A mega-round signals that robotics AI is entering its scale phase
Skild AI is in advanced talks to raise more than US$1 billion in a new funding round led by SoftBank Group and Nvidia, a deal that would value the robotics-AI startup at roughly US$14 billion. If completed, the round would nearly triple Skild’s earlier 2025 valuation and rank it among the most valuable private AI companies globally. The scale of the raise matters as much as the number. Investors are no longer treating robotics AI as a long-horizon science bet. Instead, they are pricing it as an urgent platform shift, one that could turn robots from fixed-task machines into flexible workers across factories, warehouses, hospitals, and homes. That shift sits at the heart of Skild’s pitch: build a universal “brain” for robots that works across many bodies, environments, and jobs.
From language-model hype to physical-world ambition
Skild AI is a young company with unusually heavyweight DNA. Founded in 2023 by former Meta AI researchers Deepak Pathak and Abhinav Gupta, it set out to build foundation models for robotics rather than produce hardware. In July 2025, the company unveiled its first general-purpose model, Skild Brain, designed to transfer skills across different robot types. The thesis is borrowed from the language-model boom: train a large base model on broad data, then fine-tune it for specific tasks. What makes robotics harder is the data. Robots do not generate infinite text streams the way the internet does. So Skild mixes simulation, human-action video, and real-world robot feedback to close that gap.
The timing has been favorable. After the success of large language models, capital has begun flowing toward “physical AI”—systems that perceive, plan, and act in the real world. Robotics is the sharpest test of that idea because it requires not only reasoning, but also safe control under messy conditions. Skild’s early traction attracted major backers. Amazon participated in earlier rounds, while Lightspeed and Coatue helped build its Series A base. Nvidia and Samsung joined its Series B this year, signaling that hardware leaders want seats at the table as robotics intelligence becomes a new compute market.
Why SoftBank and Nvidia are leaning in
This pending round aligns cleanly with SoftBank’s evolving strategy. The group has shifted its AI focus from purely software plays toward infrastructure and embodied intelligence. Its robotics history runs deep, from early bets on automation to building platforms that can scale globally. Skild offers SoftBank a direct lever into that future without requiring it to pick a single robot manufacturer. If a universal robotics brain wins, SoftBank wins across the whole hardware field.
For Nvidia, the logic is both strategic and commercial. Robotics foundation models need massive compute for training and high-performance chips for inference. Nvidia already dominates AI data centers. Now it wants to extend that dominance into machines that operate outside of servers. By backing Skild, Nvidia helps shape a software stack that could standardize on its hardware over time. This is the same playbook that made Nvidia central to deep learning: win the platform layer, then let ecosystems pull your chips in.
Skild, meanwhile, needs capital for the expensive parts of its roadmap. Training a general robot model requires more compute, more simulation depth, and a far wider data flywheel. It also requires partnerships with robot makers and industrial users who can deploy models in real settings and feed performance signals back into training. A US$1 billion round buys Skild time to build that network at scale instead of inching forward on smaller checks.
The bet is on “brains,” not bodies
Skild’s rise reflects a wider shift in robotics thinking. For years, the field advanced through better bodies—stronger actuators, cheaper sensors, improved mechatronics. Yet most robots still struggle outside controlled environments because their intelligence is narrow. Foundation models flip that logic. They try to make intelligence general first, and let hardware variety follow. That is why Skild is positioned as a “brain company.” It does not need to win the hardware race directly. It needs to become the intelligence layer that many hardware players trust.
If Skild succeeds, the implications are large. A universal robotic model could reduce the cost of deploying robots into new jobs. Instead of writing task-specific code for each factory line or warehouse layout, companies could fine-tune a base model on small sets of local data. That would widen the market from a few high-automation sectors into the long tail of businesses that cannot currently afford robotics integration.
Still, the path is not automatic. Physical AI brings safety and reliability requirements that language models do not face. A hallucinated sentence is embarrassing. A hallucinated movement can break equipment or hurt people. Skild’s emphasis on safety constraints and continuous learning is a sign the company understands this. Yet public expectations will rise fast after a US$14 billion valuation. The burden of proof shifts from “interesting research” to “repeatable industrial value.”
A long technical race, now funded like a short one
The next two years are critical. Skild must prove three things at once. First, that its base model transfers across robot categories without heavy retraining. Second, that performance improves reliably in real deployments, not just simulations. Third, that its data flywheel scales faster than rivals. This race is crowded. Multiple startups and big-tech labs are pursuing robotics foundation models, and some are pairing them with proprietary hardware, which can speed feedback loops.
Yet Skild now has a clear edge if the round closes. It would hold one of the largest war chests in robotics AI, plus deep ties to the world’s most important AI-compute supplier. That combination could accelerate training iteration and attract high-end talent. It also gives Skild credibility with industrial partners who want to back a likely winner, not a fragile experiment.
In Asia, the ripple effects could be strong. Japan, South Korea, and China are scaling robotics in manufacturing and eldercare, while Southeast Asia is beginning to automate logistics as wages rise. A mature, universal robotics brain would lower adoption barriers across these markets. It could also encourage regional hardware makers to standardize on common intelligence platforms, much like Android did for smartphones.
A defining moment for physical AI’s commercial timeline
Skild AI’s near-US$1 billion round at a US$14 billion valuation marks a turning point in how the market values robotics intelligence. SoftBank sees a platform bet that can sit above many robot makers. Nvidia sees the next compute frontier, one that moves AI out of the data center and into the physical economy. For Skild, the capital is a chance to industrialize a research-heavy ambition before rivals close the gap. Whether foundation models can truly generalize in robots remains a hard technical question. What is no longer in doubt is investor belief that the payoff, if cracked, could reshape work, production, and services across the world.
