Are We Paying for AI with Our Data?

Artificial intelligence has become an everyday tool for many people. Services such as ChatGPT, Claude, and Gemini can write texts, translate documents, create images, and answer complex questions within seconds. Some of these services can be used for free. At the same time, their providers require enormous data centres and tens of thousands of powerful AI chips, often supplied by Nvidia.

This raises a legitimate question: If running these services is so expensive and some AI companies have yet to achieve sustainable profits, how are these costs being financed? And are users ultimately paying with their personal data?

Why Is AI So Expensive to Operate?

Modern AI models require enormous computing power, both during training and every time they are used. Training a large language model can take weeks or even months. But the costs do not stop once the training is complete.

Every question has to be processed, and every answer has to be generated anew.

On top of that, there are expenses for data centres, electricity, cooling, data storage, networking infrastructure, research, and highly skilled employees. Nvidia benefits significantly from this development because its graphics processors and AI systems are used in many data centres.

The more people use an AI service and the more complex their requests become, the higher the ongoing costs can be. A free service is therefore anything but free to operate.

AI Companies Are Already Making Money But Not Always Enough

The claim that AI providers are not making any money at all is not entirely accurate. Many are already generating substantial revenues. Whether those revenues are sufficient to produce sustainable profits after accounting for all investments and operating costs is another question.

Some of the most important sources of revenue include:

• paid subscriptions for individual users,
• business and team plans,
• fees for the use of application programming interfaces, or APIs,
• cloud and infrastructure contracts,
• licensing agreements and strategic partnerships,
• capital from investors,
• and increasingly, advertising in free offerings.

Business customers are particularly attractive. They pay for higher performance, centralised administration, security features, and contractually guaranteed data protection.

Free consumer services, on the other hand, often serve a different purpose: They raise awareness of the service, attract new paying customers, and provide insights into how people use the technology.

Are Personal Data Being Sold? – I Asked the AI 😊

And this was the answer:

An important distinction needs to be made here: directly selling personal chat conversations to data brokers is not the same as the internal commercial use of user data.

Major AI providers generally explain in their privacy policies the purposes for which they collect and process data and the circumstances under which they share it with service providers. However, this does not automatically mean that personal conversations are being sold as a commodity to just any company.

Nevertheless, user data can be highly valuable to an AI provider. This may include not only the texts users enter, but also information such as usage times, features used, device information, approximate location, feedback on responses, and general patterns of interaction.

The more appropriate term, therefore, is often not data selling, but data monetisation and utilisation.

So, if my data (data that can have a monetary value) is used to train AI models, is it no longer accurate to call that data selling, but rather data utilisation?

But ultimately, isn't it essentially the same thing?

Whether my data is used for advertising purposes or for training AI models, it has economic value for the provider.

What Does Data Utilisation Actually Mean?

1. Improving and Training Models
Depending on the provider, subscription plan, and settings selected, user inputs may be used to improve AI models. Conversations can, for example, reveal where an AI makes mistakes, which answers users find helpful, and which features are used most frequently.

This information does not have to be sold to third parties to have significant economic value. A better model can attract more customers, retain existing users, and strengthen a company's position against its competitors.

2. Personalisation
AI services may use previous interactions to tailor responses and recommendations more closely to individual users. This can be convenient, but it can also lead to increasingly detailed user profiles.

The key questions are whether this functionality is explained transparently, whether users can control it voluntarily, and whether it can be fully disabled.

3. Advertising
Advertising is an obvious financing model for free AI services. Advertisers do not necessarily need to receive a person's name or their complete chat history. To select relevant advertisements, it may already be sufficient to analyse interests, the current conversational context, or previous reactions to advertising.

OpenAI now describes advertising functionality for certain free offerings in its European privacy policy. Depending on the settings available and the user's consent, factors such as the context of a conversation, previous chats, and interactions with advertisements may play a role. According to the company, advertisers receive aggregated information, such as data on views and clicks.

This shows that even without the traditional sale of a chat history, conversations can indirectly contribute to financing an AI service.

What Can We Do as Users?

The most important rule is simple: Never enter passwords, login credentials, patient data, unpublished contracts, or trade secrets into an AI service – whether it is free or paid.

In addition, it is advisable to:

• review privacy and data-training settings,
• make use of any available option to disable the use of your content for model improvement,
• use a temporary chat for sensitive queries, where available,
• and, in a professional environment, use only AI solutions that have been explicitly approved.

It is often recommended to delete chats from the visible chat history. However, this does not necessarily mean that all technical records are immediately removed.

For that reason, users should also check the provider's data-retention periods and any applicable exceptions.