Skild AI raises $1.4B at $14B valuation to build a general-purpose robot brain for any hardware

Summary

Skild AI has closed roughly $1.4 billion in new funding at a $14 billion valuation, one of the largest robotics software raises on record. Deepak Deepak, CEO and co-founder, confirmed the round during the January 15 appearance, noting the announcement came a day after a separate reveal around learning from human video data.

The company's core thesis is hardware-agnostic. Skild builds a single AI model — described as a general-purpose robot brain — that runs on any form factor, from humanoids and quadrupeds to robotic arms. The end-to-end model takes raw pixel inputs and outputs physical actions, effectively replacing the need for custom software per hardware platform.

Go-to-market strategy leans on existing hardware partners. Rather than manufacturing its own robots, Skild licenses its software as a service to hardware companies and enterprise end users who already have deployed equipment. This sidesteps the supply chain bottlenecks that have plagued data center buildouts, since industrial robotics hardware supply chains are already mature, even if underutilized due to prior software limitations.

Current deployments span point-to-point delivery, warehousing, and manufacturing, with a specific focus on human-collaborative environments. Deepak frames this as the critical unlock — legacy industrial robots are physically isolated from workers, reducing their ROI ceiling. Skild's model is designed to operate in open, unstructured spaces alongside people, which introduces the randomness and unpredictability that require genuine AI reasoning.

Data Scarcity Is the Defining Technical Problem

Unlike language or computer vision, robotics has no large pre-existing dataset to train on. Deepak describes this as the fundamental reason the field has stagnated for roughly 70 years despite being one of the oldest areas in technology. Skild breaks the cold-start problem through two bootstrapping sources: learning from human activity videos scraped from YouTube, Flickr, and similar platforms, and synthetic data generated through simulation. Real-world deployments then feed a data flywheel — more deployments generate more proprietary operational data, which improves model performance iteratively.

Consumer Robotics Is a Second-Order Market

Deepak pushes back on the Silicon Valley consensus that home robots are imminent. His view is that enterprise-facing consumer environments — grocery stores, hospitals, retail fulfillment — will serve as the proving ground before residential adoption becomes viable. The behavioral argument is straightforward: consumers given a robot today would likely shelve it within months, the way most people stop using a home massage chair. Familiarity has to develop through repeated public exposure before home deployment becomes sticky.