Corporate leaders at the Australian Financial Review AI Summit swapped stories of generative tools that once looked cheap, only to be overtaken by far more complex models that chew through money as fast as they process data. Agentic AI, pitched as tireless digital helpers, is pushing companies into a new era of budgeting, metering and restraint.
Behind the excitement sits a technical and economic shift. Early generative tools such as ChatGPT, launched just four years ago, looked relatively affordable because they handled short, contained tasks.
New agentic AI models work differently, acting as independent systems that can complete multi-step workflows from start to finish without human nudges. An AI agent can audit an entire database, identify an invoicing error, then draft and send an email to a supplier to fix it.
Each step consumes tokens, the fundamental units of data that AI models use to read, write and reason. The cost of these tokens climbs in line with consumption.
The financial risk comes from how brutally usage-based these systems are. An unmonitored agent can burn through thousands of dollars in tokens working on a single complex problem, especially when given broad access and long prompts.
Power and infrastructure demands grow as models get larger and more capable, further inflating operating costs. AI engineers and finance teams now talk in terms of tokenomics, designing guardrails, budgets and monitoring dashboards to track every request and cap runaway jobs.
That discipline is turning into a core competency, not a niche concern, for any enterprise deploying advanced AI at scale.
Pricing tactics from major AI providers are amplifying the anxiety. At the summit, ASX 200 executives compared the experience to dealing with a dealer, the first hits of AI came cheap or free then prices jumped once organisations were hooked and token usage exploded.
In the United States the tech industry is full of cautionary tales, from companies like Uber reportedly blowing an annual AI budget in four months to another firm said to have racked up a $US500 billion token bill in a single month. Australian outfits feel the squeeze as well, with leaders at Commonwealth Bank and other large institutions openly tracking token consumption as a board-level risk.
