Capital is moving from software that thinks to machines that do. Two financings reported in the same period illustrate the shift: Path Robotics raised a $300 million Series E at a $2.45 billion valuation, and Nidus raised a $30 million seed round led by Kleiner Perkins. One builds welding robots; the other builds teleoperation data for defence manufacturing. Together they describe a new phase of the AI boom — physical AI.
Path Robotics: Robots That Weld
Path Robotics builds welding robots powered by its Obsidian model. Welding is a useful test case for physical AI because it is skilled, dangerous, repetitive and in short supply. Experienced welders are hard to hire, and the work is physically demanding. A robot that can weld reliably, adapting to variations in the workpiece rather than following a rigid pre-programmed path, solves a real and expensive problem.
The $300 million Series E at a $2.45 billion valuation is a large vote of confidence in that proposition. It suggests investors believe the company's technology has moved beyond demonstration and into deployment at a scale that justifies a multi-billion-dollar valuation. For a hardware company, that is a significant statement: hardware is capital-intensive, slow to scale and hard to iterate, and investors have historically been cautious about it. A round of this size signals that the caution is fading where the demand is clear.
Nidus: The Data Behind the Robots
Nidus, which raised a $30 million seed round from Kleiner Perkins, works on a different part of the same problem: teleoperation data for defence manufacturing. Teleoperation means a human operator controlling a robot remotely, and the data generated by that process — the movements, decisions and corrections of a skilled human — becomes the training data that teaches the robot to act on its own.
This is the physical AI equivalent of the data flywheel that powered software AI. The more teleoperation data a company collects, the better its robots become; the better its robots become, the more customers use them; the more customers use them, the more data is generated. The moat is not the robot itself, which competitors can build, but the dataset of human expertise that trains it.
Defence manufacturing is a telling early market. Defence customers have urgent needs, large budgets and a tolerance for new technology that commercial manufacturers often lack. They are also, by necessity, early adopters: when the mission requires it, they deploy. For a young company like Nidus, defence-industrial customers offer revenue, real-world testing and credibility that is hard to obtain elsewhere.
Why Capital Is Moving to Physical AI
Several forces are pushing investors toward robots that do physical work. The first is saturation in pure software AI. The market for chatbots, copilots and text generation is crowded, and differentiation is increasingly hard. Physical AI offers a less crowded field with clearer unit economics: a robot that welds saves a measurable amount of money compared with hiring a welder.
The second force is the labour shortage in manufacturing. Skilled trades — welding, machining, assembly — face persistent shortages in the United States and other advanced economies. Employers cannot hire their way out of the problem, so they are looking for automation that works in the real world rather than in a laboratory.
The third force is the maturation of the underlying technology. Better sensors, cheaper compute and more capable AI models have converged to make robots that can handle variation — the essential requirement for real-world work — feasible in a way they were not a decade ago.
What Founders Should Take From This
For founders, the lesson is not to build another chatbot. It is to find a physical task where labour is scarce, the work is well-defined, and the economic value of automation is easy to calculate — and then build the data advantage that makes your system better over time.
Teleoperation is worth particular attention. It solves the cold-start problem: before a robot can work autonomously, a human can drive it, and every hour of human operation is training data. Founders who design their systems to capture that data from day one build a compounding advantage.
Defence and industrial customers deserve a closer look too. They are demanding, but they pay, they deploy, and they provide the kind of hard feedback that makes a product genuinely robust. A startup that can survive a defence-industrial deployment can survive almost anything.
Physical AI is no longer a science project. With $300 million going into welding robots and $30 million into the data that trains them, it is a funded, competitive market — and the founders who understand the data moat will be the ones who win it.
Angeled.com connects founders with investors.
Nidus seed round announcement ($30 million, Kleiner Perkins)
Read Next
General Translation Raises $27 Million Series A Led by a16z to Build Agent-First Localization Infrastructure
LangChain Raises $300 Million Led by CapitalG at a $3 Billion Valuation to Build the Agent Stack
Mount Thor Raises $45 Million to Run Cloud Fleets of Macs for AI Agents
OpenAI Reportedly Raises $15 Billion at a $1.25 Trillion Valuation
Q-CTRL Raises $290 Million Series C Led by Salesforce Ventures, Reaching Unicorn Status
SE3 Raises $30 Million Series A Led by Accel



