Generative AI — Startup Landscape

A brief taxonomy

(This article is based on my notes of Generative AI startups list | Dealroom.co)

The Three High Level Groups

Generative AI (GAI) startups broadly fall into one of three groups.

Group 1: General Generative AI

The first, and perhaps most important and foundational group, involves Startups that build General Generative AI models (e.g. Large Language Models), which can be used for many things, and which other Startups usually build on top of.

Group 2: Tools

These are Startups that specialize in building products that help other products and users optimize and fine tune how they use Generative AI.

These include prompt management,

Embeddings management,

Model Tuning,

Production Engineering,

And programming frameworks for developing GAIs.

Group 3: Generative AI Applications

AIs fundamentally do one thing. They answer questions. Generative AIs are a subset of AIs that generate something in response to a question. So, unsurprisingly, the most common type of GAI startup involves products or services that generate something, usually a particular type of data, and usually for some specific application.

Sub-Groups of Generative AI Applications (Group 3)

Since this is the most common group, we will further sub-divide.

Group 3.1: Text

Startups in this group specialize in using Generative AI to produce written content such as articles, reports, or even entire books. These AIs can be trained to write in a particular style or tone, or to generate content on a specific topic.

Text generating GAI typically generate copy for various media,

Research (effectively a type of search),

Or customer services.

There are also a few niche text GAIs, like those that generate text for legal documents.

Group 3.2: Code

(Code is technically 3.1 Text, but, I thought, deserved its own category)

This group includes startups that use Generative AI to generate computer code automatically. These AIs can be trained to produce code in a specific programming language, to solve particular programming problems, or to optimize existing code. This technology has the potential to revolutionize software development by significantly reducing the time and effort required to write and debug code.

Code GAIs generally fall into two groups.

Those that support professional programmers,

And those that generate code for users who cannot afford professional programmers.

Group 3.3: Audio

Startups in this group specialize in using Generative AI to produce music or other audio content like speech. These AIs can be trained to create new compositions based on a particular style or genre, or to remix existing sounds in new and interesting ways. Audio-generating AIs can be used in a variety of industries, including music production, advertising, and video game development.

Audio GAIs also divide into two types: Music,

And speech.

Group 3.4: Video

This group includes startups that use Generative AI to create video content automatically. These AIs can be trained to generate video in a particular style or format, or to combine existing video and audio content in new and interesting ways. Video-generating AIs can be used in a variety of industries, from film and television to advertising and social media.

Group 3.5: Game Assets

Startups in this group specialize in using Generative AI to create assets for video games. These AIs can be trained to generate 3D models, textures, animations, and other assets that game developers can use in their projects. Game asset-generating AIs have the potential to significantly reduce the time and cost required to develop high-quality video games.

A special subset of Game Asserts include GAIs that develop game characters.

Group 3.6: Test Data

(Technically, this group overlaps with all the others, but I kept it separate as it is its own niche)

Startups in this group specialize in using Generative AI to produce test dataset for use in other applications. These AIs can be trained to generate data that is similar to existing data, to produce data that meets certain specifications, or to create entirely new data sets. Data-generating AIs can be used in a variety of industries, including AI itself.

Group 3.7: Images

Finally, startups in this subgroup build generative AI models that generate images in response to a question or prompt. This can be useful for various applications, such as graphic design, image editing, and even generating realistic-looking 3D models. Image-based generative AI can also be used for image recognition, object detection, and other visual tasks.

GAIs generating images can be split into large images,

And smaller, more niche images like Logos, Icons, Fonts and ClipArt.

Concluding Caveats

Hope this is useful. I’ve tried to be exhaustive, but there maybe some categories I’ve missed.

‘Also, note that a perfect taxonomy is impossible because many startups overlap between categories. Finally, some of these “startups” are no longer small companies, and many which are, soon, will no longer be. Testament to how fast the field is developing…