AI tokens are catching some companies off guard and spurring high charges – National

AI For Business


As the use of artificial intelligence (AI) increases in many companies, some are reportedly facing a severe shock as they run out of AI “tokens”, which could lead to additional costs.

Last week, Bloomberg reported that Uber is putting limits on the use of AI by its employees after burning through its 2026 AI coding budget in just four months, part of a growing number of US media reports about companies being surprised by the costs of using up AI tokens sooner than expected.

AI tokens are used in large-scale language models (LLMs), such as ChatGPT and Anthropic’s Claude, and act as a type of meter that AI providers use to bill companies based on usage of the tool’s features.

As a result, unexpected costs can quickly add up for businesses.

Story continues below ad

“We’re seeing a lot of these mistakes happening, so they[companies]are buying licenses, they’re doing pilots, but they’re not actually changing their behavior,” said Patrick Farrar, CEO of AI for Canadians and head of AI at DMZ, a technology incubator and startup ecosystem at Toronto Metropolitan University.

“They’re not changing the workflows within the organization, they’re not training the people within the organization, and that’s really where you start to see costs.”

This also comes as Ottawa’s recently announced National AI Strategy aims to foster business growth in Canada, including through the introduction of AI into business workflows and the education of workers and students.

So how exactly do AI tokens work?

Related videos

Most consumers and businesses have access to some free versions of AI models with reduced or limited functionality. However, for businesses, using AI models in commercial operations requires an enterprise license, which can be acquired through a contract for a fixed number of tokens or on a pay-as-you-go basis.

Story continues below ad

If a business or consumer runs out of tokens, they will be charged a fee each time additional tokens are consumed.

Never miss the day's top stories with Canadian news delivered to your inbox every day.

Get daily national news

Never miss the day’s top stories with Canadian news delivered to your inbox every day.

Farah said a typical prompt that spans one or two sentences and generates a short response in an AI model like ChatGPT might use a fraction of one token per word, while a longer conversation could use as many as 1,000 tokens.

He says this can cost anywhere from 0.5 cents to 2 cents, but for more complex requests involving visualization and design, it can get dramatically more expensive.

“We’re now seeing companies actually running real-world (AI) workflows every day, and we’re going a step further and seeing them using what we now call agent tools, tools that run autonomously,” Farrar said.

“And those kinds of agents, or these kinds of autonomous tools, can now run all day, every hour, all week, all month. So even one type of conversation can burn 50 to hundreds of times more tokens. That’s why the financial industry may be starting to look at the bill now.”


Click to play video:


Business issue: Federal government plans to address AI copyright issues


Story continues below ad

Canadian companies are looking to innovate with AI wherever possible while maintaining efficient use of their AI token budgets.

“We’ve been pushing a lot of organizations to start working with AI tools so they can accelerate their use of AI, or at least increase their knowledge and prepare for the big changes ahead, but that comes at a significant cost,” said Annie Veillette, a partner at PwC Canada who focuses on the AI ​​Engineering and Data Engineering practices.

“Depending on the models available to employees and end users, end-of-life costs were not always well understood.”

Veillet said many companies are using these AI tools to streamline tasks that are considered tedious or relatively easy to perform but time-consuming.

These tools can work well for businesses if implemented well, but when deployed at scale, businesses that don’t keep track of their AI tokens can quickly run out and incur hefty bills.

“Now we have a lot more guardrails installed around us,” she says. “For everyday tasks, use these models, which are much lower cost.”

“For more complex and deep work, while still possible, we recommend using larger language models, which are a bit more expensive, as they offer a greater return on investment.”

© 2026 Global News, a division of Corus Entertainment Inc.



Source link