AI Token Costs Become a New Corporate Battleground

AI adoption is creating a new line item executives can no longer ignore

For much of the past year, the business case for generative AI has been framed in broad terms: faster coding, quicker customer support, better internal search, and less repetitive office work. What is becoming clearer now is that the economics of that shift are no longer theoretical. As more employees rely on large language models every day, companies are discovering that the bill does not arrive as a single software subscription. It arrives as a stream of tokens.

That emerging discipline, often described as “tokenomics,” is moving from technical teams into the finance office. The underlying issue is simple. Every prompt an employee sends to a model, every document the model reads, and every answer it produces adds to usage. At small scale the costs can feel manageable. At enterprise scale, especially when AI tools are embedded into coding, analytics, sales, and support workflows, the totals can rise quickly enough to force hard decisions about budgets and access.

WIRED’s reporting captures that transition in real time. At business software firm 8x8, employees are already using Anthropic’s Claude to draft emails, analyze customer feedback, and write code. The company says it has saved about $5 million annually over the last 18 months by canceling subscriptions to other tools that Claude could partly replace. For now, 8x8 says its annualized Claude spending remains well below that savings figure.

Early winners still expect the balance to change

That does not mean the economics are settled. 8x8 executive Joel Neeb told WIRED he expects savings and AI costs to move closer together over time as adoption expands and more complex work shifts onto the model. That is the key tension in enterprise AI right now. Companies may be able to justify current spending when pilot groups are productive and old software contracts are removed, but wider deployment changes the math.

The pattern is visible across the industry. Royal Bank of Canada disclosed that token usage jumped 500 percent in six months. Cisco said a third of its employees are using an internal AI chatbot daily, with usage becoming “pretty crazy.” Amplitude’s CEO said some top engineers are spending thousands of dollars a month or more on tokens. Box’s CEO described token budgeting as one of the most important and heated conversations now taking place.

Those examples matter because they show the problem is not isolated to one sector. Banks, network infrastructure vendors, analytics platforms, and cloud software companies are all encountering the same constraint: demand for AI tends to grow faster than the controls designed to manage it.

Token management is becoming operational strategy

Once usage spikes, executives have only a few practical levers. They can cap access, restrict premium models, route simpler work to cheaper models, redesign prompts to reduce waste, or demand that teams prove productivity gains against the spend. None of those options is purely technical. Each one affects how employees work and whether AI feels like an open platform or a tightly rationed resource.

That is why tokenomics is becoming an operational question rather than a procurement detail. A company that wants developers to use AI for code generation, marketers to use it for content drafts, and support teams to use it for customer workflows needs rules for who gets what model, for which tasks, and at what cost threshold. Without that discipline, small experiments can turn into an uncontrolled enterprise expense.

At the same time, aggressive caps carry their own risk. If the approved internal tool becomes too slow, too restricted, or too expensive to use freely, employees may work around it or limit adoption to low-value tasks. That would weaken the very productivity gains companies are hoping to unlock.

The next phase of AI spending will be judged more harshly

The first wave of corporate AI investment was driven by urgency. Executives feared falling behind, vendors raced to integrate models into every workflow, and boards wanted visible signs that management had a plan. The next phase is likely to be more demanding. Finance teams will ask which workloads truly justify expensive inference, which use cases can be automated reliably, and where AI is replacing other software versus simply adding a second bill.

The 8x8 example suggests there are companies already finding room to offset costs through consolidation. But the wider market signals a tougher reality: token usage tends to climb as soon as AI becomes useful enough to spread. In that environment, the companies that benefit most may not be the ones with the most enthusiastic adoption. They may be the ones that treat AI usage as a measurable operating system for work, not a vague innovation budget.

Generative AI is no longer just a capability question. For many organizations, it is becoming a metering question. That shift may end up deciding which AI rollouts scale, which stall, and which survive the scrutiny of the next budget cycle.

This article is based on reporting by Wired. Read the original article.