How much does generative AI actually cost?

OpenAI recently announced its new text-to-video generator, Sora. The launch has highlighted the enormous potential of generative AI, especially when it comes to streamlining content creation. This raises questions about the future of content creation. Generative AI has the potential to revolutionize how we create text, images, and video, but it also carries hidden costs.

One of these is the cost of energy consumption. To understand AI's energy consumption, we have put it into context. For example, sending an email consumes an average of 0.2 kWh in server capacity. That may sound like a small amount, but a quick Google search (which in turn uses about 0.003 kWh) shows that 347 billion emails are sent every day. That means energy consumption higher than Sweden's total energy consumption in 2022.

Similarly, using AI comes at a cost. AI models require enormous amounts of data and energy to train and run, which means that an AI-generated text, image, or video can require large amounts of data and server resources. An AI-generated text requires approximately 0.02 kWh, an image around 2 kWh, and a one-minute AI-generated video can consume up to 200 kWh. To put this into perspective, a typical 100 m² Swedish house consumes an average of 33 kWh per day. In other words, a single AI video can correspond to almost a week's electricity consumption for the average person.

• Washing a load of laundry at 40 degrees consumes an average of 1.2 kWh
• Showering for 5 minutes consumes an average of 2-4 kWh
• Driving an electric car for 10 km consumes an average of 1.5 kWh.

Generative AI or human resources?

In some cases, generative AI can definitely be more resource-efficient than using human resources, as Klarna demonstrated when they released statistics after the first month of using their AI-powered customer service assistant. For example, generating a simple image or text with AI can also be faster and cheaper than hiring a designer or writer, and AI can perform repetitive tasks more efficiently and without the risk of human error.

But when does it tip over? There are also situations where it is more resource-efficient to use human resources. For example, creating a complex video or text with AI can require enormous amounts of computing power and energy, and have a greater impact on the environment than using human creativity. In addition, it rarely gets it right on the first prompt, and several generated video drafts will be needed to achieve a good end result. By using their experience and creativity, a human resource can produce the same end result with significantly less energy consumption and environmental impact.

Who pays the price?

Today, however, it is not the user who bears the costs of this. Most AI systems are run by large technology companies that can afford to invest in the expensive infrastructure required to run AI models. They can also use profits from their other activities to subsidize the cost of AI services. Even though AI services are often free or low-cost for the user, it can be good to keep in mind what they actually cost and what impact your own use of AI has.

The conclusion is that both AI and human resources have their own strengths and weaknesses in terms of resource efficiency, and by carefully considering both the economic and environmental consequences of using AI in relation to human labor, we can use AI in a more efficient and responsible way.

Want to read more about AI? Here you will find all our articles on the subject.