On Deep Tech and Why Now is the Best Time for Company-Building
Examining Technological Genotypes and Phenotypes
The recent collapse of Silicon Valley Bank, a $200+ billion dollar behemoth that had tirelessly served the startup ecosystem for four decades, combined with the ripple effects on the banking sector as a whole, has collectively infused a sense of pessimism in the financial markets. There’s no doubt that we’re in a bear market at the moment; layoffs in big tech and the cutting of non-critical spend across the board is proof in the pudding. But, as the classic adage goes, necessity is the mother of invention: this shift from an extreme overabundance of capital resulting from easy monetary policy to financial austerity is, fundamentally, the forcing function which startups need in order to embrace a new set of disciplined and healthy habits.
There’s a reason why some of the defining companies were started in downturns, from Cisco in the late 1980s to Google and PayPal in the early 2000s. When capital is expensive, only the very best companies have access to it and the bar for staying alive is higher; the companies who survive in a downturn thrive when the market returns to normal.
The same can be said about venture funds, which is why Lux Capital GP and founder Josh Wolfe believes that 2023 and 2024 vintages can deliver some of the best returns in venture capital history, saying:
"Future returns from funds deployed in 2023 and 2024 vintage years, with declining entry prices and more orthogonal investment strategies, may prove to be the very best in nearly a generation."
This begs the question: what is an orthogonal investment strategy? For Lux, it’s investing in deep science, taking technical risk over market risk. But while Lux was one of the first firms to pour money into deep tech, they’re not the only ones.
When I started working at Starburst Ventures, a deeptech firm investing in aerospace and defense startups, in January of 2022, I first had to ask the question: what is deep tech? The companies I was looking at had a potential consequentiality that wasn’t seen in B2B SaaS plays: rocket propulsion companies, edge computing startups, and cutting-edge robotics businesses. However, while I came into Starburst thinking that deeptech was a just fringe section of the venture ecosystem only funding moonshot ideas, I soon came to learn that every product that is used roots from a deep technology.
Fundamentally, deep tech is an innovation on a core technological layer, which consequently becomes the infrastructure and foundation for more companies to build on top of. Most people see the Internet as a wave which spawned a number of generational businesses, but what they forget is it was all enabled by the deep technology which was TCP-IP and fiber optic networks. Companies like Salesforce and Amazon built cloud computing platforms (Salesforce built sales cloud and service cloud, Amazon built AWS) which enabled a whole host of cloud SaaS businesses. The iPhone is another example of a deep technology that led to the revolution which was mobile and all the mobile applications built for the App Store. And even crypto and Web3, one of the latest hype cycles in tech, has been made possible by the deep tech which is blockchain technology.
One mental model which I use to think about deep tech and the subsequent products built on top of it is the genotype-phenotype concept from biology. In biology, a person’s genotype is their genetic sequence, the specific combination of alleles in the gene. Their phenotype is how that DNA is expressed; the physical manifestation of the allelic combination. For example, genotype “Bb” is physically expressed with the phenotype of brown eyes, while the genotype of “bb” is physically expressed with the phenotype of blue eyes. In the context of tech, the deep technology layer is the genotype and the application layer in which that tech is manifested and expressed to the public is the phenotype.
Another key feature of the genotype-phenotype idea is that while phenotype is largely a function of genotype, other external factors such as one’s environment and habits can strongly influence how the phenotype comes into being. For example, though height is largely a function of genetics, people who play a lot of sports and consume nutritional foods are more likely to be taller than those who don’t. Another example is intelligence: while DNA and genetics play a huge role in determining one’s intelligence, the environment in which one is brought up and their habits also impact how the genotype translates to the phenotype.
At a technological level, while each deep technology built has its intuitive, natural manifestations and applications, it’s largely the surrounding environment and the builders developing the products who determine how the genotypic deep technology is manifested at the phenotypic, application level. For example, a lot of the Web3 hype cycle was not caused by blockchain technology alone, but precisely by how the founders building in web3 manifested that technology in phenotypes such as cryptocurrency and NFTs. Similarly, the SaaS products which were overvalued were simply zero-interest rate-driven manifestations of the underlying deep technology (cloud computing). Change the environment and the surrounding factors (ha, get it), and you’ll change the way in which the genotypic deep technology is manifested at the phenotypic application layer.
From an investing standpoint, when looking at any product, it’s so important to ask the question of what’s the genotype which underlies this product? What’s the deep technology, the core IP, which is enabling this? Because, from a returns standpoint, the infrastructure touches every layer of the stack, and consequently profits from every new iteration of what’s built on top of it. There’s always uncertainty over the applications built on top— some are game-changers while others are simply derivative ideas; investing in infrastructure is always a great bet, as it’s almost like building an index for the overlaying products.
The AI gold rush is well underway, and it’s exciting to imagine the possibilities of what can be built at the application layer— customer service chatbots, coding co-pilots, and video generators are all members of the set of things that could go mainstream as a result of this new generative AI wave. And it’s certainly possible that a good number of these products can really make their way into the echelon of things that change the way we work, live, and play.
But, for every instance of an AI product that’s developed, more money flows to the underlying infrastructure: the foundational model layer, cloud computing services, and chips. That’s why it’s important to keep an eye on these companies:
Semiconductor companies include emergent players like Cerebras, SambaNova Systems, and Graphcore, coupled with incumbent players such as Nvidia (by far the biggest winner so far, with $3.2B of data center GPU revenue in Q3 F2023, a lion’s share coming from AI), Intel, AMD (Instinct GPUs), AWS (Inferentia and Trainium chips), and Google (TPUs).
Cloud computing providers such as Amazon’s AWS, Google’s GCP, Microsoft’s Azure, and GPU cloud player CoreWeave.
Foundational model providers such as OpenAI, Character.ai, Anthropic, Stability.ai, and Adept
These companies are more capital-intensive and may have slower iteration cycles compared to a traditional software product, but the companies that overcome the massive technical risk will have few issues selling their product in the market. Now, more than ever, it is important to remember Perkin’s Law, popularized by Kleiner-Perkins founder Tom Perkins: technical risk is inversely proportional to market risk.
I’m excited for the companies that will be built in the next 5-10 years. As I wrote back in September, market corrections are simply a filtration mechanism embedded within the complex adaptive system that is the economy. Crisis can be an opportunity for the bold; the companies who realize this and either build deep technology (genotype) or consequential manifestations of that technology (phenotype) will be the ones who stand the test of time.