Microsoft recently made a quiet change to the terms and conditions for its online services to add a warning that it might restrict access for “excessive” users of its generative AI services. The change in the tech giant’s legalese was discovered by Cloudy With A Chance Of Licensing, a license watcher with a focus on Microsoft’s licensing.
The updated terms do not specify what counts as “excessive use” and how much use of Microsoft’s AI services is acceptable.
The document doesn’t offer any clarification on the exact restrictions Microsoft will be enforcing or how long they will last.
The terms and conditions document reads, “Excessive use of a Microsoft Generative AI Service may result in temporary throttling of Customer’s access to the Microsoft Generative AI Service”.
However, that’s all the document says about the potential access restriction of excessive users.
The vagueness leaves a lot of questions regarding how exactly Microsoft plans to throttle its generative AI services. However, other AI providers like OpenAI have already enforced usage restrictions, which might offer some clues.
For instance, OpenAI’s API is rate-limited by request limits and spending limits that vary depending on the usage tiers. For instance, those using the free tier of the API are limited to 3 requests per minute and 200 requests a day. Tier 5 users enjoy the highest request limits – 10,000 requests per minute and no cap on daily requests.
An excessive number of requests from a single person or organization can bog down the API for everyone else, an OpenAI document on rate limits explains.
OpenAI’s terms of service also state that if a user is believed to be “not using the free tier in good faith”, the company may start charging them standard fees or limit their access to the service.
Similarly, AI image generator Midjourney’s term of service includes a clause that states that the company would try to reasonably offer unlimited access to its services. However, Midjourney reserves the right to apply rate limits to prevent disruptions or quality issues for other users.
A Potential Sign of Bottlenecks in Microsoft’s AI Architectures
The fact that Microsoft needs to throttle access for “excessive” users might potentially hint at bottlenecks in the company’s AI architectures.
Users updating Windows to start using its new AI Copilots might have to deal with a less-than-ideal service due to degraded performance caused by an overload of Microsoft’s systems. This, in turn, could potentially deter users from using the company’s AI services in the future.
The other probable reason behind the throttling is simply that Microsoft might be trying to curtail its costs.
The company had previously warned users that scaling up its cloud-based AI infrastructure would be an expensive deal.
The GPUs needed to power AI systems account for a majority of the costs, alongside the high energy requirements. During its earnings call on October 24, the company revealed that its capital expenditures in Q1 2024 had reached $11.2 billion.
Part of this sum had been consumed by Microsoft’s investments in scaling its AI infrastructure, the tech giant claimed.