AI Is Already Costing Companies Real Money, and the Bill Is Growing Fast

The Token Bill Arrives

For the past two years, corporate America has been told AI will save money. Some companies are finding that's true — so far. Most are discovering the fine print.

According to Wired, roughly 300 companies mentioned AI token costs during earnings calls or analyst discussions in April and May 2026. That's more than triple the 93 companies that brought up tokens in the same window a year earlier.

The numbers people are throwing around are not small. Royal Bank of Canada CEO Dave McKay disclosed that token usage surged 500 percent over just six months. Cisco CEO Chuck Robbins said a third of his employees are using an internal AI chatbot every day, and "the token usage is getting pretty, pretty crazy." Aaron Levie, CEO of Box, called token budgeting "one of the most important" and "heated" topics at his company.

At analytics software firm Amplitude, some top engineers are "spending thousands of dollars a month or more on tokens," according to CEO Spenser Skates, as reported by Wired.

One Company Actually in the Black

Not everyone is in the red. Software company 8x8 told Wired it has saved roughly $5 million in annual costs over the past 18 months by canceling subscriptions to dozens of software and educational tools that Claude could replace. Their annualized bill for Anthropic's Claude is currently "well below" that figure, according to Joel Neeb, the company's chief transformation and business operations officer.

Neeb is candid that this gap won't hold forever. He expects costs and savings to converge as 8x8 expands adoption and takes on more complex AI workloads. He declined to give Wired exact total spending figures.

Most companies haven't reached that level of transparency yet.

Qualcomm's Bigger Bet

While CFOs argue about token budgets, Qualcomm CEO Cristiano Amon is making a much longer-range call: the smartphone as we know it is not the endpoint.

Amon told CNBC that Qualcomm is currently working on designs for over 40 new AI-powered devices. The form factors include jewelry, earbuds with embedded cameras, wearable pins, and watches. The common thread: something you wear constantly, something that can see your surroundings, and something that gives you immediate access to an AI agent.

"The principle is something that you wear, something that is with you all the time, something that can see the world around you," Amon told CNBC's The Tech Download podcast.

On apps, Amon was direct. They're "not dead," he said, but "those agents are going to be the new app." His example: instead of opening a banking app and navigating to a transaction, you ask an agent and it retrieves the information instantly.

The implications for Apple and Samsung are real. If agents become the primary interface, the app store model — which both companies have built enormous businesses on — faces structural pressure.

It's Already in People's Homes

ZeroHedge published a first-person account by Howard Armitage via New Atlas that illustrates where consumer AI actually stands today, away from the boardrooms. Armitage describes using ChatGPT to mock up a fence design for a neighbor in under a minute, using Grok mid-conversation to pull up a dive computer, and running a local AI through Home Assistant (using the "Nabu" voice interface) to control home devices with natural language rather than rigid pre-programmed commands.

The local-AI angle matters. Armitage specifically flags that running AI locally keeps data inside the home network rather than routing it through commercial platforms. That's a real privacy consideration that the corporate tokenomics conversation ignores entirely. The consumer use case and the enterprise use case are diverging in meaningful ways.

The Legitimate Concern

Skeptics — and there are reasonable ones — argue that the productivity gains from AI are largely unproven at scale. Token costs are concrete and measurable today. The savings are often estimated, projected, or based on soft metrics like "subscriptions we canceled." 8x8's $5 million figure is management's own estimate, not an independent audit. Companies may be rationalizing costs they would have cut anyway and crediting AI for the savings.

That concern is fair. The accounting here is genuinely murky. Most companies Wired surveyed are still trying to figure out how to monitor token usage and route prompts to cheaper models. Basic cost discipline should have been in place before deploying at scale.

At the same time, a 500 percent surge in token usage at Royal Bank of Canada is not a projected number. People are using these tools. The question is whether the output justifies the spend, and that answer varies company by company.

What the Numbers Don't Settle

The unresolved question sitting underneath all of this: as AI agents take over more tasks, do companies hire fewer people, or do they just do more with the same headcount? Several executives Wired spoke with said they are still deciding whether to expand token budgets or expand hiring to meet their goals. That decision, made across hundreds of companies simultaneously, will matter for the labor market in ways that no earnings call has fully addressed yet.