November 2023
Generative AI (GenAI) is not new, and many investors, corporations, and media outlets are surfing the hype. The technology and computational advancements that enable GenAI were developed years ago, but only recently with the explosive growth of ChatGPT in November 2022 have mainstream investors, corporations, and consumers begun to pay attention. GenAI is, however, an absolutely groundbreaking technology that has the opportunity to revolutionize large swaths of the economy ranging from education, business, law, and more.
GenAI can be thought of as machine learning (ML) 2.0. The transformer architecture founded in the self-attention mechanism is the next step in what many consider “traditional” ML. The use cases, however, are quite distinct. GenAI is inherently not suited for predictive/discriminative tasks that machine learning 1.0 models are intended for. Generative AI, aptly named, is designed for generation: the creation of data that closely resembles data the model has been trained on, be it text, image, video, or code. GenAI enables greater quantity and velocity of content creation, which can rapidly re-shape several disciplines both creative and otherwise.
Less savvy investors are setting money on fire trying to ride the GenAI wave. More discerning investors understand where in the investable universe pockets of value still exist to unlock with capital. Below I will describe the key areas investors should explore for investment in GenAI.
There is notably strong demand for generative AI solutions from a variety of industries, but the most investable areas are in the “messy middle” of the GenAI ecosystem. A sandwich metaphor is apt here, where the good stuff is in between the top and bottom layers of the stack, as illustrated simply below:
Less interesting: the application layer and infrastructure layer. Big Tech and Nvidia already dominate the infrastructure layer through ownership of the cloud computing and GPU industries. Upstarts will have a tough time displacing chip designers or cloud hyperscalers on their own turf. Additionally, the companies with the best closed-source foundation models (of which there can only be a handful) have functionally already been “acquired” (OpenAI by Microsoft, Anthropic by Google). The application layer is slightly more investable, where select companies that truly differentiate themselves with value-added fine-tuning and services can thrive (e.g., Writer). Hundreds of startups have formed in the past year that are simply skins on top of GPT-4, and provide marginal value above using the products on offer from the model providers themselves. We will soon see a graveyard of these startups form as the market comes to terms with the fact that these are features, not companies poised to capitalize on large TAMs and significant growth.
More interesting: “the messy middle.” Everything happening in between the app and infra layers is much more investable, as this is where the magic happens that enables the input of data and the output of generated content. There are several areas to explore, including but not limited to: vector search and curation, data labeling, ML orchestration, model optimization/training/observation, etc. Finding the next Snorkel.ai, Zilliz, or Pinecone should be the priority for GenAI investors right now, as tailwinds in data creation and curation across enterprises seeking to leverage AI presents a large market opportunity. Most Fortune 500 corporations are still in the early innings of their data/AI maturity journey; enterprise leaders do not want to be caught flat-footed, however, and are seeking out ways to leverage GenAI in their organizations in a safe and cost-efficient way. The startups that can capitalize on this AI-enablement opportunity are primed for growth and successful exits via IPO or acquisition.
We are still in the early innings of the GenAI revolution, but the pace of development, investment, and adoption is greater than previous landmark developments like cloud and mobile. Identifying innovators working on cracking the problems in the “messy middle” will be a winning investment strategy for early-stage and growth investors. These founders are not the ones tackling advancements in the computer science and mathematics of AI, but rather are the ones who can see the delta between the groundbreaking technology and its everyday use in businesses and for consumers. There is still quite a lot of uncertainty with the legal/regulatory environment surrounding AI very much in flux, but identifying the companies with resilience to those headwinds and clear differentiation from the infra and app layers will be worthwhile to investigate.