UC Berkeley researchers Philipp Wu (CEO) and Fred Shentu (CTO), the company grew out of groundbreaking research into robotic learning from large datasets.
The Chronology of an Accidental Megaround
The XDOF phenomenon is less a traditional startup trajectory and more a textbook case of an academic breakthrough meeting an explosive, underserved market demand.
The Academic Origins (2023–2024)
As a PhD student at the University of California, Berkeley, Philipp Wu focused his research on a persistent brick wall in robotics: how machines learn from massive, generalized datasets. Time and again, his research stalled. The primary impediment wasn’t a lack of computing power or sophisticated model architectures; it was the sheer absence of large-scale, high-quality physical data to train models effectively.
Recognizing that he could not solve the problem alone, Wu teamed up with fellow researcher Fred Shentu. Together, they developed GELLO, an innovative, low-cost teleoperation system. GELLO allowed human operators to remotely control robotic arms with high precision, generating the exact type of rich interaction data needed to train robotic neural networks. Their subsequent research paper sent shockwaves through the academic robotics community, laying the intellectual bedrock for what would become XDOF.
Stealth to Stealth-Wealth (June 2026)
XDOF formally broke cover in June 2026, announcing a $70 million Series A round. The financing featured participation from a blue-chip roster of venture capital heavyweights, including Thrive Capital, Andreessen Horowitz, Lux Capital, and Spark Capital.
At the time, XDOF leadership had mapped out a conventional operational runway. They intended to hunker down, build out their infrastructure, and scale their customer base over the subsequent eighteen months before returning to the market for additional capital.
The Hyper-Growth Pivot (Late Summer 2026)
Plans, however, were rapidly overtaken by market reality. Buoyed by an insatiable demand from frontier AI labs racing to solve embodied intelligence, XDOF’s annualized revenue rocketed toward $50 million in a matter of weeks.
Faced with this hyper-growth, venture capital firms descended on the startup, eager to buy a stake in the infrastructure layer of physical AI. Despite having just raised capital in June, XDOF was nudged back to the negotiating table by the sheer velocity of inbound investor interest.
While the exact capital total of the ongoing Series B and whether the $1.2 billion valuation incorporates the fresh funding remain unconfirmed, sources indicate the terms are fluid and nearing finalization.
Supporting Data and the Mechanics of Physical Data Collection
To understand why investors are valuing an enterprise-stage infant at over a billion dollars, one must look at the mechanics of the robotics data bottleneck.
The Internet Isn’t Enough for Robots
When OpenAI, Anthropic, and Google trained their frontier LLMs, they had a near-infinite digital corpus. The internet provided petabytes of text and imagery. But a robot tasked with clearing a dinner table, folding a fitted sheet, or sorting warehouse inventory cannot learn dexterity from a PDF.
Physical AI requires spatial, kinetic, and tactile data. It requires understanding friction, weight distribution, variable lighting, and deformable objects—variables that change with every single iteration of a task.
The Dual-Pronged Data Engine
XDOF solves this by deploying a massive, hybrid workforce and proprietary hardware-software stack. The startup captures data using two primary methodologies:
- Remote Teleoperators: Human operators situated globally steer robotic hardware remotely, executing complex manipulation tasks to generate clean baseline behaviors.
- Egocentric Collectors: Human operators wear specialized body sensors, recording their own everyday movements—such as flattening cardboard boxes, packing shipments, or folding clothes—to teach robots human-like economy of movement.
Furthermore, XDOF has partnered directly with the UC Berkeley AI Research (BAIR) lab to release what is widely considered the largest collection of high-quality robot training data ever assembled, designated as ABC.
By packaging these workflows into scalable data pipelines and annotation engines, XDOF effectively insulates robotics companies from the dirty, unglamorous, and labor-intensive work of data collection. As of its Series A announcement, the company reported active working partnerships with 20 enterprise customers, prominently including several top-tier frontier AI labs.
Official Responses and Competitive Landscape
As typical for companies navigating late-stage, highly sensitive venture capital transactions, representatives for both XDOF and 8VC declined to respond to requests for comment from TechCrunch.
Nevertheless, the competitive landscape surrounding XDOF is rapidly heating up. While XDOF currently commands the pole position in specialized robotic teleoperation data, it is not entirely alone in the market.
- Mecka AI: A specialized startup similarly attempting to capture and curate real-world data specifically tailored for robotic training loops.
- The Scale AI Pivot: Traditional human-data annotation powerhouses like Scale AI and competitors such as Micro1 (which recently secured a $500 million valuation) are aggressively expanding their operational footprints beyond text and vision datasets, setting their sights squarely on the physical robotics sector.
Despite this emerging competition, XDOF’s head start, deep ties to UC Berkeley’s academic ecosystem, and lightning-fast commercial adoption give it a profound defensive moat.
Implications for the Future of Artificial Intelligence
The impending $1.2 billion valuation of XDOF signals a fundamental evolution in how the tech industry views the "AI Gold Rush."
For years, venture capital and public markets fixated exclusively on algorithmic innovation—larger parameter counts, deeper reasoning models, and more efficient transformer architectures. But as software intelligence approaches its current physical limits, the frontier has definitively shifted toward embodied AI.
The bottleneck is no longer what a model can think, but where and how a machine can act.
If XDOF succeeds in cementing its status as the foundational data utility for physical robotics, it will hold the keys to the kingdom. Just as no modern software company can scale without cloud infrastructure, no robotics company of the future will be able to train a general-purpose humanoid or autonomous manipulator without an industrialized data pipeline.
Should the 8VC-led Series B close at the anticipated $1.2 billion price tag, it will serve as a resounding declaration from Silicon Valley: the age of software dominance is sharing the stage with physical AI, and the infrastructure layer of the robotics revolution is officially open for business.
