AI has become the year’s venture capital buzzword.
Dan Primack, Axios
Earlier this week I joined Dr. Juanita Gonzalez-Uribe at the London School of Economics & Political Science (LSE) to give a talk to her Entrepreneurial Finance class. I covered my personal journey to venture capital, what makes our partnership work at Remagine Ventures, and gave an intro to Generative AI, a current theme that connects to our investment thesis (we made 3 investments in the space).
Models like GPT have been around for years now, but it’s the accessibility of the tools, and the rise in awareness that are making generative AI go mainstream. Notably, ChatGPT has become the fastest growing consumer product, reaching 100 million unique monthly users in January just two months after launch. It took TikTok 9 months in comparison. ChatGPT has also become the #44 most popular site on the web (with 342 million visits in January) according to SimilarWeb. Incredibly, the traffic is purely organic with zero paid in advertising.
The excitement by both investors and entrepreneurs to embrace these technologies, has brought generative AI further into the mainstream. New tools are unveiled on a daily basis; at the time of writing this post I’m signed up to a number of waitlists for products that enable summarisation, video creation, 3D asset creation, presentation creation, etc. Speaking to an audience of university students, there was no doubt in my mind that this is a topic where the students have a keen interest 🙂
Infrastructure, cloud, models and application layers
ChatGPT is a very powerful tool, but is only one of 600+ generative AI tools out there. In Israel alone I mapped some 61 generative AI startups and that probably fails to capture companies in which generative AI isn’t their main product. Wix, for example, announced earlier this week that it started offering AI to write all the text for your website.
When we talk about the tech stack for generative AI, we need to consider the various components. Some startups take an API or multiple APIs (from OpenAI, Stability, etc) and others are developing their own models. Some use general data sets, and others train their models on proprietary data. NFX did a good job visualising the various layers and the level of competition for each in the illustration below.
A large majority of startups use APIs rather than develop their own AI technology, which can create a strong dependency and can make it difficult for startups to differentiate on their core capabilities. Take copywriting for example. Jasper, Copy.ai and a number of other startups offer similar services to ChatGPT, albeit with various layers of integration with advertising and CMS tools.
Several VCs are warning of hype cycle investments in AI and compare it to the crypto bubble.
Below are some recent examples:
But on the other hand, here’s an opposite take from Paul English, the co-founder of Kayak:
Eyal Gil referred to this lack of defensibility of early stage startups (especially in the AI ‘wrapper’ space) in his post “Defensibility and competition“. He argues that most early stage startups are not very defensible and that they have several ways to build defensibility over time, but ultimately being an early mover on a need that is fairly obvious to the incumbents, may only produce a short lived moat.
I decided to focus my LSE talk on the generative AI space from an investor perspective, including some of the challenges that I mentioned in my post ‘Investing in Generative AI tools‘, so I won’t repeat them here. Overall, I’m excited to continue exploring the compelling use cases for generative AI with Remagine Ventures especially as they relate to gaming, education, simulation and enhancing creativity.
You can browse through my LSE lecture slides below:
And check out this short video by HourOne:
And a short clip, courtesy of Munch: