Jeff Bezos, the founder of Amazon, is in early talks to raise about $100 billion for a new investment vehicle that would buy traditional manufacturing companies and overhaul them with artificial intelligence, according to a report in the Wall Street Journal, echoed by Reuters and other outlets.

The effort is described as a “manufacturing transformation vehicle” focused on industrial mainstays such as chipmaking, aerospace and defence, according to the news reports. Bezos has been courting some of the world’s biggest asset managers, as well as sovereign-wealth funds in the Middle East and Asia, during recent trips to the region.

The fund, which would rival SoftBank’s $100 billion Vision Fund in sheer scale, marks one of the most ambitious attempts yet to wire AI directly into the foundations of heavy industry rather than the consumer internet. If successful, it would accelerate a broader shift in which capital and algorithmic know-how migrate from software and advertising to factories, supply chains and the “real economy.”

It also underscores how a small set of tech billionaires now treat large buyout pools as strategic weapons in an emerging contest over who controls the data and industrial capacity on which future AI systems will feed.

At $100 billion, the proposed pool would give Bezos a war chest big enough to roll up mid-sized industrial firms that are struggling with labor costs, capital-intensive retooling and the investment required to digitise their plants. For governments anxious about productivity, reshoring and defence supply chains, such a vehicle could look like a timely private complement to industrial policy — or, to critics, the beginnings of an unaccountable industrial conglomerate built in AI’s image.

The various news reports call the venture a bid to “drive and speed up automation” inside acquired companies, using AI models being developed at Project Prometheus, Bezos’s industrial AI startup. Bezos serves as co-founder and co-CEO of Project Prometheus, which is separately seeking up to $6 billion in fresh capital.

“The goal is to transform legacy manufacturers into AI-first companies that can design, simulate and operate complex systems far more efficiently than today,” one investor presentation says, according to people familiar with the pitch, as reported by the WSJ.

Labor advocates warn that aggressive automation could hollow out middle-skill jobs, even as investors bet that AI-optimised plants will pay for themselves through higher throughput and fatter margins.

Subscribe now

BMW Group is bringing humanoid robots into series production in Germany for the first time, extending an ambitious push to fuse artificial intelligence with factory automation. The carmaker will run a pilot at its Leipzig plant using AEON, a humanoid robot developed by long-standing partner Hexagon, to support assembly of high‑voltage batteries and component manufacturing from summer 2026.

The move follows a 2025 trial in Spartanburg, South Carolina, where Figure AI’s Figure 02 robot handled precise, repetitive body‑shop tasks in BMW’s X3 production line. Over ten months, Figure 02 worked ten‑hour shifts, five days a week, moving more than 90,000 sheet‑metal components and supporting the production of over 30,000 vehicles, while integrating with BMW’s existing smart transport robots and robotics ecosystem.

Automakers and technology groups are racing to deploy humanoid robots as labour markets tighten, vehicle architectures grow more complex and traditional industrial automation reaches its limits. Proponents argue that mobile, human‑shaped machines, powered by increasingly capable AI systems, can slip into brownfield plants, take on ergonomically punishing or safety‑critical tasks and be reprogrammed as models and processes change, potentially reshaping factory work in the same way industrial robots did in the late 20th century.

“Digitalisation improves the competitiveness of our production – here in Europe and worldwide. The symbiosis of engineering expertise and artificial intelligence opens up entirely new possibilities in production,” said Milan Nedeljković, BMW’s production chief.

BMW describes the initiative as part of a broader “Physical AI” strategy that combines digital AI agents, a unified data platform and robots able to learn and operate autonomously on the shop floor. It has created a Center of Competence for Physical AI in Production in Munich to vet technology partners, run lab‑based evaluations with real use cases and then move promising systems into plant‑level pilots.

At Leipzig, AEON will be tested as a multifunctional platform whose human‑like form factor can carry different grippers and scanning tools and move dynamically on wheels. The emphasis, BMW says, is on complementing existing automation, taking over monotonous and physically demanding tasks, and improving working conditions rather than replacing staff.

“Our aim is to be a technology leader and to integrate new technologies into production at an early stage,” said Michael Nikolaides, who oversees BMW’s global production network and supply chain.

Subscribe now

Humanoid Robots Are Entering the Workforce

Long a staple of science fiction, humanoid robots are edging into factory floors and warehouses, but Forrester expects the transition to be cautious rather than explosive. A new report from the research firm, ‘The State of Humanoid Robots, 2026,’ argues that the machines are shifting from spectacle to tool, with early adopters chasing productivity gains while wrestling with cost and regulatory headaches. Other analysts have proposed a staged adoption model, with each phase contingent on clearing the economic hurdles of the last.

Early Deployments Are Delivering Measurable Results

Forrester’s latest automation survey finds that 69 percent of automation decision-makers are adopting or planning to adopt humanoid robots, with deployments already underway in manufacturing, logistics, healthcare, and customer service. Early adopters report 40 percent reductions in processing errors and 20 percent declines in labor costs where humanoids standardize repetitive, high-friction work, from BMW’s use of robots for ergonomically awkward assembly to KEENON Robotics cutting restaurant labor costs through automated food preparation and cleaning.

The push comes amid a broader race to automate labor-intensive tasks as companies confront aging workforces, skills shortages, and persistent margin pressure, though the debate on humanoid viability remains unsettled. Investors and boardrooms are betting that advances in generative and physical AI, coupled with cheaper cloud-based infrastructure, will allow humanoid machines to operate safely and reliably in spaces built for people, even as regulators and risk officers fret about safety, cybersecurity, and liability.

“Humanoid robots are no longer a futuristic fantasy, but a pragmatic tool for operational transformation,” says Charlie Dai, a vice president and principal analyst at Forrester. “Leaders must approach this technology with disciplined experimentation, viewing these robots as workforce multipliers that augment human capabilities rather than wholesale replacements.”

AI Advances Are Accelerating Humanoid Capabilities

The report points to advances in AI to explain the acceleration. NVIDIA’s Isaac GR00T-Dreams, for instance, is said to cut model-development time from three months to 36 hours by using synthetic motion generation and multimodal foundation models, while physical-AI techniques narrow the sim-to-real gap to improve balance and reliability in real-world settings.

Humanoids Are Augmenting Workers, Not Replacing Them

For now, Forrester stresses that humanoids are amplifying human workers rather than displacing them. In warehouses, AgiBot’s A2-W handles 30 percent of material transport with zero errors, and in Singapore’s Sengkang Community Hospital, a robot called Dexie supports multilingual dementia care so staff can focus on more complex, higher-empathy tasks.

What Is Holding Back Large-Scale Deployment

Yet the firm is skeptical that a flood of chrome-plated colleagues is imminent. High R&D costs, deployment complexity, and underdeveloped rules on safety, cybersecurity, and liability will keep most projects at pilot scale for the next two years, it says, with prudent companies advised to take a measured, human-centric approach rather than succumb to humanoid hype. We examine the pilot to production gap in detail, including the operational and economic conditions that determine when manufacturers move from testing to full deployment.

Related posts

Subscribe now

Agility Robotics, maker of the bipedal robot Digit, has signed a commercial robots-as-a-service agreement with Toyota Motor Manufacturing Canada (TMMC), turning a recent pilot into a full deployment at the carmaker’s plants in Ontario.

Digit units will be introduced to lines involving manufacturing, supply-chain and logistics tasks, with the aim of easing physical strain on workers and streamlining shop-floor operations, the Oregon-based robotics company said in a press release.

The deal follows a trial at TMMC’s Cambridge facility and will see humanoid robots integrated into existing workflows rather than driving costly retooling of production lines. Agility says its Digit platform and Agility Arc cloud software can be slotted into current layouts to handle monotonous, process-automated jobs that are hard to staff and retain, from repetitive material handling to backroom logistics. The humanoid robot economics behind deployments like this one, including payback periods and cost projections, are explored in our manufacturing analysis.

Humanoid robots have become the latest frontier in factory automation as manufacturers confront ageing workforces, labour shortages and pressure to raise productivity without sacrificing safety. Agility already counts Amazon, GXO and Schaeffler among its customers, and is pitching Digit as a general-purpose, “human-centric” machine that can walk where people walk and adapt to changing workflows using artificial-intelligence tools.

For global automakers such as Toyota, whose Canadian arm runs its largest production operation outside Japan, humanoids offer a way to automate ergonomically risky or low-value tasks while preserving the company’s long-cultivated “Respect for People” ethos on the line.

“Toyota Motor Manufacturing Canada has long been a leader in automotive manufacturing innovation,” said Tim Hollander, TMMC’s president. “After evaluating a number of robots, we are excited to deploy Digit to improve the team member experience and further increase operational efficiency in our manufacturing facilities.”

Beyond the initial rollout, Agility and TMMC plan to assess further use cases where robots and AI could augment automotive production, particularly in extremely repetitive and physically taxing jobs. Agility’s chief executive, Peggy Johnson, said the company’s next generation of Digit is intended to be the first “cooperatively safe” humanoid designed to work alongside people, enabling customers to scale deployments beyond fenced-off automation cells.

Agility argues that, as its robots learn new tasks and adapt on the fly, they will become a flexible layer of automation that can be reallocated as product mixes and production schedules shift.

Related posts

Subscribe now

Bedrock Robotics and Machina Labs said in separate announcements they have raised fresh investments to automate some of the most labour- and capital-intensive corners of the industrial economy, underscoring investors’ appetite for software-defined hardware in old-line sectors.

San Francisco-based autonomous construction company Bedrock Robotics has raised $270 million in a Series B round to advance software-defined machines for infrastructure projects. The investment, co-led by CapitalG and Valor Atreides AI Fund, brings the company’s total capital to more than $350 million and pushes its valuation past $1.75 billion. Backers include NVentures, Tishman Speyer, MIT and C4 Ventures.

Bedrock, which emerged from stealth in July 2025 with $80 million in seed and Series A capital, retrofits excavators, bulldozers and loaders so contractors can deploy supervised and, eventually, fully autonomous fleets on complex infrastructure projects.

The new capital will help shift customers from piloting individual autonomous machines to orchestrating connected fleets across large sites. A recent deployment on a 130-acre manufacturing site validated its mass excavation capabilities.

Across rich economies, ageing workforces, infrastructure pushes and subsidies for factories building chips, batteries and data centres are colliding with chronic shortages of skilled tradespeople and rigid 20th‑century manufacturing assets. Investors are betting that autonomy and AI-native production lines can unlock capacity without a commensurate rise in headcount, recasting earthmoving and metal-forming as software problems rather than purely mechanical ones.

“The construction sector is facing an overwhelming demand that it cannot fulfill,” said Boris Sofman, Bedrock’s boss, arguing that system-level autonomy will let contractors redeploy scarce operators to supervision and strategy while fleets of machines run longer with less idle time.

Machina Labs

In Los Angeles, Machina Labs has secured $124 million in Series C financing to deploy its first “Intelligent Factory” for metal manufacturing in defence and aerospace. Led by an unnamed lead investor, the round will fund a 200,000-square-foot facility housing up to 50 RoboCraftsman cells. The company plans to produce thousands of metal assemblies annually for customers in defence, advanced mobility and automotive sectors.

The Intelligent Factory uses AI-driven robotics to replace traditional tooling, enabling rapid reconfiguration for diverse parts. Machina aims to cut lead times and support high-mix, low-volume production amid supply-chain strains.

For defence ministries, aerospace primes and carmakers fretting over supply-chain fragility, such technologies promise shorter lead times, more configurable plants and a measure of industrial resilience.

“The world’s most advanced designs are being held back by 20th-century factories,” said Edward Mehr, the company’s chief executive. “This round allows us to scale manufacturing infrastructure that moves at the speed of software. We’re not just making parts, we’re reprogramming the factory itself to serve defense, aerospace, and automotive customers who can’t afford to wait.”

Subscribe now

AI is rapidly boosting robots’ capabilities, efficiency and adaptability on factory floors and in service settings, the International Federation of Robotics (IFR) argues in a new position paper.

IFR highlights how deep-learning computer vision now allows robots to see and interpret visual data for tasks such as object recognition, barcode reading, sorting and real-time production-line monitoring, while supervised learning powers defect detection, predictive maintenance, quality inspection and process optimisation. Natural-language processing lets collaborative and service robots understand and respond to spoken or written commands, and mobile robots blend data from cameras and LiDAR to perform simultaneous localisation and mapping for warehouse and shop-floor navigation.

AI for robotics promises to bring the productivity improvements seen from generative AI to the physical world, even as governments and boardrooms alike try to balance the gains from AI-enabled automation with unease over jobs, safety and the energy appetite of large models.

Robot installations are already taking over physically demanding and repetitive tasks, freeing workers from harsh conditions, even as AI adds demand for data scientists, machine-learning specialists and ethicists, and forces companies and employees to keep re-skilling to remain competitive. At the same time, the IFR notes, AI-enhanced efficiency and output may spur economic growth while intensifying pressure on businesses and workers to adapt to faster cycles of technological change.

“In future, AI in robotics will further influence how teams work, how decisions are made, and how performance is monitored,” the IFR notes, adding that improved workflows “may also raise concerns about employee surveillance or reduced autonomy.”

The federation notes that reinforcement learning, though still emerging in industrial settings, is gaining traction in motion and path planning, grasping and adaptive control, where robots learn by trial and error in dynamic environments. It singles out generative AI as the next step, predicting that models will generate code for entire robotic functions from natural-language instructions. The quality of AI-generated code should be rigorously tested IFR adds.

Developers and users, it cautions, must contend with data poisoning, biased or compromised training sets and the unpredictability of autonomous systems, since malfunctions in the physical world can have more severe consequences and human-robot collaboration must remain physically safe at all times. A growing focus on sustainability will push robotics towards efficiency and longer robot lifespans, the IFR says, even as the sector confronts the ecological cost and carbon footprint of training large AI models.

Subscribe now

Humanoid robots, long a staple of science fiction, are edging closer to commercial reality as companies begin structured trials in factories, warehouses and commercial buildings, according to new analysis from Bain & Company. The consultancy argues that the timing and scale of adoption will depend less on gee-whizz demonstrations and more on prosaic factors such as power systems, sensor density and measurable returns on investment.

Bain says the technology stack for humanoids is progressing unevenly: intelligence and perception are advancing quickly but are still short of enabling fully autonomous operation, while power systems and dense, reliable sensing remain stubborn bottlenecks. Large-scale penetration, it contends, will only come once those constraints ease, safety and privacy concerns are addressed, and early adopters can prove positive returns that justify broader rollout.

Behind the cautious timeline lies a bigger wager about the future of work and automation. Over the next decade, Bain expects humanoid robots to move from experimental deployments to large-scale use across three successive waves —industrial, commercial and then consumer — potentially reshaping labour-intensive sectors from automotive and mining to healthcare, hospitality and domestic services.

Their spread will test how quickly firms, regulators and workers are willing to trust machines that look and move like people, yet are designed to slot into tightly optimised workflows rather than to replace humans wholesale.

“Humanoid robots are on the way, but their success depends as much on economics and trust as on technology,” the report concludes.

In Bain’s schema, early industrial use cases span automotive production, mining and construction, where humanoids can step into dangerous, variable or ergonomically awkward tasks that traditional fixed or wheeled robots struggle to handle. Our analysis of industrial humanoid deployments at BMW and Toyota examines how this first wave is playing out on factory floors today.Commercial settings such as professional cleaning, healthcare and hospitality — tour-guide roles among them — are expected to follow as software improves and operators gain confidence.

Consumer adoption, in areas such as domestic cleaning and education, is likely to lag until costs fall and reliability is proven in harsher industrial environments. Bain foresees a hybrid future in which non-humanoid machines handle highly repetitive workflows, while humanoid robots act as “flexible generalists” alongside humans, who focus on strategic oversight such as planning, workflow orchestration and risk management.

Subscribe now

The world’s biggest technology companies have kicked off 2026 with announcements of ambitious plans to further develop AI for the physical world, and especially in the field of brains for robots and humanoids. Two more announcements that are noteworthy come from Microsoft, and AsiaInfo and ABB Robotics.

Microsoft has introduced Rho-alpha, a new robotics model that aims to bring the generative-AI revolution from text and images into the physical world. Built on the company’s Phi series of vision-language models, it is designed to translate natural-language instructions into control signals for dual-arm robots performing complex manipulation tasks.

Rho-alpha is described as a “vision-language-action-plus” system, adding tactile sensing to the usual camera and language inputs and with plans to incorporate force and other modalities. Microsoft is trialling the model on dual-arm and humanoid platforms and is inviting organisations to apply for an early-access programme, ahead of broader availability through its Foundry offering.

The launch underscores how big technology firms are racing to turn large AI models into “physical AI” platforms that can perceive, reason and act among humans, not just generate text or images. As labour markets tighten and factories, warehouses and hospitals hunt for productivity gains, adaptable robots that can learn from feedback and cope with messy real-world environments have become a strategic prize.

“Through these advancements, we aim to make physical systems more easily adaptable, viewing adaptability as a hallmark of intelligence,” Microsoft’s research team wrote. “We believe robots that can adapt more easily to dynamic situations and to human preferences will be more useful… and more trusted by the people who deploy and operate them.”

AsiaInfo, ABB Robotics team up

Meanwhile, AsiaInfo and ABB Robotics have launched a new “Embodied Intelligence Laboratory” in China, betting that “physical AI” will be the next frontier in industrial automation. The lab was formally inaugurated on Jan. 25 and is positioned as the practical outgrowth of the two companies’ existing strategic partnership.

Backed by Alibaba Cloud and Nvidia, the laboratory will blend AsiaInfo’s strengths in AI applications, 5G-A communications and cyber security with ABB’s know‑how in robot control. The partners plan to focus on industrial simulation and physical AI built on Alibaba Cloud’s vision‑language‑action large model and Nvidia’s simulation platform, targeting deployment in complex factory environments.

The tie‑up underlines how Chinese software, Western industrial automation and US chip design companies are converging around embodied intelligence as manufacturing digitises at speed. As factories push to cut labour costs and cope with ageing workforces, vendors from ABB to Alibaba and Nvidia are racing to turn generative and embodied AI into robots that can perceive, decide and act reliably on the shop floor.

“The Laboratory may serve as a key platform for their continuous enhancement in technological synergies and co-creation of application scenarios, with a view to advancing the implementation of the application scenarios of physical AI in manufacturing sector and setting an industry intelligence benchmark with global influence,” AsiaInfo’s board said.

Subscribe now

Humanoid, a London-based robotics startup, has run its first live factory trial with Siemens, deploying a wheeled humanoid robot to handle box-moving chores inside one of the German group’s electronics plants. Over a two-week proof of concept in Erlangen, the HMND 01 Alpha robot shuttled plastic totes from storage stacks to a conveyor, working alongside humans in regular production.

The exercise was designed to see whether a general-purpose humanoid can take on the unglamorous but necessary work of industrial logistics, rather than the carefully fenced-off duties of traditional industrial arms. In Siemens’ tote-to-conveyor “de-stacking” task, the robot repeatedly picked containers from a pile, trundled them across the shopfloor and deposited them at a handover point for human operators until the stack was empty.

Humanoid robots have become the latest obsession in factory automation, as manufacturers hunt for flexible machines that can slip into existing layouts and workflows rather than demand greenfield redesigns. From carmakers to e‑commerce giants, companies are testing legged and wheeled bipeds in warehouses and plants, hoping they can eventually tackle varied, dull and physically taxing jobs amid labour shortages and ageing workforces. Siemens’ trial suggests that, at least for tightly defined tasks, the technology is moving from lab demo to production line experiment.

“Our focus is on creating robots that deliver measurable value in real-world settings,” Artem Sokolov, the UK company’s founder and chief executive, said in a press release. “This joint PoC with Siemens showed clear potential for practical deployment of humanoid robots. We see them move steadily toward the real world, and partnerships like this one help accelerate that transition.”

The PoC unfolded in two phases, starting with in‑house development using a physical twin of the workcell to refine performance before shifting to the live factory. In Erlangen, HMND 01 ran autonomously for stretches of more than 30 minutes, handled two tote sizes and met targets including 60 tote moves per hour, uptime above eight hours and pick‑and‑place success rates exceeding 90 percent.

Both companies bill the trial as a beginning rather than a climax. They are weighing additional logistics use cases and a broader rollout across Siemens facilities, turning the Erlangen factory — described by the group as its “customer zero” — into a test bed for what may become a new class of industrial co‑workers.

Subscribe now

Arm Holdings, best known for low‑power smartphone cores, is reorganising itself around the robots of tomorrow. The British chip designer has carved out a new Physical AI division to sell semiconductors and software for machines that move through the world, from factory bots to autonomous cars, as it chases one of the most tantalising growth markets in technology.

The unit will sit alongside two existing lines — Cloud and AI, and Edge — and will fold in Arm’s automotive business, underscoring how closely carmakers’ needs now resemble those of robot-builders.

Executives disclosed the plan on the sidelines of CES in Las Vegas, where robotics and so‑called ‘physical AI’ systems are hogging the limelight. Drew Henry, who will lead the new division, argues that smarter machines could “fundamentally enhance labour” and eventually move the dial on global GDP, Reuters reports.

Arm plans to add specialised robotics staff, hoping that tighter alignment of automotive and robotics — both obsessed with safety, reliability and power budgets — will give its designs an edge over rivals.

The move reflects a broader turn in artificial intelligence from disembodied chatbots to systems that sense, decide and act in the physical world. At CES, Nvidia and others are promoting “physical AI” stacks and robot foundation models, betting that the next wave of demand for compute will come from warehouses, roads and homes rather than just cloud data centres.

As chipmakers chase this shift, they are jostling to supply the standard architectures for humanoids, industrial arms and autonomous vehicles, much as Arm’s blueprints already underpin most of the world’s smartphones.

For New York listed Arm, robots promise a way to diversify beyond handsets while preserving its reputation for frugal, deterministic compute. Chief executive Rene Haas has already pushed through price rises and floated plans to design more complete chips, and the new division gives investors a cleaner story about long-term growth in automotive and automation.

Yet competition will be fierce: Nvidia, Qualcomm and a swarm of startups also covet the sockets inside future robot fleets, where software ecosystems and safety certification may matter as much as clever silicon.

Subscribe now

Hyundai Motor Group used the glare of CES 2026 to recast itself not just as a carmaker dabbling in robots, but as an architect of a Physical AI economy built on humanoids, co-working machines and data-rich factories. Under the banner “Partnering Human Progress”, the South Korean group laid out a robotics strategy that stretches from industrial cobots on today’s shop floors to Atlas humanoids mass-produced by the end of the decade.

The initiative rests on three pillars: putting robots alongside workers in hazardous and repetitive tasks; wiring the group’s far-flung value chain into an end-to-end robotics platform; and enlisting leading AI labs to supply the brains for increasingly capable machines.

The company is also planning a Physical AI Application Centre and a dedicated robot manufacturing and foundry plant, signalling that robots are moving from side project to core industrial franchise.

The push comes as global manufacturers race to convert decades of automation into fleets of adaptive machines that can see, reason and learn from the physical world, rather than simply repeat pre-programmed motions. From smart factories in Asia to logistics hubs in America and Europe, a new contest is emerging over who can turn proprietary data, AI models and hardware scale into defensible positions in “physical AI” – the layer where software meets gears, actuators and balance.

For carmakers like Hyundai, the wager is that control of both the robots and the factories they inhabit will be as strategic in the 2030s as control of the combustion engine was in the 20th century.

“The convergence of robotics and AI represents more than a technological advancement. It is a transformative innovation that will make human life safer and more enriching,” said Zachary Jackowski, vice president and general manager of Atlas at Boston Dynamics, Hyundai’s robotics subsidiary. “By combining capabilities of Boston Dynamics and Google DeepMind through this strategic partnership, we are taking a significant step toward redefining the future paradigm of the industry.”

Hyundai has already put Boston Dynamics’ robots to work, turning the company’s once-viral machines into earnest factory hands. Spot, the four-legged inspection robot, now operates in more than 40 countries, collecting data and monitoring safety in industrial sites, while Stretch, a warehouse robot launched in 2023, has unloaded more than 20 million boxes under harsh conditions. Those deployments, Hyundai argues, are proof that its robots can graduate from spectacle to scaled infrastructure.

The more audacious bet lies with Atlas, the humanoid platform that Hyundai expects to become the largest segment of the Physical AI market. The group plans to integrate Atlas into plants such as Hyundai Motor Group Metaplant America in Savannah, Georgia, before rolling it across the network once safety and quality gains are proven.

From 2028 the humanoids are due to take on tasks like parts sequencing; by 2030 the remit extends to component assembly, repetitive motions and heavy lifting, with the long-term goal of saturating entire production sites.

To support that rollout, Hyundai is constructing what it calls a Group Value Network, effectively an end-to-end AI robotics value chain built on its automotive production infrastructure, safety know-how and affiliates’ technologies. Within this framework the company aims by 2028 to reach a scalable system capable of turning out around 30,000 robots a year, a volume it claims will put the production version of Atlas ahead of any other enterprise-grade humanoid. The same supply chain that underpins electric vehicles and batteries is meant to give the group a cost and integration edge in robots too.

The strategy also leans heavily on software, particularly AI models adapted from software-defined vehicles to robots and other physical AI products. By digitising data from manufacturing, logistics and sales, Hyundai wants a virtuous circle in which robots constantly refine their behaviour and, in turn, generate richer data for training. The Physical AI Application Centre is pitched as the hub for this loop, curating use cases and feeding improvements back into both product design and factory deployment.

Hyundai’s ambitions are reinforced by a high-profile alliance between Boston Dynamics and Google DeepMind to accelerate next-generation humanoid development. DeepMind’s Gemini Robotics foundation models, built on its large multimodal Gemini system, are designed to let robots perceive, reason, use tools and interact more naturally with humans, regardless of form factor. Together, the partners aim to develop AI controllers for complex robots performing high-impact tasks, while stressing that any scaling of robot adoption must be done safely and efficiently.

On the show floor in Las Vegas, Hyundai is keen to demonstrate that this is more than a slide-deck strategy. Its booth at the Las Vegas Convention Centre features live demonstrations of Spot, Stretch and Atlas alongside in-house creations such as the X-ble Shoulder, an automatic charging robot and MobED, a mobile eccentric droid that has already won a CES 2026 Best of Innovation award. Motional’s Ioniq 5 robotaxi and Hyundai WIA’s autonomous mobile, collaborative and parking robots round out a line-up meant to show robots embedded in both everyday life and industrial workflows.

The group’s messaging stresses “human-centred automation”, with people remaining in control even as robots take on more of the dirty, dull and dangerous work. In Hyundai’s telling, factory workers graduate from manual labour to training and supervising robotic colleagues, while the broader ecosystem shifts towards large-scale robot commercialisation across sectors beyond mobility. If the plan works, the company will not just sell cars and robotaxis, but lease out legions of embodied AI systems that move, lift and watch on behalf of clients worldwide.

Subscribe now

Nvidia is using this year’s CES to argue that artificial intelligence is finally ready to leave the chat window and inhabit the physical world. The chipmaker has unveiled a stack of “physical AI” models, simulation tools and edge-computing hardware, flanked by a parade of humanoids, surgical systems and mining trucks from partners betting that robots are about to go mainstream.

At the heart of the announcement is a family of open models branded Nvidia Cosmos and Isaac GR00T, designed to let robots see, reason and act with far less bespoke coding. Cosmos Transfer 2.5 provides customizable world models for synthetic data generation and policy evaluation in simulation, while Cosmos Predict 2.5 and Cosmos Reason 2 target vision-language and vision‑language‑action tasks, especially for humanoids that must coordinate full‑body movement in messy real-world settings.

Nvidia says partners such as Franka Robotics, NEURA Robotics, Humanoid and Salesforce are already using GR00T-enabled workflows to train and validate new robot behaviours, from industrial manipulation to video search and summarisation.

The push reflects a broader scramble among technology firms and manufacturers to extend generative AI from tokens to torque, fusing large models with sensors and actuators in factories, homes and hospitals. Nvidia’s proposition is that a unified stack—spanning open foundation models, cloud-native orchestration and edge modules built on its Blackwell architecture—can lower the cost of developing “generalist-specialist” robots that learn many tasks, rather than the single‑purpose machines that dominate shop floors today.

With robotics now the fastest-growing category on Hugging Face, and with open frameworks like LeRobot integrating Nvidia’s Isaac and GR00T tools, the company is positioning itself as the default infrastructure provider for an emerging ecosystem of AI-powered machines.

“The ChatGPT moment for robotics is here,” declared Jensen Huang, Nvidia’s founder and chief executive. “Breakthroughs in physical AI — models that understand the real world, reason and plan actions — are unlocking entirely new applications,” he said, pitching Nvidia’s combination of Jetson processors, CUDA software, Omniverse simulators and open models as the catalyst for partners “to transform industries with AI-driven robotics.”

To anchor those ambitions in silicon, Nvidia has launched the Jetson T4000, a $1,999 module at 1,000‑unit volumes that brings its Blackwell GPU architecture to autonomous machines with up to 1,200 FP4 teraflops of AI compute and 64GB of memory in a 70‑watt envelope, promising roughly four times the performance of the previous generation.

Jetson Thor and IGX Thor, aimed at humanoids and industrial edge systems respectively, are being adopted by developers such as Boston Dynamics, NEURA Robotics, LG Electronics, Caterpillar and Archer to boost navigation, manipulation and safety‑critical autonomy.

On the software side, Nvidia is releasing Isaac Lab‑Arena, an open-source framework for large‑scale robot policy evaluation linking into benchmarks like Libero and Robocasa, and OSMO, a cloud‑native orchestration tool that coordinates synthetic data generation, model training and software‑in‑the‑loop testing across on‑premise and cloud hardware.

The company’s collaboration with Hugging Face will see GR00T N models and Isaac Lab‑Arena folded into the LeRobot library, with humanoids such as Reachy 2 and the tabletop Reachy Mini certified to run Nvidia’s VLAs and large language models on Jetson Thor and DGX Spark. From LEM Surgical’s Dynamis system, which uses Cosmos Transfer and Jetson AGX Thor for autonomous surgical arms, to Caterpillar’s mining and construction fleets, the message from Las Vegas is that for Nvidia, the next big AI market is not another chatbot—but the robots rolling, walking and flying out into the world.

Subscribe now

Inbolt, a Detroit-based maker of vision-guidance systems for robots, has unveiled a new bin-picking technology that mimics human dexterity. By mounting a 3D camera directly on the robot’s arm and infusing it with proprietary artificial intelligence, the system allows machines to spot, seize and position parts amid the chaos of unstructured bins — where objects lie jumbled, obscured or shifting. This promises cycle times under one second per pick and success rates up to 95 percent in live automotive production.

The innovation upends conventional bin-picking setups, which depend on costly overhead cameras, intricate calibrations and rigid grasp points prone to failure if bins budge or parts evade detection. Inbolt’s on-arm approach generates endless grasping strategies in real time, analyses objects in-hand after gripping, and tweaks trajectories mid-motion for precise placement.

Already deployed in over five factories, it slashes hardware needs to one camera per robot, easing scalability across varied bins and sites. Powered by Nvidia hardware, the system thrives on minimal computation for robust performance across part shapes.

As manufacturers grapple with labour shortages and demands for round-the-clock output, Inbolt’s advance underscores a pivotal shift in industrial automation: from brittle, vision-limited robots to adaptable “physical AI” that rivals human improvisation.

Globally, bin-picking bottlenecks hobble assembly lines in automotive, electronics and logistics, costing billions in downtime; this on-arm AI could unlock flexible, low-cost autonomy, accelerating reshoring and countering rising wages in Asia. With firms like Tesla and Foxconn scaling humanoid robots, such breakthroughs may finally tame the “last metre” of unstructured manipulation, propelling factories towards true lights-out operation.

“Traditional bin picking systems are too rigid for real factory conditions,” says Albane Dersy, Inbolt’s chief operating officer. “We designed our solution to adapt in real time, able to see, grasp, and adjust the way a human would. That level of flexibility is what manufacturers need to reach truly autonomous production.”

Key perks include resilience to moving bins, affordability sans overhead rigs, and swift integration — heralding cheaper, nimbler robotics for an era of volatile supply chains.

Subscribe now

Mouser Electronics has begun shipping Infineon Technologies’ latest PSOC Edge machine-learning microcontrollers, aiming to bring more intelligence to devices at the periphery of the network in homes, factories and warehouses. The launch extends Mouser’s role as a gatekeeper for new chips, while giving Infineon another route into fast-growing edge-AI niches.

The PSOC Edge family combines an Arm Cortex-M55 core with an M33 coprocessor, Infineon’s NNLite accelerator and support for Helium DSP and Arm Ethos-U55, allowing neural networks to run continuously at low power inside a single, highly integrated chip. Target applications range from smart speakers and home appliances to collaborative robots, industrial controls and more sophisticated human–machine interfaces.

Infineon pairs the silicon with its ModusToolbox environment and DEEPCRAFT Studio AI, which offers pre-built models for voice, audio and vision, reducing the effort required to graft machine learning onto existing embedded designs.

The deal reflects a broader shift as chipmakers race to pull machine learning out of distant cloud data centres and into everyday objects. Edge-AI microcontrollers promise faster response times, lower energy use and better privacy, by processing data locally rather than streaming every sensor reading back to a server. With global electronics makers scrambling to differentiate commodity hardware through software and AI, distributors such as Mouser have become crucial channels for seeding new architectures among engineers and small design teams.

Mouser is backing the PSOC Edge rollout with two development kits that are meant to make experimentation easier. The KITPSE84AITOBO1 PSOC Edge E84 AI evaluation kit focuses on rapid prototyping of ML and edge-AI workloads using DEEPCRAFT Studio, while the KITPSE84EVALTOBO1 board adds a SODIMM system-on-module carrying the E84 MCU plus an AIROC Wi-Fi and Bluetooth combo, MIPI-DSI display support, audio interfaces, and Arduino and mikroBUS headers for expansion.

For Infineon, whose microcontrollers already populate industrial and IoT systems, the PSOC Edge line is pitched as a way to hardwire machine learning into the next generation of “always-on” smart devices.

Subscribe now

Galbot, a Chinese maker of humanoid robots, has raised more than 300 million dollars in fresh capital, lifting its total funding to 800 million dollars and valuing the company at $3 billion. The round, which the company says is the largest single financing in the embodied AI sector, further entrenches its position in a field that blends large AI models with general‑purpose robots.

The money comes from investors in China, Singapore and the Middle East, signalling that capital allocators across Asia and the Gulf remain keen on hardware-heavy bets so long as they come wrapped in an artificial-intelligence story. Galbot, based in Beijing with research centres in Shenzhen, Suzhou and Hong Kong, claims to be the first company to build, in-house, the full stack of high-quality datasets, embodied foundation models and robotic hardware for its humanoid systems. Its flagship Galbot G1 robot is already deployed in manufacturing, logistics, retail and healthcare, with more than a year of continuous real‑world operation behind it.

For global investors, the deal underscores how humanoid robots have become the next contested terrain in the wider AI arms race, drawing comparisons with US efforts such as Tesla’s Optimus and a crop of American and European startups. As Chinese companies double down on “embodied intelligence” to offset labour shortages and to automate factories, warehouses and service work, the combination of proprietary data, large models and vertically integrated hardware is emerging as a favoured strategy. Galbot’s $3 billion price tag suggests that, in China at least, humanoid robotics is shifting from speculative prototype to an investable industrial platform.

“Galbot is the world’s first company to fully achieve full-stack in-house development across hundreds-of-billion-scale high-quality datasets, embodied foundation models, and robotic hardware,” the company said. “These innovations have laid a solid technological foundation for the large-scale deployment of humanoid robots.”

The company says it already has paying customers. In industrial manufacturing, it has signed up CATL, Bosch, Toyota and Hyundai, and claims to be the first company to deploy humanoid robots in real autonomous operations on factory floors, with orders for thousands of units. In smart-city services, its Galbot Store — an autonomous retail outlet run by G1 robots — operates in more than 30 cities, while logistics sites use its machines around the clock and hospitals such as Xuanwu deploy them in wards, pharmacies and for way finding.

With the new capital, Galbot plans to accelerate research, scale deployments and expand overseas, pitching itself as a standard-bearer for China’s bid to lead in general-purpose robotics. The company says it aims to “advance embodied AI technology” and speed up the industrial rollout of autonomous humanoids worldwide, a promise that will now be tested in export markets and in the unforgiving economics of full-scale automation.

Subscribe now

Aptiv, an industrial-technology supplier best known for its automotive electronics, is teaming up with Vecna Robotics to build a new generation of autonomous mobile robots (AMRs) for warehouses and factories, betting that better perception and orchestration software can squeeze more productivity out of increasingly automated supply chains.

The collaboration will combine Aptiv’s sensing, compute and embedded software with Vecna’s AI-driven material-handling platform in a joint solution pitched as safer, smarter and cheaper than today’s systems.​

Under the deal, Aptiv’s perception portfolio — including its PULSE surround-view camera and ultrashort-range radar — will be integrated into Vecna’s AMRs, alongside machine-learning models for real-time perception and dynamic path planning in busy, mixed-traffic environments. Vecna will contribute its autonomy stack and CaseFlow workflow tools, which coordinate humans, forklifts and robots, while Aptiv supplies the underlying compute platform based on VxWorks and Wind River’s Helix virtualization technology. The companies say the result will be more flexible deployments that avoid the fixed infrastructure and brittle layouts common in older warehouse-automation projects.​

Warehouse robotics has become a test-bed for the convergence of automotive-grade sensing, cloud-scale computing and generative AI, as logistics operators attempt to cope with volatile e-commerce demand, labour shortages and rising capital costs. Global spending on automation has accelerated since the pandemic, but many facilities still rely on islands of semi-automated equipment that struggle to interoperate, making orchestration software just as important as the robots themselves.

As rivals from startups to industrial conglomerates race to offer “autonomy-as-a-service”, suppliers that can bundle perception, safety and workflow optimisation into a single stack hope to lock in long-term platform relationships with fleet operators.​

“Automation is transforming the way goods move through warehouses and factories, with devices that sense, think and act in real time,” said Javed Khan, executive vice-president for intelligent systems at Aptiv. “Together, we’re making automation safer, smarter, and more affordable—helping customers meet the demands of modern logistics.”​

Vecna’s chief executive, Karl Iagnemma, said the tie-up would yield “more robust and cost-optimized robotic products” that cut operating costs and improve throughput while avoiding major infrastructure upgrades. Both firms pitch their approach as a way to scale deployments quickly, with Vecna’s Pivotal orchestration platform adapting to changing workflows and Aptiv’s virtualised compute designed to support different robot configurations over time.

Aptiv plans to showcase Vecna’s CPJ co-bot pallet jack at CES 2026, signalling that the robots are meant as much for marketing its software-and-sensing stack as for hauling pallets around draughty sheds.​

Subscribe now

Microscopic robots the size of single-celled organisms are edging closer to practical use, as researchers demonstrate cell-powered machines that can sense their surroundings, run simple programs and change course without human supervision. Built using standard chip-manufacturing techniques, the devices suggest that autonomy at the microscale may soon be as much an information-processing challenge as a mechanical one.​

Engineers at the University of Pennsylvania and the University of Michigan have built sub-millimetre robots — roughly 210 by 340 by 50 micrometres, comparable to a paramecium — that integrate photovoltaic power, temperature sensors, memory, a processor, optical communications and electrokinetic actuators on a single chip. Each robot can be mass-produced lithographically, with yields above 50 percent, and in principle could be manufactured for about a cent apiece at scale.​

For four decades, micro-robotics has promised swarms of machines swimming through blood vessels, patrolling industrial process lines or mapping microscopic ecosystems, but has been held back by the inability to shrink computation, power and locomotion into volumes below a cubic millimetre. Most existing micro-robots are either tethered to bulky external control systems, or hard-wired for a handful of fixed behaviours, limiting deployment beyond carefully controlled laboratories.

By moving a fully fledged, if tiny, computer onto the robot itself and powering it with light, the new work points to a future in which general-purpose micro-robots can be cheaply programmed and reprogrammed to carry out diverse tasks in messy environments — from targeted drug delivery to nano-manufacturing — without custom hardware each time.​

“By moving computation to the micro-robot, we reduce both the cost and operational overhead to a bare minimum, paving a path to widespread adoption,” the authors write, arguing that versatile, penny-priced machines could ultimately find uses from microsurgery to studies of living systems.​

To squeeze autonomy into a power budget of about 100 nanowatts, the team designed a custom complex-instruction-set processor in a 55-nanometre CMOS process, paired with a few hundred bits of onboard memory. Specialised commands such as “sense temperature” or “move for N cycles” compress dozens of low-level operations into single instructions, allowing the robots to execute useful behaviours despite their tiny program space.​

Immersed in fluid, the bots propel themselves by driving current between platinum electrodes, generating electrokinetic flows that translate and rotate the chip at speeds of around 3–5 micrometres per second in straight motion and up to 0.3 degrees per second when turning. In thermal-gradient experiments, the robots measured local temperature, compared it with prior readings and switched between arcing searches and on-the-spot rotations to climb toward warmer regions, reporting temperatures with a resolution of about 0.3°C in less than 1 cubic millimetre — performance that Pareto-dominates current miniature digital thermometers.

Subscribe now

Mercado Libre, Latin America’s largest e-commerce and fintech platform, has struck a commercial agreement with Agility Robotics to deploy Digit, a humanoid robot, in its fulfillment operations in San Antonio, Texas, with an eye to expanding across the region. The robot will first handle warehouse tasks that support order fulfillment, before the partners broaden its use to other logistics workflows in Mercado Libre’s network.

The companies say the rollout is designed to address roles that are hard to staff, especially repetitive and physically demanding tasks in warehouses. By shifting those jobs to robots, Mercado Libre expects to improve ergonomic safety, reduce labour gaps and lift productivity, allowing employees to move into work that adds more value.

The tie-up underlines how global e-commerce and logistics operators are moving from pilots to commercial deployment of humanoid robots as labour markets tighten and service expectations rise. Agility’s Digit, which has already moved more than 100,000 totes in live commerce operations, is pitched as a way to slot automation into existing facilities without major redesigns, while its cloud platform, Agility Arc, links fleets of robots to warehouse-management systems, conveyors and mobile robots.

For Mercado Libre, which operates in 18 Latin American countries, humanoids offer a route to scale logistics capacity without matching it one-for-one with headcount.

“At Mercado Libre, we are constantly exploring how emerging technologies can elevate our operations and improve the experience for our employees and millions of users,” said Agustin Costa, the firm’s senior vice-president of shipping. “Our partnership with Agility Robotics and the deployment of Digit in our facilities is a significant step forward in our vision to create a safer, more efficient, and adaptable logistics network.”

Digit is a human-scale robot that walks, lifts and moves materials and totes through existing aisles and workflows, designed to be dropped into standard warehouses without costly infrastructure changes. Agility says Digit can learn new tasks and adapt to changing workflows, filling high-turnover roles and keeping goods moving through continuous, repeatable processes.

The deal adds Mercado Libre to a client list that includes GXO, Schaeffler and Amazon, all experimenting with humanoids in industrial settings. Daniel Diez, Agility’s chief business officer, said the company is “incredibly proud” to support Mercado Libre’s workforce and operations and argued that autonomous humanoid robots are now capable of “performing meaningful work and delivering real value” in commercial facilities.

Subscribe now

A new market study, from Market.Us, suggests that self-reconfigurable robots are shifting from research curiosities to a serious pillar of next-generation automation, with revenue projected to grow from $1.31 billion in 2024 to about $7.05 billion by 2034 at an annual clip of 18.3 percent. The numbers hint at more than a niche: modular robots are being framed as a way to hedge against technological and demand uncertainty in factories, labs and even space programmes.

The first signal is who is buying. Education and research account for more than half of current demand, reflecting the technology’s roots in universities and publicly funded labs, yet the report points to rising use in industrial automation, inspection and hazardous environments. That mix matters: academic deployment underwrites the core hardware and algorithms, while early industrial pilots test business cases around multi-task platforms, higher asset utilisation and lower maintenance downtime.

The second signal is where the power sits in the value chain. Homogeneous modular robots command around 71.8 power of the market, suggesting that users still prefer simple, repeatable building blocks over exotic heterogeneous systems. Chain-based architectures, with 45.3 percent share, map neatly onto near-term applications such as climbing, confined-space inspection and loco-manipulation tasks that conventional arms or mobile bases struggle to handle.

Geography reinforces the pattern. North America holds just over 48 percent of global revenue, buoyed by research-heavy ecosystems and collaborations such as HEBI Robotics’ modular “inchworm” systems and Boston Dynamics’ work with Toyota Research Institute on large behaviour models for adaptive manipulation.

For now, that concentration keeps technical know-how and early standards formation in a handful of hubs, even as Asia and Europe build their own programmes in space exploration, manufacturing and education.

Strategically, the most important claims are economic and organisational rather than technological. The report cites productivity gains of 15–25 percent in multi-task environments and the ability to keep operating after losing up to a quarter of the modules, positioning modular platforms as a hedge against both downtime and design lock-in.

Robots-as-a-service models from companies such as KUKA, together with NASA-backed modular blocks from HEBI and furniture-building swarms from Roombots, frame reconfigurability as a service layer rather than a one-off capital purchase.

Plenty could still slow the transition. The need for robust connectors, power transfer and coordination pushes up costs, and the lack of common module and protocol standards threatens fragmentation. Yet as distributed control, docking mechanisms and AI planning improve, self-reconfigurable robots look less like an esoteric research topic and more like an operating system for physical work in factories, cities and off-world projects alike.

Subscribe now

SoftBank and Nvidia’s plan, as reported by Reuters first, to pour more than $1 billion into Skild AI, at a mooted $14 billion valuation, underlines how quickly capital is shifting from text models to AI that controls machines in the physical world. If completed, the deal would also cement Skild as one of the most richly valued robotics-software bets of this AI cycle.

What Skild represents

Skild, founded in 2023 by former Meta AI researchers and backed early by Amazon and Lightspeed, is building a hardware-agnostic “brain” for robots rather than making robots itself. Its foundation model is trained to generalise across robot types and tasks, from warehouse work to household chores, positioning the firm as an operating layer for “embodied AI” rather than a single-use application.

The company has already raised $300 million in a Series A and $500 million in a Series B this year; a new round would almost triple its last $4.7 billion valuation, unusually fast even by AI-startup standards. That escalation reflects investor expectations that a small number of platforms will dominate training data, model performance and developer ecosystems in robotics.

Why SoftBank and Nvidia care

For SoftBank, the talks are part of Masayoshi Son’s attempt to rebuild an AI and robotics thesis after the bruising Vision Fund years, including a recent $5.4 billion deal for ABB’s robotics business. SoftBank has been an investor in Skild since at least its Series A, and the latest round would deepen exposure to software “brains” that can sit atop its growing stable of hardware assets.

Nvidia already supplies Skild’s GPU infrastructure and has been expanding from chips into full AI stacks, including robotics simulation and software tools. A larger equity stake would help Nvidia defend its platform position if robotics workloads become a major source of demand for its next generation of accelerators.

Signals for the robotics race

The size and speed of the round mark embodied AI as the next arena for platform battles, alongside foundation models for language and images. It also raises the bar for rivals such as Tesla’s Optimus and Figure AI, which are tying software more tightly to proprietary humanoid hardware.

As one person familiar with the talks told Reuters, the deal would be “one of the largest private financings in the robotics AI sector to date,” underscoring how capital is clustering around a few perceived winners. Whether Skild’s universal brain can move from pilots to large-scale deployment will now be watched as a bellwether for the commercial readiness of general-purpose robots.

Subscribe now