Investing in Technological Supercycles: AI, Web3 and Non-consensus Founders
What do AI and Decentralized Protocols Have in Common?
The piece was originally published here.
At Leonis Capital, we are particularly interested in "supercycle investments" – these are companies powered by technologies that could have a lasting, positive impact on society, often for 50 to 100 years. We have a strong bias for early-stage investments in these companies, favoring non-consensus founders who think long-term, stay ahead of the curve, and strive to create real value with supercycle technologies.
You might wonder, aren't “supercycles” and “being nonconsensus” contradictory? After all, supercycles are defined by new trends and perhaps some market frenzy.
To be sure, in every supercycle, there will be many consensus founders who are keen to “ride the wave”. They might build great companies short term but investors and the overall society are paying a premium for the temporary momentum and not for the company’s true value. Non-consensus founders, on the other hand, pay little attention to “the new hype” but focus on building ground-breaking technology slowly and steadily, without much fanfare. They have unique insights about their industries and often toil away for years perfecting their products and services before seeking outside validation. They are hidden gems because they capture the value created by technological supercycles without demanding the consensus of the hype.
We believe the next tech evolution is leading us to the next exciting supercycles that are driven by AI/ ML and decentralized protocols. So we started a venture firm to back non-consensus entrepreneurs who have the technical insights and unique POVs in these two supercycle tech catalysts.
What are Supercycles?
The global economy is prone to cycles of expansion, peak, recession, and recovery. A supercycle is a sustained period of expansion often driven by technological revolutions.
Source: Adapted from Allianz Global Investors “The Sixth Kondratieff – Long Waves of Prosperity” (January 2010)
In economics, supercycles are often referred to as "Kondratiev waves" or "long waves." The most well-known supercycle is the Industrial Revolution, which began in the late 1700s and triggered a cycle of industrialization, economic growth, and global commerce.
Steam engines and electricity were examples of supercycle technologies that transformed our human society. More recently, the internet, invented in the 1980s, has become such a ubiquitous “invisible” tech fabric that we don’t even think too much about it. And yet 40 years later, we are enjoying the application layer of the internet - of Apple, Amazon, Alphabet, and Microsoft.
These are supercycle companies – most of them were born 20-30 years ago and they will be around 20-30 years later, if not longer: partially because of the visionary entrepreneurs and stellar management team, but mostly because they were founded and built on a paradigm-defining technology – the internet, or the information technology.
We believe that the next supercycle will be driven by AI and web3/blockchain technology. But before we dig into why we think these two technologies are supercycle technologies, we want to introduce a new concept – the "Supercycle-Consensus Matrix".
The Supercycle-Consensus Matrix
As we mentioned in a previous essay, economist Carlota Perez argues that supercycle technologies go through four phases: irruption, frenzy, synergy, and maturity. In the irruption stage, the technology receives relatively little investor attention. However, in the frenzy stage, the technology is hyped and the market's enthusiasm causes a "consensus-supercycle" effect - it becomes common knowledge to buy into the trend, as everyone is doing it.
In the consensus-supercycle effect, people often overlook the fundamentals and instead focus on the "right resume," the "bragging rights," and “what's hot”. It's our biological instinct to follow the crowd. But data shows that in these categories, companies are often flooded with capital and over-inflated valuations that are likely to decline or crash. Too much capital and ego can make it difficult to focus on customers and create real value.
Non-consensus founders, however, might have entered the field prior to the frenzy or might choose a path that is slightly different. They possess unique insights about the technology or industry that are not yet widely accepted by the market. Apple's Steve Jobs and Steve Wozniak are prime examples of pre-hype entrants. As such, non-consensus founders still capture the upside generated by the supercycle technology but their companies are often undervalued by the market. At Leonis Capital, we proactively seek out these asymmetries.
The founders in the non-supercycle categories can also be successful. But ultimately, they will miss out on much of the new technological revolution. They do more to disrupt old industries than build new market paradigms.
AI as a Supercycle Technology
Back in 2020, we published an essay on our viewpoint of AI and why it matters. That was a time when AI / ML was not as “hot” as it is today. In the essay, we tried to articulate why AI-first companies might have a different business model than traditional SaaS companies and why ultimately it will have a long-lasting impact on our society.
Since then, more people are aware of GPT-3 and DALL-E; even more, people are playing with multiple AIGC applications, from copy.ai (a business-oriented one) to Replika and PhotoforTinder (more joke-like ones). Investors have "suddenly" decided to pour hundreds of millions of dollars into AI companies. In an otherwise lackluster VC environment, AI is bucking the trend, creating unicorns and generating buzz in Silicon Valley. Examples include:
Stability AI, the company behind the popular image-generation model Stable Diffusion, received 101 million in a seed round led by Coatue and Lightspeed Venture Partners. Stability AI’s kick-off event in San Francisco’s Exploratorium was described by The New York Times as “a coming-out party for generative AI” and the Coatue partners called 2022 the “AI moment.”
Jasper, an AI content platform powered by GPT-3, raised a 125 million series A in November 2022, putting the company at a 1.5 billion valuation.
Earlier this year, HuggingFace, a company striving to build the “GitHub of ML,” reached a 2 billion valuation with its 100 million series C.
At the same time, generative AI has become “the new thing” in Silicon Valley, with models like Stable Diffusion and ChatGPT fascinating the Twittersphere. In a previous article, we shared our views on the emerging generative AI landscape. One of our conclusions is that many aspects of generative AI are over-hyped – the over-crowded fields, the pseudo-new market categories, and the applications that over-promise. AI didn’t suddenly become a solution to all of our problems. Nor does it mean that AI I is only now starting to be a game changer.
AI has gone through its 2017 hype to become a more mature, application-ready technology today. The AI models in 2022 are much more capable than those in existence 5 years ago and have clearer application cases. Since OpenAI's release of GPT-3 in 2020, AI labs around the world have raced to develop bigger and better language-, image-, and video-generation models, such as DeepMind's Gopher (175B parameters), Microsoft and Nvidia's Megatron-Turing NLG (530B parameters), China’s Wudao 2.0 (1.75T parameters), and Meta’s Make-A-Video.
The rapid advancement of the underlying technology has enabled an AI renaissance that is touching ever more industries. For example, large language models (LLMs) are quickly getting bigger and better over the last 2-3 years. Current-generation LLMs like ChatGPT, or GPT-3.5, are significantly better than their predecessors released a few years ago, such as BERT (2018), T-5 (2020), and even GPT-3 (2020). Here is a visualization from our previous essay.
AI copy-editing tools such as Jasper and Copy.ai have seen widespread adoption over the last two years and are rapidly becoming standard tools for content generation and marketing. AI-generated images are becoming increasingly realistic and impressive. The Midjourney-created piece "Théâtre D’opéra Spatial" even won a digital art award at the 2022 Colorado State Fair. Generative AI has reached other areas like code generation, where Github's Copilot is used by more than 1.2 million developers and writes nearly 40% of their code automatically, and video generation, where RunwayML has been using AI to edit the Michelle Yeoh movie Everything Everywhere All at Once.
Jason Allen’s piece, “Théâtre D’opéra Spatial,” generated by Midjourney, took first place in the digital art category at the Colorado State Fair. Source: The New York Times
From niche to “hot”, has anything fundamentally changed? Probably not. It’s the natural progression of the tech evolution – from irruption, hype, and synergy to maturity. The technology is getting better but the potential of the technology has always been there.
Unlike many in the VC business who believe that 2022 is the "AI moment" or the "beginning of generative tech" or that the generative AI hype is “absolutely justified,” we believe that the creative explosion we are seeing in AI today is just a part of the AI supercycle.
Take the impressive AI chatbot ChatGPT as an example. The model gained an astonishing 1 million users in less than a week and started rumors that it could supplant Google’s search engine. But if one reads OpenAI’s documentation closely, ChatGPT is basically an improvement upon an existing model InstructGPT, released in January 2022, but “with slight differences in the data collection setup.” In other words, it was not a key technological breakthrough but just another step in the improvement of language models.
The hype of generative AI will probably inflate the prices of consensus-supercycle companies that build on eyeball-catching technology. However, real market value comes not from fancy technology but from the real-world problems that AI can now solve. In our previous essay, we cautioned against AI projects that overpromise and those that are essentially hammers looking for nails. Instead of being blown away by the tech hype, non-consensus "AI-first" founders will need to place customer needs first and foremost.
Web3 and its Supercycle Potential
Now, let’s look at the Web3/ blockchain space. We also put out an essay on why web3 is a big deal after the downturn of the general web3 market. Since then, the macro has looked even worse with the sudden fallout of FTX and its widespread contagions. However, has anything fundamentally changed about decentralized protocols? Definitely not.
In fact, one can argue that the crypto scandals grabbing headlines this year – the FTX debacle, the Terra Luna crash, and the collapse of Celsius and Three Arrows Capital (3AC) – represent a handful of pseudo-web3 companies that are centralized exchanges or highly speculative financial products. On the other hand, most of the decentralized/automated protocols have been much immune from the human errors/ greed that was presented in both the Terra Luna and FTX cases.
The key (and the original) idea of web3 is decentralization, yet the recent scandals all involve centralized financial exchanges – essentially what we view as “pseudo-web3” projects.
Fred Wilson from USV published a post last month titled “Taking A Long Term View Of Web3”. The main point in his piece is that short-term turbulences should not shake up our confidence in the fundamental value of decentralization.
Having gotten involved in the blockchain space in 2012, we have witnessed the Mt Gox saga which resulted in a huge loss of customer funds and therefore depressed Bitcoin’s price for years. However, what happened next is the quiet building period that lasted about two years, within which the engine of Ethereum was humming under the hood. In 2015, Ethereum was officially launched. It opened up the next bull and bear cycle driven by various ICOs that peaked in 2017.
After the “ICOs” cycle, the market went quiet for 2-3 years until the “DeFi summer” came along in 2020. This was a year when the DeFi total value locked (TVL) saw a 20-fold increase from $700 million to a whopping $15 billion. These were times when the market was excited about the “crypto revolution” and how DeFi will “reshape financial services” forever.
Source: defipulse.com, image from Yield App
And now here we are. 2022 has been through a lot, with Terra Luna, 3AC, and now FTX. The system is going through another system shock – how long will it last this time? Nobody knows.
At the same time, we are seeing the hope in decentralized projects, which are increasingly challenging and replacing their centralized, pseudo-web3 counterparts.
Over the last year or so, trading volumes on decentralized exchanges (DEXs) are quickly catching up with centralized exchanges (CEXs). In the aftermath of the FTX collapse, there has been a significant shift in crypto trading behavior, with trading volumes on DEXs skyrocketing to a whopping $27 billion in the week after FTX’s crash. Meanwhile, CEXs are seeing large outflows, as trust in CEXs tumbles.
If you pay attention, whether it’s Mt Gox, scammy ICOs, or even the FTX debacle, the center of the collapses are human errors and greed. The original blockchain mantra is “In math we trust.” In other words, web3 is not fully here yet. But it will be. Many of the centralized entities for web3 thrived by the crypto hype and fell of human greed and systematic over-leverage.
Is 2022 web3 similar to the dot com burst in the early 2000s? It rhymes.
Many people wrote off the internet after the dot com bubble burst two decades ago. However, the power of supercycle information technologies, such as cloud computing and mobile phones, ultimately transformed the internet into an economic engine that is the backbone of the global economy today.
Web3, with its fundamentally different logic of organizing human knowledge, has the potential to be an even larger supercycle technology than the web2 internet. But the technology remains in its very early stages. The boom and bust cycles that we see in crypto and DeFi today are characteristic of technological supercycles driven by overexcitement and disappointment.
Leonis Capital is not a token fund by design. Firstly, we are not convinced that tokens are the best way to align with the founding teams in the long term. Secondly, we believe using the filter of "token market cap" to evaluate a project could be misguided, as it allows short-term market manipulation, which could distract the teams from focusing on the fundamentals.
Instead, we believe that in order for web3 to take off, more effort needs to be put into building the infrastructure and usability – the decentralized protocols that power web3 applications and the user interfaces that make them accessible to the masses. The negative headlines that we hear about are almost exclusively the most speculative projects in web3, which, unfortunately, were the consensus supercycle companies over the last few years. We still hold our belief that the next wave of opportunities will be in protocol-driven smart software that takes more patience and a longer-term vision to build.
AI and Web3 – The Intertwined Dance
How might AI and web3 – two supercycle technologies – be intertwined?
AI and decentralized protocols are “supercycle technologies” that go through periods of boom and bust but ultimately set new paradigms and create new markets. At the same time, we’ve been paying attention to some of the initial interactions and overlaps between these two technologies.
In essence, AI is about how data is interpreted and generated, in an autonomous way; whereas Web3 is about how data is stored and verified, in a decentralized manner. We believe entrepreneurs can create iconic companies leveraging either of these powerful technologies or even both!
We are already seeing interesting applications of AI and web3, particularly in the field of financial services. AI can be used to analyze financial data and predict market trends, which can be beneficial for DeFi trading. For example, Semiotic Labs, a company that Leonis has invested in, uses AI to improve the Graph protocol. They have also developed the first reinforcement learning agent to optimize trading routes and engage in web3 market competition.
Generative AI can also be used to create and animate NFTs. The key idea is that AI-generated content can be the commodity, while web3 offers a solution for human identification, management, and ownership.
For example, Bored Apes have come to symbolize the NFT craze. But AI researcher Yannic Kilcher trained an AI model, a generative adversarial network (GAN) to be exact, to generate an infinite number of Bored Apes. His app even allows you to control the style of the ape and turn your own headshot into a Bored Ape NFT. Try it out yourself here!
While AI seems to undermine the uniqueness and “collectible” quality of NFTs, it also democratizes art creation and makes NFTs affordable and accessible to the masses. (Sorry, Jimmy Fallon and Paris Hilton, your Bored Apes probably aren’t worth six figures!)
On a more serious note, we can imagine AI and blockchain technology being combined to track supply chain movements, with AI optimizing the supply chain using shipping and transaction data and smart contracts self-executing the terms of agreement and storing data on-chain that’s publicly verifiable; In healthcare, AI can analyze medical data and suggest treatment options and patient records can be stored safely on the blockchain as private digital assets. AI can be used to conduct biometric recognition and the blockchain can securely and transparently store and share ID data.
The possibilities of AI x web3 are limitless. We are still in the early stage of the AI-Web3 interplay.
Parting Thoughts
Supercycle technologies like web3/blockchain and AI have gone through plenty of ups and downs. But as the underlying technologies continue to improve, new applications will continue to emerge, regardless of whether we are in an "AI hype" or "crypto winter" or the other way around. It is important to take a long-term view when evaluating the technologies and the companies building on top of them.
Consensus-supercycle companies are often overhyped and overpriced, and as a result, may underdeliver. On the other hand, non-consensus supercycles companies are the hidden gems in a technological revolution.
At Leonis, rather than chasing new hypes or ditching old trends, we choose to focus on the value that supercycle technologies can add to our society and their long-term implications for the future.
That’s why we are most interested in “non-consensus supercycle” companies that will still be around 20, 50, or 100 years later and will continue driving economic growth and shaping humanity’s future.