Not All Deep Tech is a Startup
Venture scale returns are the vast minority of viable businesses, and deep-tech is especially hard.
The quick summary:
Software has outlier qualities for driving extremely large returns: zero-cost replication, instant scaling, perfectly variable opex, reliable infrastructure, open-source libraries.
Despite this, building software unicorns is exceedingly rare and difficult. Software founder expectation value is highest for a non-venture growth model.
Deep tech usually has none of those qualities and is even harder to build, making viable venture scaled businesses even more difficult
High capital requirements to revenue must be met by similar massive revenue generating leverage, either through software-derived value, compounding economies of scale, high margin-consumables, or nearly infinite capital goods durability.
Technology is a ladder to value creation. You have to reach for a rung that’s actually there, and climb your way to the orbital-shipyards of tomorrow.
Software Is Amazing
For decades software drove outsized investment returns with extremely fast growth and high margins, features of a product environment with mature infrastructure and zero-cost replication. The fundamental cost-basis of providing software services has also continuously declined through steady improvements in compute scale, storage, and interconnect, while the underlying market of users has grown to be billions.
Despite all this, numerically the odds as a founder are exceedingly small to be considered a ‘win’ from a venture fund’s perspective. It’s incredibly rare to reach $100m in revenue and hit $1 billion in valuation, and a great deal of startups driven to ruin in search of tail-end outcomes could’ve lived a healthy life as a cashflow-generating independently owned businesses. Venture capital comes with strings attached, like pushing for growth, loss of board control, ongoing dilution, exit pressure, and potentially misalignment of goals.
The vast majority of businesses in the United States are privately owned, but startups capture disproportionate interest with dramatic stories of world-changing ideas, massive investment rounds, media frenzies, storied falls and titanic victories.
Recently the expectation-value for a founder pursuing an independently owned, boot-strapped or single angel-round funded company are even higher - “LLM-powered everything” drastically supercharges the small team versus the large incumbent with speed in iteration, and cheap scalability in traditional cost-driving areas like sales, marketing, customer support and qa testing.
It’s cheaper and easier than ever to start a new business, the market is massive, and for every unicorn opportunity, LLM’s have enabled 10,000x more independent business models that can generate real cash flow without raising money from professional investors. The oceans may run red with fleets of privateer founders taking a chunk out of software whales, turning incumbents margins into their profits.
These features - reliable infrastructure, perfectly scaling opex, zero-cost replication, large markets - make software an entrepreneurs dream. And now we can create it with natural language.
We simply speak and technology is created. It’s magic.
Deep Tech is Insane
The frontiers of engineering research have none of those qualities - niche fields with small pools of talent, poor standardization of tools, sparse and limited supply chains, massive technical risks in multi-year timelines where last-minute changes drive cost and delay overruns. These risks come on top of any baked in scientific risk or limitations of fundamental engineering capabilities - take a look at the state of error correction in quantum computing for a painful tutorial.
Theses issues of talent, tooling, logistics and timelines manifest in practice as a Great Filter known as the “Valley of Death” on the survival of new business ventures seeking to translate the frontiers of applied sciences into product lines through at-scale manufacturing. The long payback period for financing manufacturing projects means even well-established industry players have a difficult time raising private debt to fund expansion and scale, motivating the development of IFCUS, a government-backed loans program to co-underwrite manufacturing and industry specific debt for growth projects.
It’s risky, the payoffs are uncertain, and the commanding heights of the industrial economy are often surrounded with barbed wires of regulatory capture and piled corpses of past startup attempts. It’s grim, it’s the world of glass chewing Balrog Elon’s from the deepest mines of tech.
Why do we bother trying?
Creating New Worlds
The upside to successful deep tech ventures is you build a science-fiction future worth looking forward to. The more we master the natural world the greater our leverage on providing lives of prosperity and abundance to every person on the planet, and the trajectory of industrial capitalism with all its faults is clear - improving the material quality of life, curing diseases, and lifting people from repetitive physical toil into comfortable jobs of more agentic creativity.
This is the fundamental promise of basic science research and development, and not something unique to “Deep Tech Venture Companies” - so where do we draw the line? Both have extremely long capital payback periods, immense amounts of technical risk, and potentially massive impact.
To understand what makes a good deep-tech company as opposed to just engineering research and development, it helps to understand what specific leverage on value generation the completed technology will provide.
Here are a few patterns, by no means exclusive:
Software-Derived Value
It’s the iPod + iTunes model: once built or sold, the hardware platform can churn out software-enabled zero-cost of replication services. Hardware is the wedge, but it’s a software play. The first true example of this was in radio communications, where the marginal cost of a broadcast is quite small, but also applies to satellite constellations, new computing hardware, and even the App Store’s 30% revenue share.
Compounding Economies of Scale
This is the von Neumann Constructor pattern - where a manufactured item is an input to its own production process. For example, consider a company that sells robotics for machining industrial goods, and builds its robotic machines using its own robotic machinery. Getting to production volume means economies of scale, which lets you develop and build the factory cheaper and cheaper. Gigafactory is just the beginning of this trend.
High Margin-Consumables
This is the Juice-as-a-Service model where a capital-intensive platform enables the recurring sale of high-margin consumables to reach massive value leverage. Fundamentally this dynamic drives the entire biotech industry, for both platforms like NGS, single-cell sequencing, transcriptomics, as well as therapeutics. In 2023 Illumina derived 68% of its revenue from consumables sales while only 15% directly from equipment. In therapeutics, the high-capital cost platform is the entire pre-clinical, clinical, and manufacturing apparatus, which enables the deployment of therapeutics whose marginal costs is a tiny fraction of their ticket price.
Extreme Capital Goods Durability
The Too-Cheap-to-Meter pattern applies for anything proposing to provide a commodity good but in a revolutionary new way, such that the long-run capital costs of deployment average out over such a volume of production the per-unit depreciation is essentially negligible. Nuclear power is the prime example - the power generated from a nuclear plant is just as good in 30 years, unlike the performance of a computer chip which rapidly becomes obsolete.
The Monopsony Exception
The exception to all of these is to sell to the government, specifically the Defense Department, which is much less price sensitive and a lot more function-oriented - if you can provide a better fighter jet, missile, tank, or bomb, that delivers a decisive military advantage, then they are perfectly happy to pay a premium for it. In time, that high-cost defense application hardware can reach better economies of sale to become price-competitive in less performance-demanding civilian markets. The DoD is all about establishing beachheads.
Normal Business are Good, Too
Not all companies in the space of atoms have to be deep-tech, or venture-backed - a quality 5-axis CNC machine and experienced operator can pay back capital costs in 5 months of producing motorcycle parts, after all.
That’s still building in the world of atoms, manufacturing a better physical environment. And its long-term profitable, unlike the vast majority of ‘deep tech’ ideas - in the end, don’t assume long engineering development timelines and large capital expenditures necessarily guarantee a venture-scale outcome if successful.
At the end of the day, it has to not just be valuable, but have outsized leverage in value production.
Great post Andrew, I will be in SF in April and would love to meet if your around. Cheers!