Robotics Landscape

The robotics space is evolving rapidly, with conversations currently revolving around data quality, humanoids versus specialized robots, and where value accrues in the ecosystem. This week, we want to take a 5,000-ft view of the space, identify the major categories of the robotics landscape, provide some preliminary views on value accrual to those categories, and identify the major roadblocks the industry will likely face moving forward.

The Robotics Stack

Hardware & Components: These are the physical components that make up the robot. This could include everything from actuators to chips to the nuts and bolts that keep everything together. While there are exceptions to the rule, this is the part of the stack that is most likely to be commoditized. Often, value lies not in the components themselves but in the expertise required to pull them together into a functioning, useful artifact. However, recent supply chain constraints within the industry, with a revised focus on U.S.-sourced parts specifically in the defense industry, have opened up opportunities for tech-enabled manufacturers to carve out a niche of this enormous market.

Robot Platforms & Form Factor: This bucket refers to the robots themselves, some taking humanoid form while others take the form of specialized robots. We have written about where the industry stands on this bifurcation, and the answer is likely somewhere in the middle. In the near term, it is very hard to separate the hardware from the end application, and there is likely a segment of skills that will never be separated from the hardware form factor due to their complexity and edge cases. We think value accrual in the near term (next 3-5 years) will be for specialized form factors, and in the long term (10+ years) will be for generalized platforms like humanoids.

Developer Tools & Middleware (Includes Simulation & Digital Twin): This bucket includes everything from hardware-focused IDEs to the Robot Operating System (ROS) open-source software commonly used to build in robotics. Konvoy has spent a large amount of time in and around the DevOps space and we believe that the qualities that accrue value in the software development lifecycle are likely to translate here. Specifically, platforms that are deeply embedded into developer workflows (creating data or governance moats, and network effects) will increase switching costs.

Perception & Sensing: This is how a robot sees and feels its surroundings (the cameras, LiDAR, radar, IMU, gyroscope, accelerometer, and touch sensors that take in information, plus the software that turns that information into a picture the robot can act on). Whether value accrues here is unclear because this layer is getting squeezed from both sides. From below, the sensors themselves are getting cheap, so the hardware is commoditizing just like the components bucket. From above, the AI world models are ingesting and processing this data at rapid rates. These models now know how to go straight from raw camera input to action in one step.

AI & Autonomy Software: This part of the stack is arguably the hottest right now for fundraising, with companies building world models that understand physics well enough to let robots interact with the physical world without explicit instructions. Many companies are attempting to build out datasets for these models, training on specific tasks. Others are creating their own models, and the jury is still out on whether or not they will trend towards commoditization in the same way that many LLMs are doing today.

Application Layer: These are the tasks that the robots will inevitably complete and, as mentioned above, are currently difficult to separate from the platforms themselves. Over time, we expect a marketplace of robot apps to be downloadable for generalized platforms (like humanoids), but this will be limited to skills with limited edge cases where failure is not catastrophically detrimental or where specialized hardware is required.

Fleet Management & Orchestration: Lastly, how you manage fleets of robots, monitor their health, update them, or manage their location is equivalent to the observability platforms like Datadog in DevOps. These platforms are capable of accruing enormous value.

The Roadblocks

Robotics startups have raised $18.8 billion in 2026, compared to $15 billion in the full year of 2025, and while a significant amount of talent and capital is being dedicated to this industry, there are still major hurdles to overcome before robots are widespread.

1. Moravec's paradox: the things humans find effortless (grasping, balancing, perceiving) are the hardest to automate. Moravec argued this was because the skills that we find intuitive like walking or finger dexterity are things that took ages to perfect.
2. The long tail: a robot can be 95% reliable in a demo and still useless, because of the last 5% of edge cases.
3. Hardware iteration is slow and expensive: you cannot ship a hardware fix overnight.
4. Sim-to-real gap: what works in simulation rarely transfers cleanly. This is because simulations are composed of hundreds of assumptions and oversimplifications.
5. Integration tax: a working robot is mechanical + electrical + perception + control + software, and the system is only as good as its weakest subsystem.
6. Unit economics: hardware margins are inherently worse than infinitely scalable software. Additionally, many of the data capture techniques and implementations can require serious white-glove service for initial implementation and beyond. The most scalable companies will need to find other ways to monetize beyond just the sale of the robot.

The Paths Forward

So how does the industry move forward while navigating each of these roadblocks? There seem to be three paths:

1. Focus on easy, replicable tasks: These tasks could be low or high value but will likely lack defensibility, given that there is no value that comes from additional data to solve edge cases because of the simplicity of the task. For example a robot arm stacks uniform boxes off a conveyor onto a pallet in a fixed pattern which has minimal variation or edge cases in a static environment.
2. Augment tasks and learn over time: We are already seeing this approach with different types of tele-operations, where humans are controlling robots in the near-term with the intention of moving towards full autonomy. Kodama (a Konvoy portfolio company) is an example of this in the logging industry. We are also seeing another approach where employees are given XR headsets that monitor their movements, capturing data with the intention of creating robots that can execute those tasks. This segment likely has the highest defensibility. Data captured early improves product quality, allowing for customer expansion and the capture of more data to solve for edge cases.
3. Augment human behavior: Some tasks may be so complex or dangerous, or the edge cases too vast, to rely on full autonomy. In those cases, it may make sense to switch to focus on augmenting the humans instead of replacing them entirely. We are already seeing this with certain highly skilled labor roles, such as surgery.

From an investment perspective, the challenge for each of these buckets is determining what is able to scale at a clip that is capable of large-scale returns. Unlike software, we are now forced to navigate complex manufacturing processes, bill of materials risks, and global supply chain concerns in a way that never existed when software was at the core of the market.

Takeaway: The robotics landscape continues to evolve, and there is a tension between hardware and software. While there are segments of the stack that will likely be pure software plays, much of the innovation is taking place in the messier middle ground. Data moats, switching costs, and network effects, which are the cornerstones of any large business that accrue value over time, are being derived from the physical world. That is what makes this market harder than the software cycles that came before it. Defensibility still exists, but it will have to be earned through mastering manufacturing and supply chains. We are excited to meet the teams, figuring out how to build those moats.

‍