In this interview from the Machina AI Summit 2026, Andrew Lonsberry, chief executive officer and co-founder of Path Robotics, joins theCUBE's John Furrier to discuss how AI-powered robotics is closing the skilled labor gap threatening US manufacturing. Lonsberry frames the problem in stark terms: the average human welder in the United States is 55 years old, with 20 to 30% projected to retire over the next two decades — a shortage no amount of capital can solve. He details how Path's robots, powered by the Obsidian neural network, reduced welding labor from 150 human hours to 9 for one customer — a 91% reduction — and explains why physical AI robots improve continuously over time, inverting the traditional automation model where day one is always the best a machine will ever perform.

The conversation also explores the data-first philosophy that has shaped Path Robotics since its founding. Because welding is a destructive process, Lonsberry notes, building a foundation dataset required deploying robots to customer floors early and aggregating real-world data across every deployment in North America — a strategy he describes as "machine learning starts with machine failing." He details the Weld World Model and the company's plans to extend that approach horizontally across manufacturing tasks as the broader data flywheel accelerates. The interview also examines the evolving human role in physical AI: Path's Mission Control team, many of whom transitioned from hands-on trades, now oversee robot fleets, diagnose anomalies and ensure data quality — jobs Lonsberry sees as safer and more fulfilling than the work they replace. With over $360 million raised and revenue growing 15x over the past 15 months, he outlines why converging compute access, maturing algorithms and a rapidly expanding deployment base will drive step-function improvements in robot capability in the years ahead.

In this interview from MACHINA 2026 in Paris, Jason Yueh, chief technology officer of VicOne, joins theCUBE's John Furrier to discuss why cybersecurity is the missing foundation beneath physical AI's rapid move into production. Yueh explains that VicOne, a Trend Micro subsidiary, provides full lifecycle protection for robots and autonomous systems — spanning software bill of materials (SBOM) management, hardware component origin detection to comply with U.S. BIS supply chain rules and Europe's Cyber Resilience Act (CRA), in-device protection and continuous monitoring through dedicated robotic security operations centers.

The conversation also explores the AI-specific risks that emerge as robots grow more capable, including prompt injections and adversarial image patches that can cause a robot to ignore a stop sign without any traditional software compromise. Yueh notes that while fixed-arm robots and warehouse autonomous mobile robots are well-established, humanoid and quadruped robots are now crossing from demonstration into real production deployments in open environments such as security patrol. He underscores what he sees as the critical gap in the industry: most robot makers are still focused on functional safety while treating cybersecurity as secondary, even though one cannot exist without the other. From securing seven of the top twenty passenger vehicle companies to enabling medical robots and autonomous airport operations, Yueh outlines how the physical AI industry must close that gap to build safe, production-ready systems at scale.

In this interview from the Machina AI Summit 2026 in Paris, France, Amanda McMaster, interim chief executive officer of Boston Dynamics, joins theCUBE's John Furrier to discuss how physical AI is accelerating from research demos into enterprise deployment. McMaster details the company's landmark decision to commercialize Atlas, its humanoid robot, transitioning from hydraulic to electric actuation and pairing that shift with AI to meet surging market demand. She explains how AI has compressed behavior authoring from roughly a year to just a few hours, with close to 99% reliability — a step-change that is unlocking new classes of real-world applications at a pace the industry has never seen before.

The conversation also explores how Spot, Boston Dynamics' quadruped robot, is already delivering value in industrial settings through visual, thermal and acoustic inspection — replacing manual walkthroughs and catching anomalies without requiring facilities to undergo costly retrofits. McMaster outlines a two-pronged AI strategy that combines internal development with partnerships with leaders such as Google DeepMind, targeting sub-two-year ROI payback for customers. She details a "data bank" concept designed to help enterprises make sense of the operational data robots collect over time, laying the groundwork for a platform ecosystem built on top of Boston Dynamics hardware. Backed by Hyundai's vertically integrated supply chain and a new manufacturing facility in Savannah, Georgia with capacity for up to 30,000 robots per year, McMaster makes the case that 2027 could be the breakout year for physical AI — and that robotics IPOs may not be far behind.

In this interview from the Machina AI Summit 2026, Abhinav Gupta, co-founder and president of Skild AI, joins theCUBE's John Furrier to discuss how a single universal robot brain can power a heterogeneous fleet of machines across real-world physical environments. Gupta traces his path from Carnegie Mellon research and Facebook's FAIR robotics lab to co-founding Skild AI, driven by the conviction that while robot hardware rapidly commoditizes, the intelligence layer is where durable value accumulates. He explains why the "ChatGPT moment" for robotics will unfold gradually over the next 18 months rather than overnight — shaped by trust, safety and the hard physics of deploying machines in unpredictable environments.

The conversation also explores how Skild AI's omni-bodied Skild Brain unifies learning across heterogeneous robot fleets, allowing quadrupeds, humanoids and industrial arms to share data and transfer skills between form factors. Gupta describes a defining moment when a robot with broken ankle motors adapted its gait in real time without having been trained for that failure — an unexpected demonstration of in-context learning that revealed the full power of cross-form-factor data pooling. On the data side, scaling laws are beginning to emerge: a deployment at NVIDIA's H100 GPU factory now requires less than 24 hours of training data, down from more than a month just one year earlier. With more than $2 billion raised at a $14 billion valuation and NVIDIA among its investors, Skild AI is moving toward what Gupta sees as a 2027 breakout year — when trust, safety and fleet economics converge to bring physical AI into enterprise environments at scale.

In this interview from the Machina AI Summit 2026, Jeff Cardenas, co-founder and chief executive officer of Apptronik, joins theCUBE's John Furrier to discuss how humanoid robots are transitioning from experimental technology into practical enterprise deployment. Cardenas frames adoption across three phases — back of house, front of house and the home — explaining why manufacturing and logistics represent the most immediate proving ground for the industry. He argues that mobility is largely a solved problem, pointing to autonomous vehicles as evidence, and that dexterous manipulation is now the critical frontier separating early pilots from mass-scale deployment.

The conversation also explores how Apptronik is addressing the data challenge at the core of robot training through Robot Park, a 90,000-square-foot data factory designed to collect robot-generated, human-generated and synthetic data across multiple modalities. Cardenas details the company's modular approach with Apollo, which can operate on both legs and wheels depending on the application, underscoring that robots will not be confined to a single form factor. He also addresses the global competition in robotics, framing it as the space race of our time, and acknowledges China's manufacturing strength while arguing the U.S. holds a clear lead in AI models. From a $935 million Series A — believed to be the largest in Texas history — to active pilots with customers including Mercedes-Benz and GXO, Cardenas outlines how Apptronik is positioning itself to lead the next phase of physical AI at scale.

In this interview from the Machina AI Summit 2026, Jonathan Hurst, co-founder and chief robot officer at Agility Robotics, joins theCUBE's John Furrier to discuss the inflection point driving humanoid robots from controlled work cells into commercial deployment at scale. Hurst traces Agility Robotics' decade-long evolution from university spin-out to commercialization, explaining how real-world deployment of its Digit robot forced the team to understand what enterprises actually need — from insurance coverage and ROI to IT infrastructure compatibility. He details why bin handling in narrow spaces is the ideal first use case for humanoid robots, requiring two arms for the lift and two legs for balance in environments never redesigned for machines.

The conversation also explores Agility Robotics' data strategy, where digital twins, simulation and reinforcement learning combine in a layered approach — analogous to a student learning an instrument through demonstration and then thousands of hours of practice. Hurst breaks down the make-versus-buy calculus behind the company's engineering philosophy, noting that cameras and chipsets can be sourced off-the-shelf while custom actuators represent core intellectual property no supplier currently provides. He also touches on the fleet intelligence model, where skills learned by one Digit robot propagate across the entire fleet. From building certified safe data buses to AI-driven human detection systems, Hurst outlines why the two-year safety engineering effort behind Digit V5 is not a constraint on ambition but the foundation that finally allows humanoid robots to step out of the cage and scale.

In this interview from the Machina AI Summit 2026, Inhi Cho Suh, chief executive officer of Niantic Spatial, joins theCUBE's John Furrier to discuss how real-world foundation models are giving robots the spatial awareness they need to navigate unpredictable physical environments. Suh explains that Niantic Spatial does not build robots — it builds the perception and navigation layer that makes them viable in dynamic, real-world environments. She outlines how the company's Visual Positioning System (VPS) and high-fidelity 3D reconstruction technology allow robots to navigate complex indoor-outdoor spaces where GPS fails, with Coco Robotics already using the system for last-mile urban delivery. A newly announced USDZ format converts 3D Gaussian splats into aligned polygonal meshes, enabling faster and safer robotics deployment. Suh also underscores the scale of the opportunity, noting that physical AI addresses roughly 80% of the economy — making it four times the potential of digital AI.

The conversation also explores what sets Niantic Spatial apart in an increasingly crowded world-model market. Suh argues the company's core advantage is grounding everything in real-world data — a "real to sim to real" approach that eliminates the hallucinations common in purely synthetic training pipelines. She highlights how Niantic Spatial's computer vision technology generates high-resolution 3D Gaussian splats from commodity cameras costing just a few hundred dollars, removing the need for expensive LiDAR systems that most competitors rely on at scale. A recently announced partnership with Spexi adds aerial and drone footage to the company's data acquisition capabilities. On the enterprise side, Suh identifies safety as the single biggest barrier to deployment in spaces where humans and robots interact, and outlines vertical opportunities across energy utilities, industrial manufacturing and logistics — including asset maintenance in oil refineries and week-to-week reconstruction site monitoring. From rethinking digital twins for brownfield conversion to advancing semantic scene understanding as the next research frontier, Suh makes the case that Niantic Spatial is positioning itself as a foundational layer across the robotics, logistics and industrial AI markets.

In this interview from the Machina AI Summit 2026, Andrew Wooten, co-founder and chief product officer of Robo AI, joins theCUBE's John Furrier to discuss how a novel video-to-action approach is bridging the gap between robotic demos and real-world industrial deployment. Wooten explains that despite the proliferation of industrial robots, virtually none are capable of intelligent manipulation — most are locked into pre-programmed motions with no ability to handle variability. He outlines the core problem: the Vision-Language-Action (VLA) model approach fails because it cannot achieve the data diversity needed to generalize across real-world tasks. Robo AI's answer is the Direct Video-Action (DVA) model, which pre-trains on internet-scale video and post-trains with robot actions — a paradigm validated at a large automotive factory before the company emerged from stealth in 2026.

The conversation also explores Robo AI's go-to-market strategy and early commercial momentum. Wooten details why manufacturing and logistics are the ideal first proving ground: multi-shift operations, built-in maintenance infrastructure and acute labor shortages make the economic case for intelligent robotics compelling in ways that consumer or hospitality environments cannot yet match. He underscores a customer-first development philosophy — walking factory floors, mapping full end-to-end tasks and identifying edge cases before any technology is committed to code. Having raised $450 million in a Series A with a team approaching 100 people, Robo AI is now hiring across hardware engineering, AI research in computer vision and video generation, data infrastructure and sales. From targeting its first commercial deployments later in 2026 to positioning for what Wooten expects to be a breakout 2027, Robo AI is building toward a future where robots move beyond spectacle into sustained, measurable industrial impact.

In this interview from the Machina AI Summit 2026 in Paris, David Reger, founder and chief executive officer of Neura Robotics, joins theCUBE's John Furrier to discuss how cognitive robotics is moving from research demos into real-world deployment and why solving the physical AI data problem could unlock the industry's ChatGPT moment. Reger argues that the most important shift of the past 18 months isn't technological — it's a mindset change. Since ChatGPT proved that AI is real, attention has shifted to making it physical. He explains why robotics has lagged: unlike language models trained on existing internet data, embodied AI requires sensor data that simply doesn't yet exist, along with compute architectures built around reflexive nervous systems rather than static brains. With projected labor shortfalls of 6-7 million workers in the US and more than 80 million in China by 2030, Reger frames cognitive robotics — machines that can hear, see and feel — as the only scalable answer.

The conversation also explores how Neura Robotics is deliberately tackling the full hardware and software stack, a choice Reger defends as the only path to feeding robots the right data. He uses the analogy of swimming to illustrate why vision-language-action models alone aren't sufficient: physical tasks require a reflexive nervous system operating at millisecond speeds, not just a centralized brain. Working with NVIDIA, Qualcomm and Amazon, Neura is developing that architecture and plans to open source its sensor and compute stack to accelerate ecosystem growth. Reger also details Neuraverse, an agentic AI orchestration platform that sits above the robot itself, enabling enterprises to train their own specialized models through a "brain account" and license that knowledge competitively rather than surrender it to a centralized system. From redefining compute architecture for physical AI to building a platform where businesses can own and monetize their operational know-how, Reger outlines why an open, platform-first approach is the only model that can responsibly scale cognitive robotics at global scale.