Over the last 25 years, three main hardware platform shifts have defined an era of tech innovation.
Personal computers led to the democratization of compute access from large corporations and their mainframes to the individual. This spurred the creation of software programs designed specifically for consumers and kicked off a microprocessor war to build smaller and smaller processors to support smaller computers for consumers. It also led to the creation of Graphical User Interfaces (GUIs) as software use wasn’t limited to technical users.
Mobile phones changed how we interact with the internet. Access to the internet was no longer limited to a static location. It also led to the cottage industry of app developers which could leverage Apple and Android’s app stores to reach billions of people. By circumventing the need for companies to spend massively upfront on distribution, it opened the door for who could build apps.
Cloud-based technology would not be possible without upgrades to internet bandwidth capabilities, semiconductors, and distributed systems / grid computing. Cloud products allowed for continuous updates and cheaper distribution, enabling the types of margins that VCs have come to invest behind today. It also allowed software companies to outsource the overhead of maintaining their own servers, meaning they could invest in their products more.
We are entering a fourth era. Cheaper and higher-quality hardware components, advances across the traditional hardware and robotics (defined broadly) stack, and new software breakthroughs in transformer models are enabling robots to become the next computing platform. This new platform shift creates two main opportunities:
Hardware and robotics development will eventually look like software development - iterative, fast and highly customizable. To enable this change, there will need to be a suite of tools that support the design, testing, and creation of new pieces of hardware.
Accessing high-quality sensing and imaging data will become commoditized. This creates two main opportunities. (1) Companies providing this data will need to offer insights on top of data collection, and as insights become commoditized, offer actions and services. They may build these new features in-house or partner with software companies that can decrease their time to market. (2) New vertical SaaS companies will be built using the commoditized data as their “system of record”. Some of these companies will offer auxiliary services leveraging the data collection tools - drones, cobots, and cameras, on top of the core workflows.
Trends
Several key trends make now an opportune time to build in these two areas. They are split between macro and technical.
Macro
Global onshoring due to energy and industrial resilience / independence concerns.
Desire to compete with China’s cheap exports.
Shrinking industrial labor force in developing countries → need for greater output / productivity per worker.
The wars in Ukraine and Gaza have shown how much modern warfare has changed and how much defense suppliers need to change as well. There is a clear need for rapid iteration within a tech-enabled battlefield. Primes are not fit to be the sole suppliers of this new battlefield.
*Parts of traditional SaaS are becoming saturated as areas for greenfield development shrink and gains over incumbents become more incremental.
Micro (Technical)
Drones have gotten incredibly cheap and high-quality.
Huge jumps in OCR (specifically convolutional neural networks and vision transformers) + CV tech (specifically segmentation and pose estimation) make understanding optical data much easier. Pile on LLMs that can both be fed instructions and action upon the cleaned-up optical data much easier.
Low-Earth Orbit (LEO) satellites have increased from 500 → 8k in the last 25 years. By 2030 there are projected to be tens of thousands. The price to launch one of these satellites has steadily decreased while streaming data back to Earth at low latencies has markedly improved as well
Price, availability, and capability of cloud compute have gone way down.
LiDAR and other sensors have become more robust in detecting objects in more granularity, at greater distances, and in more adverse conditions. Edge computing and new advanced chips have made in-memory compute for complex tasks possible. This further expands what actions devices cameras or sensors can perform without an internet connection.
Improvements in actuator and motor tech have made robots and machines much more agile and able to perform a broader array of precise tasks, narrowing the gap between what humans and robots can do.
CNC machinery has gotten better internet connectivity along with a more flexible 5-axis motion that allows them to be orchestrated into more customized processes.
Additive manufacturing and 3D printing have rapidly increased and enabled the production of complex parts with more customization and faster turnaround times.
You can order a customized anodized piece of sheet metal for $35 in 3 days!
Tying it all together
Parts of the ecosystems I have described are already becoming fairly well built out, with multiple strong companies in them. As an investor, it’s my job to think about what a different future looks like and who stands to benefit from it. Below are a few predictions:
Hardware development begins to look like software development
A tooling ecosystem, similar to JIRA, IDEs, and testing suites for SWEs, will arise for mechanical, simulation, and manufacturing engineers. Intuitive tools, some with Co-Pilot features, will replace the stagnant on-prem CAD, CAE, and CAM tools that exist today. Some will operate as point solutions, and others will combine CAD, CAE, and CAM functions along with other features like project management and issue resolution. Importantly, these will be cloud-based tools that don’t have the same localized compute restraints that current tools have. Further down the road, they will be able to suggest changes such as replacing one material for another because it can survive higher temperatures or changing the placement of a bolt to maximize durability. This new tooling ecosystem will cause material organizational changes as engineering roles won’t be so strictly dictated by the software that they use. There will be less Mechcanical Enginer vs. Manufacturing Engineer vs. Simulation Engineer, and more general Engineer roles.
Another part of the developer tooling ecosystem that exists to support software development is data logging and observability. Robots, drones, and cameras can now generate (and with the right pipelines) send reams of data back to cloud servers. Companies today have tools that provide them with observability, but the next-gen of these products will be able to sift through all of the data and attribute events to consumable issues using NLP. Similar to how NewRelic is able to identify a runtime issue when a job has been sitting in the queue forever, these observability tools will be able to interpret sensor data and provide digestible context to it. An interesting part of this will be solving for what are the robot’s responsibilities to identify what it should vs. should not be doing and then translating that from sensor data.
Multimodal generative models + additive manufacturing will materially decrease the time it takes to design and prototype hardware in industries in which these steps were the biggest bottlenecks. Testing these new pieces of hardware in different environments can be enabled virtually and across a wide number of edge cases as generative models can create synthetic testing scenarios.
Some companies will rebuild manufacturing setups from the ground up. 5-axis CNC machinery and the “done-in-one” production process will become staples for leading manufacturers. CNC machinery and additive manufacturing will be built with better internet connectivity, allowing for these machines to be synced using a central OS (or even next-gen CAD tools). New Shop Floor Management tools, which will look similar to Warehouse Management Systems, will crop up to orchestrate human workflows within manufacturing shops, allowing companies to scale more efficiently. There will be an ecosystem of robotics and machine system aggregators that will make integrating these devices as simple as a few API commands.
From 500 LEO satellites in orbit 10 years ago to tens of thousands by 2027, we will have unprecedented visual and sensory coverage of the Earth. This presents a multitude of opportunities for software to build on top of this architecture. (1) We will need dedicated CI / CD solutions that can send new software and models to run on these satellites along with multi-tenant cloud solutions that can run multiple bespoke models for different customers. (2) There will need to be compression software that can help developers optimize their models to run on edge on the satellites and stream data back. (3) There will undoubtedly be excess capacity for these satellites across all the providers, and so similar to GPU aggregators, there will be satellite aggregators that by connecting to each specific LEO provider can aggregate coverage of a desired area and re-sell to end customers. (4) Whether it’s those same aggregators or someone else, people will build marketplaces that house apps that can be run on satellites for different use cases.
Collecting data becomes commoditized, leads to new SaaS
Autonomous drones, enabled by swarm technology, will be paired with software platforms to monitor large areas for hyper-specific risks. They will then communicate back to cloud servers who will use generative ML models to turn these risks into either a) workflows for humans or b) workflows for the drones. Larger drones with higher-resolution cameras and sensors will act as orchestrators who can better identify issues that smaller worker-bee drones can investigate further or act on.
Low-earth-orbit satellites + advances in mesh technology will enable sub-cm GPS. This in turn will allow developers to build a whole set of better software and hardware experiences. More realistic AR experiences, safer autonomous vehicles, and more automated EV charging payment experiences will be possible with this more precise location tech. There will also be a cat-and-mouse game of building spoofing / jamming resistant GNSS services.
Manufacturing, industrial companies, and civil services companies will not only catch defects with their products as they happen but will proactively identify the issues that led to those defects and alleviate them. These fixes will either be piped into existing task management systems or communicated to dedicated robotics who can fix them. In situations where there are multiple fixes, they will provide cost estimates and potential future damage estimates to help companies prioritize which to address first.
Increase in drone + LEO + OSINT data will result in companies / governments being able to access near real-time visuals in many parts of the world. Multi-modal search will make combing through this data possible. Advances in satellite tech like on-orbit refueling, and highly efficient electric satellite thrusters, along with the sheer number of LEO satellites will make it easier than ever to monitor certain areas. This will result in countries and companies protecting sensitive areas with tools that can disable or blind satellites from viewing a certain area.
The airbase / airport of the future will be autonomous or hybrid autonomous / manual. This will require an operating system that can handle traditional scheduling and workflow activities, as well as one that can connect with autonomous vehicles to appropriately route and task them. As transformer models continued to be used to improve weather forecasting, these airbase operating systems could automatically suggest and implement route changes to aircraft and supporting ground vehicles.
If you are building towards a future where hardware creation looks more like software - iterative and with a large cast of supporting tools, I want to talk to you! If you are building a platform that leverages data from robotics, IOT sensors, CV-enabled cameras, drones, or any other manner of hardware, I want to talk to you! My email is nickb at northzone.com.