How do AI companies earn money?

Taivo Pungas -

Which business models benefit from AI?

This is an audience question I recently got at a webinar aimed at early-stage startup founders. The answer is almost like enumerating all business models because AI is like Javascript. Which businesses could benefit from Javascript? It can create value at almost every company.

I’ll make the question more specific: what are typical ways companies earn money with AI today?

You can sell single decisions. AWS Textract, for example: you can send them an image and they send you back all the text detected in the image. So they have an OCR API, and they sell you each request for something like 0.1 cents. It’s a decisionmaking service, which is not a new model but has become common for AI offerings because it’s the simplest way of selling an algorithm.

You could do a subscription model where the client pays a fixed fee every month for using the software and perhaps has usage-capped packages, or the fee depends on the number of users. It could be as simple as selling packages of single decisions, e.g. 1,000 decisions per month for $10. Or, each decision might not be explicitly counted, like in the case of x.ai which sells a scheduling AI assistant as a service, where the price does not depend on the number of meetings scheduled.

You can sell a software license. There is a Silicon Valley/Ukraine based startup that sells a SLAM SDK. SLAM means simultaneous localisation and mapping, which is a key component in building autonomous robots, virtual reality rigs, etc. So their client purchases a license for using that software and integrates it into their robots, mobile phones, headsets, etc. Then that device would have relatively good capabilities for solving that problem. The AI provider could charge a few dollars per device sold or simply a lump sum of ten million dollars for the license.

Other models like revenue-sharing are also possible but less common.

The model that makes the most sense is mainly a consequence of other things like

  1. what model your clients are used to;
  2. how the software will be deployed (e.g. onto an edge device vs in the cloud vs on-premise) — which itself is affected by data sensitivity, intellectual property, and other concerns;
  3. the ratio of fixed vs variable costs of offering the service to a new client.

OpenAI’s capped-profit model

If you missed it in 2019, OpenAI pioneered a capped-profit model which is an interesting combination of non-profit and for-profit models. Under that model, any investor or employee’s potential returns are capped, above which all profits go to the OpenAI Nonprofit. The cap “is negotiated in advance on a per-limited partner basis” and for the first round of investors, the cap is 100x their investment.

The capped-profit model is really an ownership model, not a description of how value is delivered and captured. You might think it’s weird for OpenAI to be innovating something so fundamental when they have no idea what value their work towards general AI will provide. It’s not: if they capture a significant chunk of the world economy, humanity would rather not give it to a couple of VC funds.