Understanding business models for AI companies
We are in the middle of the AI revolution. Apart from AI turbocharging advertisements and content understanding, numerous startups are building models, AI platforms, tools & frameworks, and products.
In this note, we examine the two main business models for AI companies.
1. AI-infra as a service (open source business model):
A business model based on open source is similar to a freemium model. The users get the software free but are willing to pay for premium features for additional value. Here are some common ways of providing additional value :
a. Cloud services:
There are excellent ML frameworks available that are free to use. Since ML/AI is a very compute and data-intensive field, users would need software services for data management, data lake, compute services — GPU/TPU, etc.
There’s a flywheel effect here. The more users level up with their model lifecycle, such as larger models, automated retraining, model monitoring, etc., the more they will need data and compute services ($$$).
Examples of frameworks backed by cloud services are TFX & VertexAI by Google cloud, Sagemaker by AWS, Azure AI by Azure, MLFlow by Databricks.
b. Specialized accelerator frameworks:
Specialized frameworks are tied to specific hardware platforms for accelerating your workloads (training & inference). For example, Nvidia triton inference server can accelerate model inference on GPU significantly. It is available for free but you are locked in to GPU hardware platform ($$$).
3. Pay-per-use inference (Model hosting):
There have been several breakthroughs in language and vision understanding (& generation) in the last decade. Along with the breakthroughs, the usage of these foundational models has increased manifold.
Foundational models are touching 100B parameters and beyond. This requires sophisticated machinery to host the model and use it for inference (both batch and real-time). This is a business model being used by companies such as Hugging-face (NLP, multimodal) and Deep-gram (ASR).
2. AI-powered products :
With spectacular AI-powered advances in NLP, CV, ASR, translation, and other domains, there is a thriving marketplace for AI-powered products.
The suppliers in this marketplace are domain solution providers: chatbots, robotics, enterprise support, fintech (risk prediction, fraud detection), content search, etc.
The consumers are companies using AI-powered products in their B2B/B2C businesses for competitive advantage. The usage is billed accordingly.
A few examples of AI-powered products are -
a. Digital Assistants: Software that helps you to do task X better. Examples are Grammarly for writing, Calendar apps for scheduling meetings, etc.
b. Chatbots: Chatbots can improve customer support through automated support.
c. AI-powered surveillance devices: From home security (Ring) to enterprise-grade security, AI supercharges security.
d. Standalone Recommendation services: Content recommendation services such as movies, and tv-shows have become quite popular.
Here, we went through two distinct ways of monetizing AI software :
- AI infra as a service.
- AI-powered products.
Let us know if you have come across other ways of monetizing AI software.
