Technology executives have long predicted that the cost of artificial intelligence would decline as the technology advances and becomes more widespread. This prediction has partially come true: AI adoption is growing, data centers are expanding computing power, and the cost of running AI models—known as inference—is dropping. However, AI usage, particularly for the most sophisticated tasks, is increasing so rapidly that businesses are now paying more than ever through the use of tokens, the fundamental units of data that large language models read and produce.
The Token Economy: How AI Billing Works
AI tokens are the basic units of text or data that models like OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini use to process language and generate text-based answers, code, and images. For instance, when a software engineer asks Claude to produce code for a project, the model breaks down the prompt into smaller data units to understand the request and generate output. Complex tasks such as coding or creating agentic AI workflows—where AI agents perform multi-step requests with minimal human intervention—consume tens of thousands or even hundreds of thousands more tokens than a simple chatbot query, according to Andrew Forde, head of AI in Canada at KPMG LLP.
“If you’re writing a new piece of software, you could hit 150,000 tokens versus 100 tokens for a paragraph-long text response from an AI,” Forde said. Users are charged for both input and output tokens, meaning companies pay for every request or prompt their employees make to AI, as well as the resulting response.
Skyrocketing Costs and Corporate Responses
In a now-viral example, one company reportedly spent US$500 million in a single month on Claude licenses from Anthropic PBC, according to Axios. Other companies are withholding employee raises due to unexpected AI spending or stating that human employees are now cheaper to hire than deploying AI. The surge in token consumption is driven by businesses increasingly asking AI to handle more sophisticated requests, such as coding and agentic AI workflows.
“It is orders of magnitude higher to do the workflows that businesses are starting to experiment with,” Forde said. “When you start to do more agentic work, where you have multiple AI agents going off and doing deep work to get the outcome you asked for, this has a steep cost in usage.” For example, Royal Bank of Canada—the world’s third-most advanced bank in AI, according to the Evident AI Index—reported that its AI token use jumped more than 500% in its second quarter. Meanwhile, Shopify Inc. chief technology officer Mikhail Parakhin noted in an April podcast that the company has achieved 100% daily AI tool usage across its staff and funds “unlimited tokens for everybody.”
