Generative AI services like GitHub Copilot do not appear to be profitable for Microsoft yet. The cost of maintaining AI models far exceeds the revenue.

Microsoft wants to distinguish itself as a leading provider of generative AI and the software giant is willing to do anything to achieve this. Billions of dollars have already been invested, but the ROI is currently very negative. Training and maintaining models costs much more than the direct revenue that generative AI already brings in.

The Wall Street Journal cites GitHub Copilot as an example: A subscription to the AI ​​assistant costs ten dollars per month, but offering the service costs Microsoft twenty to eighty dollars per month per user. GitHub Copilot is used for coding by about one and a half million developers. For Microsoft, this means losses in the millions per month.

Microsoft isn’t the only one spending more than it makes on generative AI. OpenAI’s ChatGPT is also loss-making: without Microsoft’s generous investments, the company would be in dire financial straits.

Expensive chips

The problem is twofold. Training AI models primarily requires advanced (and therefore very expensive) chips. GPU market leader Nvidia is one of the few companies that is already benefiting from the generative AI hype. The company makes tens of thousands of euros in profit per chip sold and makes billions in profits. To solve this, Microsoft is said to be working on its own AI chip, which it may unveil during Ignite.

But it’s not just chips that are expensive. AI models require a lot of data and computing power, which must come from the cloud. To illustrate, processing each search query you enter into ChatGPT to get a correct answer costs 36 cents. Add up the many questions ChatGPT receives per day and you quickly reach a daily cost of $700,000. In order to make its AI services easier, Microsoft is also said to be working on a leaner backend.

Good example

The example below appears to be Adobe. Adobe is already managing to make its Firefly products profitable. The Company sets a “limit” for use of the AI ​​Tools based on your monthly subscription price. If you exceed this limit, the performance of the tools will be significantly reduced.

With a little imagination, you can compare this method to the limits that mobile providers set for “unlimited” data: if you exceed the magic limit, your connection will become noticeably slower. Adobe is using this measure to “hedge itself on the cost side,” the company told the Wall Street Journal.

In any case, the high costs and lower sales will not stop Microsoft from continuing to roll out generative AI services. The last major Windows 11 update introduced Windows Copilot and from November the AI ​​assistant will also be available for 365 applications. Microsoft wants to charge an additional fee of thirty euros per month for this. Whether this is the profitable formula remains to be seen.