Solving Hard(ware) Problems
…or why this software investor has a crush on hardware
It’s hard to find a VC who doesn’t love software, and for good reason.
For one, you don’t need much to get started. There are no supply chains to sort out, no factories, no purchasing huge amounts of inventory up front…in short, nothing slow or expensive enough to keep you from getting off the ground. Especially in the age of the cloud and AI copilots, you’re only ever a few clicks, prompts, and copy-pastes away from a functional skeleton of a product.
Software is also a lovely thing to scale. SaaS companies in particular have high gross margins (no IT admin) and recurring revenues (subscriptions), a combination which means they can re-invest a substantial piece of their revenue into (hopefully) predictable growth.
It’s not very surprising then that we’ve built, invested in, and bought a ton of software over the last few decades. Particularly in the 15 years or so since the global financial crisis, macro tailwinds like smartphone adoption and the maturing of cloud infrastructure have made it easier than ever to build and distribute software, and we’ve collectively obliged. In 2023, it’s difficult to find an industry completely untouched by software or to find someone who doesn’t recognize at least a handful of shiny B2B SaaS logos.
But what about hardware? I think that outside of oohing and ahhing at our bezel-less iPhones, most of us in software tend to forget that hardware even exists. We use software that runs in the cloud on top of data from other software that runs in the cloud…easy! You’d be forgiven for thinking that all of software runs in an actual cloud with no physical link to the rest of the world, which DALL-E tells me would look like this:
As a general rule of thumb, VCs are mostly afraid of anything involving hardware today, at least more than they were in the last century. After all, the first VCs got their start funding grease-melting guns, generators, and radioactive isotopes before spending a few decades concentrating mostly on semiconductors. I can already hear Christoph and the rest of our software-loving P9 team telling me that they would have preferred to invest in SaaS if it had existed, and he’s probably right.
Still, it’s clear that things are different today. The numbers tell us that the overwhelming majority of VC dollars go to pure software companies. More intangibly, my impression is also that most investors shut down at the first mention of “hardware” and immediately associate it with “bad,” regardless of whether the company is actually making hardware or using it underneath a software products.
In many cases, I understand. Some pure hardware businesses have capex requirements that would break a typical venture fund’s model. But in most other cases, this self-imposed allergy is perhaps a bit too harsh.
Hard problems = hard ware
So why am I so interested in hardware now?
For one, we know that despite traditional VC wisdom that hardware is too hard, there’s plenty of evidence that solving big problems that involve hardware in some way can lead to really big outcomes. This tweet sums it up nicely: 8 out of 10 of the largest public companies by market cap rely on hardware in some way.
More importantly, DALL-E’s cloud picture above is clearly wrong. Something, somewhere grounds your favorite CRM software or delivery app (or the data they rely on) to the physical world with hardware, and in many cases even makes that software possible in the first place. Performance improvements in chips and memory let us build increasingly demanding systems. Network infrastructure makes it possible to move data around at the speeds required to deliver low/zero latency user experiences. Sensors gather the image, speed, location, temperature, and other data we often need to make software useful.
In other words, hardware is almost always in the picture. Particularly when we zoom out on the big questions we look set to tackle in the next decade of tech, almost all of them seem to involve hardware in a big way. Here I’m thinking about:
- Clean energy production
- EVs and mobility
- Agriculture and food security
- Techbio and data-driven healthcare
- Safe and efficient manufacturing
- Defense and collective security
- Space
These are all hard problems to solve, and it’s unlikely that we’ll have them squared away come 2030. Still, there does seem to be a sense of urgency attached to most of them thanks to climate change, a global pandemic, war, an energy crisis, and other painful phenomena, and that combined with a general feeling of “these things are important” leads me to believe that we will accelerate our progress on them in the coming years.
In any case, meaningful progress on any of them involves going much lower than the cloud to manufacture, measure, transport, cultivate, etc before we can even start to build software on top. I like this take on it: software may have eaten the world, but hardware will play a big role in “saving it.”
So with the macro for hardware and hardware-enabled software looking pretty good, it seems natural to me that we should be paying more attention to how to make it less…hard. The starting point for that is understand what problems people actually face with hardware today. I’m no expert, but to me this includes things like:
For people producing hardware:
- Organizing interdisciplinary engineering teams that often do hardware, firmware, embedded software, and everything in-between
- Providing the tools that allow developers to run their software on your hardware
- Meeting strict security and compliance requirements
- And many others that I’m unaware of :)
For people building software that interfaces with hardware in some way:
- Building and maintaining integrations and data infrastructure to be able to read data from physical devices
- Centralizing data from physical devices into a central control plane/software layer for visibility
- Cleaning this data, making sense of it, and delivering insights from it
- Optimizing software to be able to get the most out of a given type of hardware
- etc
While these two buckets of people face very different challenges, I put them all in the same camp of facing “hardware meets software” challenges that I think are interesting opportunities to go after in the next decade of tech.
Take AI as the first example of this: transformers were so ̶t̶r̶a̶n̶s̶f̶o̶r̶m̶a̶t̶i̶v̶e̶ disruptive largely because of the insight that we could make better use of the parallelization capabilities of modern GPUs to train models on huge quantities of unlabelled data. In other words, hardware know-how enabled a software breakthrough, and in this case the dominant player in the industry happens to be a company that owns not only the hardware (GPUs) but also the software interface for controlling it (CUDA).
We also already see a new generation of hardware meets software players that look like they’ll bring down barriers to entry for companies looking to tackle hardware-intensive problems:
- SpaceX is radically reducing launch costs for satellites
- Varda is making it possible to manufacture things in space
- Hadrian is reshaping the defense industry with outsourced component manufacturing;
- And probably many more to come
I don’t have a crystal clear thesis on the right business models or strategies for success that can be applied to all of the above, but there are a handful of areas I’m excited to dig deeper into.
Hardware meets software…where?
I’m paying close attention to 4 themes in this space, which I broadly categorize into 2 buckets:
Software for hardware and low-level teams
- Engineering OS and design
- Embedded systems tooling
Software with hardware knowhow
- Small AI
- SaaS and data infrastructure for physical assets
Before diving deeper, here’s quick overview of the companies on my radar thus far:
Software for hardware and low-level teams
Someone has to build the hardware in the first place. And here there is so much software to be built to help these teams catch up to the pure software world. During my first internship in VC I worked with a few others on Partech’s investment in Memfault, and I saved this screenshot from our memo of a decades-old connector hat firmware engineers still use to debug devices:
The bottom line: go ask anyone working on hardware, firmware, embedded systems, or broadly anything lower than application software, and you’ll find that tooling is ancient. Decades of innovation up the stack seem to have largely left these teams behind, and there are at least a few sub-categories where modern players can capture the $Bs spent on legacy or in-house solutions:
1/ Hardware engineering OS and design
Iteration and velocity are challenging when you have interlinked hardware and software problems. Changing your supplier for a physical component can impact the behavior of other components, the whole system, and the software on top. Designing components or even the whole device often requires you to mix and match specialized tools made decades ago. New companies in the space can unseat legacy requirements management software and design tools to give mixed discipline teams an OS for building the next generation of products.
Companies I like in this space:
Engineering OS:
Design:
- Flux (PCB design in the browser)
2/ Embedded systems tooling (devtools, observability, security)
as you keep trickling down the stack, someone, somewhere is writing code that actually controls physical hardware or runs lightweight software in very resource constrained contexts (e.g on IoT devices and very small chips). These teams have special development flows, security concerns, observability needs, etc. that most often can’t be met by copy pasting tooling from up the stack. The number of consumer electronics, appliances, and medical devices collecting sensitive data (and the regulatory and compliance scrutiny we apply to how they do it) is exploding, and I expect to see more software companies tackling these issues in the coming years.
Companies I like in this space:
- Memfault (IoT reliability platform)
- Bugprove (firmware security)
- Notch (drivers for firmware)
- Embedd (ML-generated drivers)
- Luos (microservices architecture for electronics)
Software with hardware knowhow
These software products bridge the hardware-software gap and are typically built by teams with a deep understanding of how specific hardware works.
3/ Small AI
Over the last decade we’ve seen companies develop a lot of know-how in designing systems to combine edge and centralized processing for AI-powered systems, such as in computer vision (hello, Intenseye). As we now look to apply the latest advances in AI to new use cases where (i) compliance and security constraints and (ii) latency issues make sending data elsewhere challenging, I expect we will iterate very quickly on how to move some/all of our models to the devices themselves.
Companies I like in this space:
- Ggml.ai (bring large models to commodity hardware)
- Plumerai (embedded ML toolkit, mostly video/detection today)
4/ SaaS and data infrastructure for physical assets
P9 alumnus Clément has written and spoken extensively about this. He says that becoming the system of record for assets and infrastructure by collecting data from the physical world can be very strategic for software providers. I agree :)
In some cases, even if software companies aren’t collecting data themselves, they can radically improve the efficiency of solving physical infrastructure problems by organizing the workflow around everything from design to production to permitting to accounting related to physical infrastructure.
On the data infrastructure side, there’s also a lot of work to be done in different industries to allow companies to efficiently ingest data from hardware and even deploy their software to edge devices.
As a fund we’ve invested in many companies applying software to physical problems and interfacing with hardware, most recently in energy with companies like Ampere, Kerith, and Solarize in energy.
This is a broad category where it’s tough to have a thesis on everything. I’m particularly interested in 2 sub-segments:
Companies that use software to tackle the design of complex (often physical) infrastructure, like:
- Continuum Industries (electricity transmission, water pipes)
- Hamina (wireless networks)
Companies that improve data infrastructure at the edge, like:
Opportunities
All the concerns people typically have about hardware are valid, but it’s not all gloom and doom. In fact, for companies building software combined with/enabled by/distributed alongside hardware, there are a few interesting advantages to think about:
- Defensibility and lock-in: the know-how required to either procure or produce and then extract insights from hardware is significant, and adds an extra layer of complexity to would-be challengers. Also, in cases where players own both the hardware and software layers (e.g. NVIDIA), it’s incredibly difficult for new entrants to compete (because they have to build the hardware and entire ecosystem of tooling on top of it) or for companies to consider moving to a different platform.
- High switching costs: the investments and effort that go into installing hardware somewhere where it can be useful for software (on a lamppost or a satellite, in a field or a hospital, inside a car, etc) make that software stickier. Replacing or removing software often means paying to remove hardware, which changes the equation for customers.
- Distribution and upsell opportunities: Software vendors that dip a foot in the hardware world can often enter with one or the other (or via the distributors of one or the other) and upsell later.
Did I miss something?
Do you have thoughts on emerging “hardware meets software” trends or are you building a company you think is relevant here (even if it doesn’t fit squarely into one of the above)? If so, I’d love to hear from you: alex@pointnine.com
