TL;DR

We are backing teams building in the physical world

“Software is eating the world” – Marc Andreesen (2011). Yes, it has (maybe more than we needed it to in some cases). Yes, it will continue to – but we believe physical machines will be its next (and most important) distribution engine. Over the last few decades, SaaS has automated the lives of desk workers. That transformation is now spilling out into the real world.

Why this matters now

For years there has been hype surrounding IoT and robotics, with many empty promises. So why is this time any different? We believe the stack has been rebuilt and the infrastructure is now in place. First-gen “smart” devices were fragmented, cloud-dependent, lacked security, and couldn’t close the loop between sensing and acting. Today’s world offers edge AI, improved models, standardized (and cheap) hardware, and ubiquitous connectivity to fill those gaps, letting devices operate autonomously and coordinate as fleets.

Where we’re focused

• Autonomous robotics: Task specific bots replacing scarce labor in ag, construction, logistics, and more
• Physical security automation: Cameras, access, and alarms that detect and respond without human guards
• Asset inspection & infrastructure visibility: Drones, rovers, and mounted vision systems turning power lines, pipes, and bridges into self reporting networks
• Vertical orchestration platforms: Software “command centers” that manage updates, routing, and analytics across thousands of deployed devices

Why are we so excited?

• New budget pools: These products compete with labor spend, not IT line items – unlocking 5 – 10x larger TAMs
• Compounding data moats: More deployments = more data = better performance = more deployments. Hardware flywheels look a lot like the best SaaS loops
• The next iconic platforms: The sheer magnitude of greenfield (in some cases actual greenfield) creates a foundation for generational companies to emerge. Early-stage companies who develop an early data advantage can evolve exponentially, allowing them to compete with incumbents

Introduction

In recent decades, tech innovation has been dominated by tools that automate knowledge work. From CRMs to ERPs and every workflow tool in between, we’ve seen the daily work of desk jobs digitized, delegated, and streamlined. Software ate the world; some would argue more than we needed it to. Now, that transformation is spilling out into the physical world.

During those years, most investors avoided backing hardware-centric companies due to the capital intensive, lower-margin nature of these businesses. While others continue to shy away from investments that do not follow the traditional SaaS playbook, we see this as an opportunity to generate alpha for our investors while backing the teams building the future of automation.

The opportunity ahead is not to optimize workflows inside a browser tab but rather to automate workflows on farms, in warehouses, on job sites, and throughout the physical economy. For years, digital infrastructure outpaced physical infrastructure. But as hardware becomes better and smarter through the advent of more powerful AI, and connectivity becomes ubiquitous, that gap is closing. The result is a new generation of full-stack technology companies building software-enabled machines that solve real-world problems. This shift will be just as consequential as the SaaS explosion that came before it, but instead of driving productivity from behind a screen, it will reshape the physical environments where work actually happens.

Why IoT 1.0 Fell Flat

Before we get to what’s working now, it’s important to understand what didn’t work the first time around. The term “Internet of Things” was coined to describe the idea that everyday physical objects – from thermostats to tractors – could be connected to the internet and made smarter through data. That vision caught on quickly. By the early 2010s, nearly every major electronics company was promising a “smart” version of its core products. But the reality rarely lived up to the hype.

Fragmentation was the first major problem. Devices were built with proprietary stacks, meaning that even within the same home or building, a user might have to manage five different portals to control five different “smart” tools. There was limited interoperability, lack of a shared standard, and no incentive for vendors to cooperate. Industrial IoT suffered the same fate, with vendors offering isolated solutions that didn’t integrate into broader systems.

Latency and cloud dependence caused additional issues. Devices relied on round-trips to remote servers for basic functionality. A sensor might detect something, but the decision to act on that data would be made in the cloud, introducing delays. This architecture couldn’t support real-time control which is essential for anything safety-critical or time-sensitive.

Security and maintenance also created friction. Many IoT devices were difficult to update, and some lacked basic encryption or identity management. That led to highly publicized attacks like the Mirai botnet, which exploited thousands of insecure cameras and routers. Over time, companies grew wary of deploying devices that became liabilities rather than assets.

Finally, and perhaps most importantly, there was no closed loop. These devices could sense the environment but couldn’t act on it. They could collect data but couldn’t close the loop to drive outcomes. As a result, many became data silos that fed dashboards but didn’t deliver value.

Why Now: The Stack Is Ready

We are now at an inflection point from a technological and macro standpoint. Many of the technological limitations that constrained the last generation of physical automation have been resolved. What’s emerged is a radically more capable technology stack that combines the affordability of modern hardware with the intelligence of software and the reliability of enterprise-grade infrastructure.

Edge AI and computer vision are at the center of this shift. It’s no longer necessary to ship data to the cloud and wait for a response. Thanks to advancements in chips like NVIDIA Jetson and Apple’s Neural Engine, complex machine learning models can now run directly on low-power edge devices. These models can interpret video feeds, detect anomalies, and make decisions locally, with millisecond latency. This allows machines to operate in dynamic, unstructured environments (fields, factories, city streets, etc.) where they must sense and react in real time.

Low-latency connectivity via 5G, Starlink and distributed edge compute further unlocks real-time control. For example, Sub-20ms connections via 5G or Starlink let a remote operator pause an autonomous excavator or reroute a drone mid-flight as quickly as if they were on-site. At the same time, micro data centers sitting in a cell tower can fuse each robot’s sensor data, push fresh maps, and broadcast new models in real-time. The result: fleets of machines now see the same environment, share learnings instantly, and act as a coordinated system rather than isolated devices.

Hardware has also become cheaper, more modular, and easier to deploy. Sensors that once cost hundreds now cost tens. Open-source frameworks and prototyping kits allow teams to go from design to field testing in weeks instead of months. And because hardware components are increasingly standardized, integration risk is lower. The era of custom-built, single-purpose machines is giving way to general-purpose platforms that can be updated and repurposed via software.

All of this is happening against a backdrop of powerful macro tailwinds. Labor markets are tightening, particularly in industries that rely on physical labor. Wages are rising, labor participation is falling, and companies are struggling to fill roles in agriculture, logistics, manufacturing, and construction. At the same time, geopolitical instability and reshoring efforts are forcing companies to bring production closer to home – often to higher-cost labor markets. In this environment, automation becomes more of a strategic necessity than a cost saving mechanism.

Where We’re Focused

Innovation is happening faster than ever before, which can make it difficult at times to assess persistence. As such, we are focused on hardware form factors where software creates a disproportionate leap in value. These are categories where the physical hardware is already commoditized or well understood, but the addition of AI, orchestration, and connectivity creates exponential value.

1. Autonomous Robotics

In December 2023, we made a bet based on labor challenges in agriculture. While available farm labor declined by 20% over the prior two decades, we did not see incumbent OEMs (John Deere, CAT, etc.) making the necessary investments in autonomy to address this issue. After getting introduced to the team at Burro, it became clear that their advantage in mobility and autonomy would enable long-term differentiation and strategic value, regardless of which direction hardware capabilities/preferences trend towards. Burro’s autonomous robots augment workforces, enabling farms and nurseries to improve yields while solving for labor shortages and wage pressure. As we evaluate similar issues across industries such as manufacturing, construction, and logistics, we’re excited about companies that develop an early software edge, and compound that advantage through performance flywheels.

Side note: Humanoid robots are attracting a lot of attention (and capital) right now. It’s pretty evident that these devices will work out in the long run, if you assume any rate of improvement in balance, actuation, and large action models. However, we don’t think they will be ubiquitous for all use cases (i.e. in farming, a tractor is more effective than a human, and an autonomous tractor will be more effective than an autonomous human). We think in most settings, task-specific machines will outperform, cost less, and scale faster.

Examples Across other Industries:

• Corvus Robotics: Deploys unmanned warehouse drones that scan and track inventory with higher speed and accuracy than manual approaches.
• Dusty Robotics: Automates the layout process in construction, utilizing robots to print digital building plans directly onto jobsite floors
• Gather AI: Layers advanced computer vision software onto off-the-shelf hardware devices (drones/cameras) to digitize warehouses, automating materials handling and inventory management processes
• Shinkei Systems: Radically improves the quality and efficiency of our seafood supply chain using robotics to automate traditional techniques and humanize the fish harvesting process

In all of these examples, the value isn’t just in the hardware. It’s in the software; localization algorithms, real-time mapping, fleet coordination, and learning loops that improve performance with each deployment.

2. Physical Security Automation

Physical security is undergoing a shift from manual oversight to autonomous response systems. What was once reliant on guards, routine patrols, and reactive video review is becoming a closed-loop, intelligent network of machines that detect, decide, and act without human intervention.

Cameras don’t just record anymore; they identify weapons, intrusions, and unsafe behavior in real time. Doors can now adapt dynamically based on context, identity, or protocol violations. Alarms no longer wait for human triggers – they engage proactively with voice prompts or environmental cues. A new generation of intelligent hardware systems are making our spaces safer by eliminating human involvement.

This transformation is particularly acute in industries where safety, compliance, and operational uptime are non-negotiable – like construction, education, logistics, and public infrastructure.

Example Companies:

• Dragonfruit: Retrofits existing cameras with agentic software to automate asset protection for retailers. The software is camera agnostic
• Lumana: Transforms traditional IP cameras into intelligent agents. Allows them to understand, detect and act on critical events and incidents

3. Asset Inspection & Infrastructure Visibility

Critical infrastructure (power lines, oil pipelines, bridges) is hard to monitor, expensive to maintain, and dangerous to inspect. Traditional approaches involve manual patrols, helicopters, or site visits. AI and computer vision are changing that.

Modern inspection tools use drones, fixed sensors, and mobile platforms to continuously capture data about the physical state of infrastructure. Machine learning models can then detect anomalies, predict failures, and guide preventative maintenance.

Example Companies:

• Buzz Solutions: Automates the analysis of power line inspections using AI to detect faults, corrosion, and vegetation encroachment.
• Noteworthy AI: Mounts edge‑AI smart cameras on everyday utility fleet vehicles to automatically geolocate poles, inventory components, and flag defects in real‑time

This is one of the most immediate applications of software-enabled hardware, delivering clear cost savings and risk reduction in asset-heavy industries. We believe that the companies that turn towers, pipes, and bridges into self-scanning, self-reporting networks, will lock up maintenance budgets and data advantages for decades.

4. Verticalized Orchestration Platforms

Orchestration platforms serve as the connective tissue for distributed fleets, enabling operators to monitor performance, push updates, coordinate behavior, and analyze outcomes in real time.

These platforms are what make intelligent hardware scalable. Without them, deploying dozens (or thousands) of machines across job sites, warehouses, or facilities would be operationally infeasible. While large horizontal orchestration platforms like AWS IoT and Azure Digital Twins do have natural insertion points into companies (because of their cloud offerings), we believe that verticalized solutions that are specialized, edge-first, and cloud-agnostic can create a wedge for themselves.

Example Companies:

• InOrbit: Offers a spatial intelligence platform that serves as the centralized command center for optimizing a fleet of autonomous robots in warehouse settings
• BrightAI: Powers the future of critical infrastructure through a centralized operating system (Stateful OS) and multimodal sensors – which together enable asset & site visibility, autonomous drone inspection, and workforce wearables / copilots.

Why Investors Should Pay Attention

This wave of physical automation is not a niche trend. It’s a new computing paradigm – one that takes everything we learned from enterprise software and applies it to the physical economy. The impact could be just as large.

• TAM Explosion:
  • Tapping into new budgets: These companies won’t be competing for traditional software spend – they will unlock entirely new budgets by automating roles historically filled by people. Security personnel, warehouse staff, and field technicians are labor-intensive functions with high recurring costs. By replacing or augmenting humans with software-driven machines, these platforms now compete for compensation budgets which are often 5 – 10x larger than software line items.
  • Automating the great outdoors: Indoor (warehouse) automation has been prevalent for years. Continued advancements in computer vision will open up huge markets outside of the confines of factory walls. In theory, this should grow TAM linearly with the number of square feet that machines are able to automate.
• Learning Curves: Hardware performance will compound. More deployments mean more data. More data means better performance. Better performance leads to more deployments. The companies that get there first will enjoy defensibility that resembles the best SaaS flywheels.
• Historic Parallels: When computing left the lab and entered the office, we got SAP and Microsoft. When it left the office and entered the consumer world, we got Google and Apple. Today, it’s leaving the screen and entering the physical world. The companies that define this era could be just as iconic.

The last decade was about automating screens. The next one will be about automating the world around us. This time, it won’t be about dashboards and data collection. It will be about systems that see, decide, and act – at scale, in real time, in the real world. For entrepreneurs, this is one of the richest seams of opportunity. For investors, it may be the defining theme of the next decade. Now is the time to start paying attention.