The End of the Blank Check: Why Tech Firms Are Finally Metering AI

For a brief, intoxicating period, the technology sector convinced itself that generating endless streams of artificial text was synonymous with productivity. The phenomenon even earned a moniker: “tokenmaxxing.” Corporate leaders, terrified of appearing left behind, encouraged employees to interact with artificial intelligence models as much as possible, with some going so far as to erect leaderboards to gamify the cash burn. Now, financial gravity is reasserting itself.

Australian software developer Atlassian has decided that handing employees an unlimited corporate card for compute power might not be a sound business strategy. The company is rolling out digital “AI wallets” for its research and development staff, imposing monthly spending caps ranging from $500 to $2,000 across four products, including Claude Code. Once the funds are exhausted, access pauses until a top-up is approved, though management has yet to deny a request. A company spokesperson framed the move diplomatically, stating that Atlassian provides a significant budget for our builders to leverage multiple AI tools and insisting the caps actually represent an increase in authorized spending. Yet, the underlying message is clear: the era of unmetered experimentation is over. Notably, the firm recently eliminated 1,600 jobs, explicitly citing artificial intelligence as a contributing factor.

Atlassian’s sudden interest in accounting stands in stark contrast to the broader industry's recent excesses. Uber reportedly depleted its entire artificial intelligence budget within four months. Amazon management has been forced to instruct staff to stop using the technology merely for the sake of using it. The underlying unit economics explain the panic. A token, the basic unit of measurement for a model's response, roughly equates to four characters; processing a document the size of the US Declaration of Independence consumes about 1,695 tokens. While OpenAI charges $5 per million tokens for its flagship GPT-5.6 Sol model, and Anthropic demands $10 for its Claude Fable and Mythos equivalents, the volume of automated requests is exploding.

The culprit behind these soaring costs is not just chatty employees, but autonomous agents. As Gartner analyst Arun Chandrasekaran points out, systems are now spawning their own sub-agents to complete independent tasks, generating massive chains of prompts. Even as the baseline price of models drops, the sheer volume of automated consumption is driving bills into the stratosphere. Consequently, firms are scrambling to shift simpler tasks to cheaper, open-source, or open-weight models that can run locally.

The broader corporate landscape is slowly waking up to this fiscal drain. A recent Elastic survey of 500 senior Australian executives revealed that 80 percent fear high token consumption is being falsely equated with actual productivity gains. More tellingly, 32 percent of respondents have already paused, cancelled, or rolled back their artificial intelligence deployments due to spiralling costs. Currently, a mere 9 percent of Australian organisations impose limits on token or API consumption. Atlassian’s metering system is an outlier today, but it represents a necessary maturation. Throwing venture capital at an algorithm does not magically generate a viable business model, and basic cost control remains a prerequisite for corporate survival.

Written by Andreas Hofer andreas.hofer@alpineweekly.com