What is Hard Tech?
Deep Tech vs Hard Tech vs Software
TL;DR
With peak SaaS now in the rear view mirror and many software/SaaS companies struggling to fill in to their very high later stage valuations, Hardware is suddenly cool again. Hard tech has emerged as a category which heavily overlaps with deep tech, and even to some extent with software with an emphasis on hardware and manufacturing.
What’s historically been seen as attractive about hard tech companies has been that while they are high on technical risk the high barriers to entry for new players, often make for low market risk - which means they have sticky products with very deep and wide moats to defend their market positions against competition. However, not all hard tech companies are created equal: many are actually much lower on the technical risk front but that means they may come with significant market risk the founders and investors don’t appreciate full out the gate.
Investors should avoid falling into the oversimplistic heuristic of “hard tech = technical risk, software = market risk” and instead try to really understand what each individual companies risk profile looks like.
Founders should also appreciate that just because they are “hard tech” doesn’t mean their isn’t market risk and that many of the same forces - networking effects, minimum viable customer base, CAC costs, etc. apply to them as well. At a minimum, the incumbent solution (as clunky as it is) must be overcome at scale for the business to flourish.
Perkin’s Law
One of the more frequently thrown around terms when discussing hard tech is Perkin’s law. One of the early tenets of tech investment was a quip from Tom Perkins (of Kleiner Perkins fame) that “Market risk is inversely proportional to technical risk”- later known as Perkins’ law.
“Market risk is inversely proportional to technical risk” - Perkin’s law
The Early Days of Silicon Valley: Technical risk and hardware dominate
Early success stories in VC like Intel, Genentech, 3COM, Cisco, Apple, etc. were triumphs of technology who dug robust technology moats that made them de facto monopolies in their verticals. Their competitors took many years to overcome the moats, which gave them tremendous valuation multiples. The DoD dominated as a customer in the early days of Silicon Valley - this made the market a one dimensional technical problem (generally shove as much processing into as small a space as possible) to understand and solve for. Market risks, where they existed, were mostly in adjacent markets or from competitors attacking the low margin parts of their business (eg- Intel’s battle in the 80s with the Japanese and Koreans over the memory market) rather than head to head competition for market share: a la Uber and Lyft, or AWS and Google Cloud.
The era of market risk: desktop software to the internet to SaaS
In the last twenty five years of the .com, social, mobile and SaaS investment waves, start ups focused on business models with less to minimal technical risk, but lots of market risk. Terms like “networking effects” and “blitzscaling” became popular as software companies expanded into new markets mostly based on the same fundamental technologies with minimal new tech development required. Cloud computing/SaaS, opensource, object-oriented programming, low code/no-code and other paradigms minimized the amount of true technological development (ie R&D) required to deploy and maintain new products. This isn’t to underplay the role of software engineering: there is tons of innovation in these companies. But you didn’t need to invent a new paradigm of processing or cross a new feature size node of lithography to add a like button to Facebook - it was a software integration exercise (I realize that I’m being somewhat glib here, but I think everyone would admit that a like button is easier to implement than going from 30 to 16 nm feature size on an IC) and a market adoption problem.
With these extremely low technical barriers to entry, the name of the game was having a clever go to market and overcoming the “cold-start problem” so that your network becomes sufficiently sized reach escape velocity. This meant expanding your user base as quickly as possible, deploying quickly and building a codebase with new features, rather than achieving some technological breakthrough that could only come through heavy experimentation and developing new core technologies.
As you can see from Figure 2, for much of the last 20 years of silicon valley, the focus was on new categories and business models enabled by software, rather than purely new technology, especially in the hardware domain. Metcalfe’s law, coined by Robert Metcalfe who founded 3com and invented ethernet, underscored the importance of these network effects: “the value of a network is equal to the square of its number of connected nodes”.
Metcalfe’s law, that “the value of a network is equal to the square of its number of connected nodes” was in the forefront of many of the 2000-2020 era go to market plans. Market risk considerations dominated over technical risks
Today: Past Peak SaaS and Hardware is back in vogue
Fast forward to today and its generally accepted that we are past peak SaaS and that endless capital no longer exists to Blitzscale new software businesses ad infinitum. With higher interest rates we’ve seen higher interest in building real world things that are harder to make but come with stickier customers and more enduring value.
Fortuitously this coincides with some truly new technologies like AGI, nanomaterials, autonomous systems breaking out of the laboratory and being much close to commercialization. Despite the high capital requirements of many of these ambitious products, early stage hard tech and deep tech is now seen as the hot investment by investors who have been burned by Web3, Crypto and with SaaS investments that are coming up short in the growth stage as interest rates and cost conscience CFOs clip their valuations.
However, not all hardware and deep tech companies are created equal. There exists a tremendous opportunity to build new hardware companies based less on deep technical development, but more on new business models. Just as software can be truly deep tech and involve tremendous research and development to achieve amazing new technological breakthroughs (I’m looking at you OpenAI) vs companies with virtually no tech development and 100% market risk (food order management), the same exists in the hardware paradigm. There are categories of hardware companies that are very necessary to build which come with shorter R&D timelines than what you would think of as a tech company. They lean more towards the market risk of Perkin’s law with perhaps one exception: it’s much harder to walk away from a hardware product (try trading in a new car for a different one) than it is to unsubscribe from a SaaS product.
What is Hard Tech?
I think it’s important to define what hard tech is and what it isn’t.
I’m pretty intuitive: I define hard tech as industries that are developing, building and selling actual hardware, or (arguably) software that supports the development and building of actual hardware (note that selling isn’t included in there, because CRMs are not hard tech). Hard tech has some overlap with, but is differentiated from deep tech, which includes some pure software plays like AGI and Blockchain technology. It also includes some biotech/medical sub-categories which I don’t include because they are really in an entirely different vertical. Deep tech also includes a lot of technologies that are low TRL, with some understanding that the core problem they are seeking to solve may still involve a lot of science/research (as opposed to development) with the expectation that they are many years away from go to market. The best examples of this are things like Fusion, which even the fusion fanboys will tell you is at least 10 years away from commercialization.
I’ve made an attempt to rack and stack the categories in the Venn Diagram in Figure 3 below (thanks to Liz Stein from US Innovative Technologies Fund (USIT) for input here). I realize that in doing so, I’ll probably elicit a lot of opinions/comments about “you forgot me” so please feel free to pipe up in the comments and I can keep updating this. I think this is important as we try to understand the boundaries of this new investor space.
Hard tech vs hard tech adjacent software
Four categories on the edge of hard tech and software in the diagram above that I believe are worth mentioning: Software for Manufacturing, Software for Hardware, Modeling & Simulation (often described using the overused buzzword of “digital twin”) and Cybersecurity focused on protecting hardware. All of these software centric companies deserve to be thrown in with hardware in the hard tech category, because they have expertise and go to market plans that require significant understanding of the hardware and hardware market.
Some examples of Software for Manufacturing include: First Resonance and Epsilon3 (which I am an investor in) and Duro. These are generally tools designed to handle some portion of the product lifecycle management (PLM). Historically these tools have been glorified aides for mostly flat files (pdf, word, even excel spreadsheets) that are hard to maintain and disconnected. First Resonance’s Ion software handles planning and supply chain and makes it much more dynamic. Epsilon3’s test management suite makes traditional flat file test procedures as editable and manageable as software maintained in Github, including ways to handle the data generated from those tests more effectively. Duro labs makes a product meant to help with the CAD design and drawing management process called Duro One. These products aim to shake up and streamline an outdated set of products many hardware companies are still using purely out of inertia (more on this later).
Software for Hardware involves software focused on unlocking new capabilities in hardware. Some examples include: my 2nd company Spartan which makes a signal processing suite called Clarify which enhances automotive radar resolution and performance through embedded software. Companies like Light.io (now defunct) which tried to extract 3D ranging data from camera data for autonomous applications. Another good example from the automotive space is apex.ai which builds an operating system for automotive processors and sensors. All these firms involve large amounts of embedded software (as opposed to cloud) and oftentimes have to be managed within the quality specs of larger companies that they are suppliers to, such as AS9100, ASPICE and CMMI. An intimate understanding of hardware, embedded programming and the physical world and how hardware interacts with it is integral to these companies and their technical differentiators.
Modeling & Simulation - sometimes called “digital twins” or “multi-physics simulations” is another software but hardware adjacent category. These are companies which aide design and test for complex hardware systems through new software tools and technology that reduce prototyping cost and speed or unlock new insights. Some examples include former Defense Undersecretary Dr. Will Roper’s company Ishtari which makes digital twins of jet engines and other complex systems. Multi-physics simulations company Pasteur Labs is using neural nets to speed up computationally taxing simulations. Finally some companies like my first company Epirus has a “digital twin” product they offer as part of their Smartpower suite which allows them to simulate amplifier behavior with unprecedented accuracy, enabling significantly higher levels of control.
Finally, in the software adjacent to hardware categories are cybersecurity companies focused on protecting hardware. One example of this is Galvanick, led by my friend Josh Steinman, which seeks to protect SCADA (that’s microcontrollers and other logic units that manage infrastructure systems) from cyberattack. The most famous incident of this sort was the Stuxnet virus, which crippled Iranian centrifuges in the late 2000s only to then escape and cripple thousands more systems worldwide before Kaspersky Lab’s discovered it and added it to their anti-virus software suite. As the number of these devices proliferates and IoT devices quickly exceed world population (I saw one estimate putting the number of IoT devices worldwide at 1 trillion by 2030), software intended to protect these devices will become just that much more critical.
Roots of hard tech software in SpaceX Warp Drive
Many of the “software for manufacturing” category companies, particularly in South Bay, are founded by former SpaceX engineers who had exposure to Warp Drive.
One of SpaceX’s differentiators from incumbent aerospace players was its embrace of software as not just a product, but a way to engineer and manage how you build the product. Historically, incumbent aerospace companies leaned on a jurassic era set of software tools with names like PDM or VM (some still do) or worse yet internal toolsets that they didn’t allocate sufficient capital to modernize. None of these tools talked to each other, they were hard to use, harder to search and were barely one step removed from electronic filing cabinets. In short, they sucked (and in many places, still do).
Warp Drive was the name used internally by SpaceX for the proprietary ERP system they developed which gave them a leaps and bounds increase in productivity vs traditional aerospace contractors. SpaceX started with a blank sheet of paper, hired a bunch of really sharp SaaS software folks to build them a supercharged toolset and changed all that — and most of the legacy contractors have been playing catch up ever since.
Now, all those x-SpaceXers are brewing their own versions of Warp Speed that are more narrowly tailored to their own particular specialties but more generalized in design to bolt on to many different product company models besides just building rockets and satellites (though it can do those too). They are then turning around and selling them to other SpaceX alums starting rocket companies, car companies and other things. This is just one example of how having a CentiCorn like SpaceX with hundreds of alumni founding new companies after leaving creates fertile soil for new product development.
Hard Tech vs Deep Tech
What differentiates hard tech from deep tech is really the amount of outright scientific development and engineering required to make the product a reality. Typically if major parts of your system are in the Orange part of Figure 4 (TRL 1-3), you have significant R&D, not just NRE ahead of you to get to a minimum viable product and you are almost certainly deep tech.
For Deep tech companies the go to market is seemingly pretty simple (or so the seed stage investors would like to believe, growth stage types know it never is): invent a product no one else has even come close to in terms of capability and convince the public its safe and more economical to use than the alternative. In this respect, they are differentiated from hard tech companies who oftentimes are competing with incumbent solutions that are well understood and may create a fairly wide and deep chasm to widespread adoption.
For instance, I don’t think anyone would argue that the first company to make fusion power work and commercialize it will effectively have a monopoly and become richer than Croesus overnight. That’s why (as I outlined in a previous article) there are 72 companies funded right now to try and make fusion happen. As such, it clearly fits in the deep tech category.
The same could not be said for a company that is doing additive manufacturing (3D printing), or designing new types of windmills for renewable energy. There are clear competitors that they will need to differentiate themselves from and considerable market risk from adjacent technologies. It is clear that there is not the same monopoly potential there is vs someone with the secret to unlocking the next order of magnitude improvement in energy density afforded by fusion power. Which brings me to my next point
Hard Tech companies often come with market risk the founders don’t appreciate at launch
Hard tech companies, unlike purely deep tech companies, are up against a clock to get in the market before competitors come in with alternatives. These are made even harder by the fact that often times the sales cycles for hard tech products can be quite long - measured in quarters or years rather than weeks or months. Evaluation periods where customers actually use samples need to run long enough to discover any infant mortality effects and iterating takes longer due to the need for redesign and fabrication. Measure twice, cut once is the rule.
The reward for this however, is that hard tech is certainly stickier than pure software: its harder to fire a lidar vendor once they are built into your car model for the year than it is to churn a SaaS license. On the automotive front, vendor agreements commonly run in the decades, with long tails for guaranteed spare parts that need to be supplied and stocked often by federal statute. Military contracts can also run 25 years+, so if you’re luck enough to get into a Program of Record, you may be able to expect to put your kids through college on the proceeds. So while it may not be a pure monopoly like in the sense of whoever invents fusion power or like what Genentech had when they synthesized synthetic insulin in the early 80s, you certainly will enjoy a high degree of product stickiness with the ability to focus sooner on growing margins through economies of scale and optimization rather than constantly fighting off competing products.
Tying this all together
Hard tech is a unique category differentiated from Deep tech and Software. It focuses on hardware and hardware focused software companies which support them with relatively manageable technical risk (ie its an engineering problem, not a technology problem). The lower technical risk means faster time to market, but also carriers the risk of more competition able to enter the space to compete. Hard tech companies also tend to operate on longer go to market cycles then pure software plays, but that bring with it stickier product market fit that may even go into decades if you are lucky.
Several subcategories of software exist that support hard tech and deserve to be included. These products are integral to manufacturing 4.0 and are helping unlock new capabilities and performance in newly designed systems. Many of them are inspired by innovations coming out of major players like SpaceX, that literally broke the mold on modern manufacturing and aerospace design processes.
It’s important for hard tech founders to realize that there is a business side to what they are doing and its not just an engineering problem. Go to market is usually much harder than you think here with prototyping and evaluation periods that always take longer and are much more delayed than you initially thought. Even experienced hands in the hard tech space may run into the classic credibility problem of hesitancy to work with you because you are an upstart with an uncertain future that could put a larger company at risk if you go out of business - as the old adage goes “no one ever got fired for buying IBM.” These are all real things that founders need to consider as part of how they enter the market and cross the chasm.
As a category, I think Hard tech brings with it a sizable set of differentiators versus other categories and has a place in any portfolio with a thesis supporting it (personally, this is why I’ve made a career out of it). Deep tech often times comes on too long of a timeframe to be palatable for LPs lacking 10+ year time horizons. Hard tech offers some category exposure to deep tech without the high likelihood that the large amount of technical research investment upfront on low TRL products won’t pan out.
Really cool read, thanks!