AI bills surge as token pricing boosts corporate costs

Companies including Uber report monthly AI bills climbing from tens of thousands to hundreds of thousands of dollars as major language model providers shift to usage-based token and per-request pricing. The change is appearing on corporate invoices as pilot projects move into regular production.

IT leaders say that two years ago many firms secured lower, fixed or promotional rates and rolled out internal AI tools. As pilots scaled into production and teams adopted agent-driven workflows, monthly costs that once ranged around $10,000 to $50,000 have risen to mid- or high-six-figure sums at some larger organizations. Per-token billing causes costs to increase with the number of model calls, adding variability to budgets.

Raymond Kok, chief executive of Mendix, discussed on a podcast that companies are shifting from experimentation to identifying business-relevant AI use cases and budgeting for software agents in a similar way to staff costs. Kok said he expects prices to rise further after major model providers pursue public listings.

The pricing shift is prompting changes in governance and IT practice. Firms are restricting access to certain models, tracking token consumption by production agents, and evaluating whether high-volume or sensitive workloads should run on dedicated on-premises servers to limit variable cloud fees. Some IT teams are considering hybrid setups that use public models for lower-volume tasks and local models or hardware for heavy use.

Vendors and cloud providers are commercializing more capable models, and some customers expect promotional pricing to end as providers scale and seek returns following large financing events. Finance and procurement teams are increasingly requesting clearer forecasts of AI spend and introducing internal chargebacks to allocate costs across business units.

Organizations are weighing which AI tasks justify continuous, high-volume model calls and which can be redesigned to reduce token use or moved to architectures with steadier pricing. Those choices affect budgeting, governance and where companies run their AI workloads.