This post is the first in a series that will explore March Capital’s approach to Generative AI in both our portfolio and investment strategy.
“It was the best of times, it was the worst of times…it was the spring of hope, it was the winter of despair.” Charles Dickens’ famous opening line of A Tale of Two Cities perfectly describes the duality of the state of the venture capital industry today.
The Current Venture Landscape & Economic Reset
10-Year US Treasury yields soared to over 4% in 2022 after a prolonged period under 2%. This precipitous rise in interest rates and, therefore, cost of capital, brought record-high public market valuations back to earth over the past twelve months. Snowflake (SNOW), the high-flying tech darling, is a perfect example. After reaching north of a 150x EV / NTM revenue multiple in September 2020, SNOW is now trading at 17x.
Private market valuations have followed suit with pre-money valuations for Series A-D funding rounds falling at an unprecedented pace (50%+ from Q1 2022 to Q4 2022), while venture dollars deployed declined to 2019 levels.
Fig. 1, VC Valuation Declines During Economic Resets; Pitchbook (April 23); Note: Valuation decline shows average median pre-money valuations for Series A – D funding rounds.
These indicators suggest the industry is suffering through the “worst of times,” or at the very least, a meaningful reset.
The Rise of Generative AI
For Generative AI, however, the last twelve months have represented the “best of times.”
Nearly $14B has been invested into Generative AI through the first half of 2023. This represents nearly 40% of total VC investment this year to date and an astounding 3x increase on the total capital invested into the sector from 2022. Additionally, Generative AI companies are receiving valuation premiums between ~50-130% on median Series A-D investments.
Fig. 2, Premium on Generative AI Valuations by Stage in 2022; Pitchbook (5/23); Note: Data compares Generative AI median pre-money valuations to median pre-money valuations across all deals.
All this excitement leads us to ask, are we experiencing a watershed moment in technology akin to the birth of software in the 1990’s, the rise of mobile in the 2000’s, or modern SaaS in the 2010’s? Or, is Generative AI more hype than substance?
To answer this question, we’ll first take a step back and examine Generative AI in more detail.
What is Generative AI?
Generative AI is a category of artificial intelligence (AI) models leveraged to produce net-new assets such as text, images, audio, video, code, and synthetic data. These models have served as the next evolution of machine learning (ML). Before Generative AI, ML models were largely predictive in nature, classifying and identifying patterns in existing data.
ChatGPT (Generative Pre-trained Transformer) is the most recognizable application of Generative AI to date – powered by GPT-3.5, a large language model (LLM) developed and popularized by OpenAI, which leverages a transformer architecture (developed by Google in 2017) designed to process sequential data (i.e., language data) in parallel. By abstracting away underlying architectural complexities and engaging users in an intuitive chat-based web interface, ChatGPT saw the fastest adoption of any modern internet platform and catalyzed an explosion of startups seeking to push the envelope with this new technology.
A new startup ecosystem, with some overlap to the broader AI ecosystem, has emerged around Generative AI. Large public companies such as NVIDIA (NVDA) and Advanced Micro Devices, Inc. (AMD) dominate the market share of the infrastructure behind Generative AI. These companies provide the Graphics Processing Units (GPUs) on which various types of Generative AI models are built and run. Leading companies building large multi-purpose models with this infrastructure include: OpenAI, Google, Facebook, Microsoft, IBM, Cohere, Anthropic, and more. Various applications focused on both horizontal and vertical-specific use cases sit on top of such models. These applications vary from a light user interface layer built atop open-source models to full-stack applications utilizing proprietary data and / or models. Companies building in the application layer include Hippocratic AI, Enterpret, Runway, and Typeface. Finally, alongside the infrastructure, model, and application stack sits the enablement layer. AnyScale, OctoML, Snorkel, and Modular are examples of companies building within the enablement layer to help various elements of the Generative AI stack operate more effectively and efficiently.
Fig. 3, The Simplified Generative AI Tech Stack.
Generative AI: Hype or Substance?
At March Capital, we believe that Generative AI has tremendous potential as a transformative technology. In fact, it is projected to generate $100B+ in revenue by 2030. With that said, VC attitudes toward Generative AI are currently at or near the peak of inflated expectations and will likely traverse the “trough of disillusionment” before finding the “plateau of productivity”.
For this reason, we are keeping a few first principles in mind as we consider our investment approach to the sector:
- A sustainable competitive advantage is still required. A “Generative AI” moniker is not a sufficient long-term competitive advantage. Investable companies must still possess some proprietary offering – technology, team, network effects, etc. – to effectively attract and retain customers.
- Unique data provides a right to win. Expanding on the point above, Generative AI models are only as good and as unique as the underlying data on which they are trained. Companies with proprietary, unique, or difficult to access/compile datasets have a competitive advantage.
- Distribution is crucial for scale. No matter how advanced and unique a company’s technology, they must still go-to-market. Companies with strong existing distribution or other GTM advantages have a head start in the Generative AI race. Upstarts and incumbents alike are taking advantage of this new technology. Startups that rise to the top will either have a unique advantage over incumbents or they’ll be able to define, dominate, and capture value in an entirely new category.
These are just a few of our considerations as we evaluate companies within the Generative AI landscape.
Generative AI within the March Capital Portfolio
While Generative AI is the latest craze, March Capital has a long track record of AI investing. We believe existing companies can combine data, technology, or distribution advantages with Generative AI techniques to achieve or cement category leadership. Several companies within our portfolio are doing just that. These include companies utilizing Generative AI within analytics (i.e., ThoughtSpot), customer experience (i.e., Uniphore and ASAPP), computer vision (i.e., Parallel Domain and Spark Cognition), and Healthcare (i.e., Generate, Tessera, and Suki).
Generative AI has undeniably launched a spring of hope in an otherwise wintry VC climate. Over the course of this blog series, we’ll explore some of the exciting things that our portfolio companies are working on before sharing our Generative AI thesis in more detail. To begin, in our next installment, we’ll dive into how ThoughtSpot is utilizing Generative AI to transform analytics.