The new era of Robotics and the Future of Industrial Automation
The industrial world is at an inflexion point.
Embodied AI, robotics powered by advanced perception, reasoning and autonomous action has achieved technical maturity that was unthinkable from even 5 years ago. The latest humanoids signals that advancement - Figure AI's Figure 03 humanoid robot or the latest NEO robot from 1X, which were both recently released in October 2025 can now load your washer and fold your laundry.
While we are not here to talk about humanoids, they are a useful proxy: what looks like magic in the consumer domain is already bleeding into industrial automation. The same advances in perception, motion planning, simulation and large-scale training that power these humanoids are now shaping the next generation of industrial robots.
And yet despite the maturity of the technology, robots aren’t everywhere yet.
Why aren’t robots everywhere yet?
Data bottlenecks and Moravec’s Paradox: Robots excel at tasks humans find hard i.e. doing a backflip but struggle with “easy” human tasks requiring common sense, fine motor skills or multitasking. Unlike LLMs which can train on trillions of freely available text tokens, robots require embodied, real-world interaction data. Collecting this data is expensive, slow and fundamentally limited. Factories are full of variability and edge cases which makes it really hard for models to generalise. As a result, simulation-trained robots often fail when deployed in messy and unpredictable environments.
Integration challenges: Factories are built on decades of layered equipment - legacy PLCs (Programmable Logic Controllers), MES/ERP software, operator routines etc. None of these systems speak the same language. Deploying a robot is rarely plug-and-play and so each deployment feels more like a bespoke engineering project.
Economic and ROI barriers: Manufacturers don’t ask, “Can the robot do the task?”. They often ask “What is the payback and how does this scale beyond a pilot?” The first robot often triggers a realisation: to make automation worth it, you need operator retraining and workflow redesign or even additional robots. And so the ROI moves from unit-based to system-based which is way harder to justify from most manufacturers’ point of view. Trust and downtime risk remain major hurdles as manufacturers need near-perfect reliability because downtime is costly.
So why now?
AI breakthroughs and talent inflow:
There has been a noticeable shift in top AI talent leaving large labs to launch robotics startups, mirroring the dynamics we saw during the LLM breakthrough era. In the same way that BERT, GPT-3 and the “scaling laws” moment catalysed a wave of foundation-model startups, robotics is experiencing its own inflection driven by the emergence of generalist datasets and models.
Academic and commercial research in robotics is accelerating rapidly, with a tenfold increase in published papers since 2020 - 2024 alone saw more robotics research than the entire decade of the 2000s.
A major catalyst has been Open X-Embodiment (O-XE), a massive cross-robot dataset and training framework released in late 2023 by Google DeepMind, Stanford, Columbia and 30+ robotics labs. O-XE effectively unified robotic data across dozens of robot types, enabling the training of generalist models that can transfer across hardware. This did for robotics what large-scale corporate did for LLMs.
It is no coincidence that many of the researchers behind O-XE and successor models (RT-2, RT-X) left to start companies such as Skild AI and Physical Intelligence, building “robotic brains” that were simply not feasible before datasets of this scale and diversity existed. The talent inflow is a strong signal that robotics is entering its own foundation-model moment.
2. Falling cost curves for hardware
The unit economics for industrial robots are shifting sharply.
The average price of traditional industrial robots ranges from $25,000 to $100,000, depending on functionality, payload and complexity. However, rapid advances in China’s robotics manufacturing ecosystem, particularly from firms like Estun, Efort and Elite Robots has accelerated a wave of low-cost, high-performance six-axis arms. These companies benefit from vertically integrated supply chains, locally produced servos and controllers and increasingly sophisticated AI-vision modules. The result is a generation of robots priced far below Western incumbents, expanding global supply and forcing a repricing across the industry. For example, the IFR Robotics Statistics report quotes that the average price of newly installed industrial robots in Asia is at around $16,000 in 2024 vs roughly $46,000 for Europe.
Edge compute has also undergone a step-change in affordability and performance. Hardware such as NVIDIA Jetson modules, Qualcomm RB-series chips and various ARM-based accelerators has experienced dramatic reductions in cost per TOPS (Tera Operations Per Second) over the past five years. Compute that once required a $3,000 industrial PC can now be delivered by a $300 - $700 edge module with onboard GPU acceleration. This collapse in compute cost means that advanced perception, vision inference and motion planning can run directly on-board the robot, eliminating the need for expensive centralised controllers and reducing total system cost.
At the same time, modular designs, open-source hardware and increased competition are expanding supply and driving economies of scale. Vision and tactile sensors have become more compact and cheaper while commodity actuators now deliver performance levels that previously required custom engineering.
These trends collectively mean that while it remains difficult for Western companies to win on hardware alone, the overall cost to build and scale versatile robots is falling quickly.
3. Demographic/Labor economics are forcing automation
In advanced manufacturing economies, labour shortages are increasingly acute. A recent survey by the UKG Workforce Institute found that 61% of U.S. manufacturers struggle to fill critical roles. This is very much a problem across the US, Europe, Japan and China.
The problem is mainly driven and compounded by an aging workforce, skills mismatches driven by technological change and the declining appeal of manufacturing roles for younger generations. In practice, this means it takes months to hire an operator, putting SME manufacturers that are too small for traditional automation projects at risk of not being able to fulfil existing contracts or grow as fast as they could. Combined with geopolitical reshoring and “just-in-case” production strategies, these gaps are pushing manufacturers toward automation as an operational necessity.
Where we are looking at
Historically, investing in robotics has been challenging. This is because compared to SaaS, robotics demanded heavy upfront capex, long R&D cycles and complex GTM pathways dominated by integrators.
But the playbook is changing as laid out above. Rapid advances in edge compute, commoditised hardware (from motors to arms) and increasingly powerful AI foundation models are collapsing the cost curve and unlocking software-driven robotics. Much of the “hard” part of robotics like perception, planning, manipulation is being abstracted into reusable models. And as more software-centric plays enter industrial automation, the potential for higher multiples for exits increases.
We expect this to play out across multiple domains:
1. Intelligence and autonomy layer
Due to the data bottlenecks mentioned above, this is the highest-leverage layer of modern robotics stack: The foundational models, data pipelines and autonomy infrastructure that make robots adaptable rather than scripted. This layer includes include VLA, motion planning, Simulation, Sim2Real, Reinforcement learning (RL), teleoperation, tactile sensing, agentic AI, no code/natural language processing, datastack, analytics engine etc.
2. Operating system (OS) / Middleware and enabling infrastructure layer
This is the connective tissue that sits between robots and enterprise systems which enables connectivity, orchestration and integration. Industrial automation today is still dominated by PLCs (Programmable Logic Controllers), fieldbuses and integrators which is what makes robotics deployment slow, expensive and non-scalable. This layer includes solutions building in fleet management, orchestration, factory integration middleware layers, edge computing platform.
3. Applications and vRaaS models
These are execution-layer robotics: full-stack, verticalised systems that solve specific, high-value factory problems. By contrast, general-purpose robots tend to generate more hype than revenue. They require long R&D cycles, are usually heavy capex and their go-to-market remains unclear - all of which make them challenging bets for venture returns. Vertical applications here includes robotics automating electronic manufacturing, PCB testing, general assembly etc.
If you’re building or investing in this space, please reach out, I would love to chat!