Fusion5 warns that the period of free or subsidized AI has ended. As providers transition to token-based pricing models, AI operating costs will become a major budget item for enterprises.
Fusion5 CEO Kristy Brown stated that over the past three years, businesses have essentially benefited from “the most expensive tech experiment in history” funded by AI companies, but costs are now gradually being shifted to customers.
Tools like Claude, Gemini, Microsoft Copilot, and various coding or customer service agents are adopting token-based pricing models or a combination of subscription and usage-based fees.
According to Ms. Brown, unlike traditional software, every prompt, every reasoning step, and every action by an AI agent generates real costs. Autonomous agents can even execute transactions or make decisions on behalf of businesses.
An agent stuck in a reasoning loop can consume a massive amount of tokens and generate significant costs before anyone notices.
Fusion5 recommends that businesses build an ontology, a structured knowledge layer that fully describes the enterprise such as organization, processes, data, brand, and expert knowledge. This helps AI find the right information from the start, reducing reasoning steps and cutting computational costs.
If data is lacking or enterprise knowledge is not digitized, AI will spend a lot of time searching, which both increases costs and easily creates errors. According to Fusion5, businesses will “pay twice”: once for computing power and once for the resulting errors.
The article states that corporations like Tesla, Uber, Meta, and Amazon have begun restricting employee use of AI agents to control costs and evaluate investment efficiency.
Fusion5 cited data suggesting that four major US tech giants expect to spend nearly $700 billion on AI infrastructure this year. Meanwhile, OpenAI is reportedly spending about $60 billion annually on computing power while generating only about $13 billion in revenue. Microsoft also announced that Azure exceeded $100 billion in revenue for the first time and Microsoft 365 Copilot reached over 30 million paid licenses.
Fusion5 proposes establishing an Agent Operations Centre (AOC) to monitor AI agents in real time, track costs, set up alerts, control safety guardrails, and hand over processing to humans when agents operate outside allowed boundaries. The company plans to launch this service in the fourth quarter, targeting about 1,400 customers in New Zealand and Australia.
📌 The enterprise challenge is shifting from “whether to use AI” to “how to operate AI efficiently and control costs.” As AI agents become increasingly autonomous and token-based pricing models become widespread, AI governance will become a new operational function. Businesses that invest in data, internal knowledge, and agent monitoring systems will have a greater advantage as AI costs are no longer subsidized by providers as before.
