AINEWS
CFOs Are Starting to Ask Which Software They No Longer Need Because They Have AI
AI spending is the easy part. The harder question is what companies quietly cancel to pay for it.
AI spending is the easy part. The harder question is what companies quietly cancel to pay for it.
The cloud communications company 8x8 gave its employees access to Anthropic's Claude and quietly canceled dozens of software and educational subscriptions. The result, by the company's own estimate, was roughly $5 million in annualized savings, with the Claude bill running "well below" that figure, according to Joel Neeb, 8x8's chief transformation and business operations officer (WIRED).
That math, simple as it sounds, captures a shift that is starting to show up across corporate earnings calls. The interesting question for finance teams is no longer how much AI costs. It is what they can stop buying because they have it.
WIRED's review of recent earnings-call transcripts found that roughly 300 companies addressed AI token questions in analyst discussions in April and May 2026, compared with 93 in the same months a year earlier. Tokenomics, the unit cost of what an AI model reads and writes, has become a new line item in boardroom conversations (WIRED).
At 8x8, the substitution logic ran in plain view. The company, with about 1,800 full-time employees, rolled Claude out across the organization and audited its software stack. Dozens of subscriptions, including educational and tooling licenses, were dropped because Claude could absorb the work. The $5 million figure is an estimate, not an audited number, and Neeb declined to share the exact total Claude spend. He acknowledged that the savings-to-cost ratio will compress as more employees adopt AI and as the work itself gets more complex.
That caveat matters, because the substitution story is not yet uniformly true. Royal Bank of Canada disclosed that token usage surged 500% in six months. Cisco CEO Chuck Robbins said a third of employees now use an internal AI chatbot daily, and "the token usage is getting pretty, pretty crazy." Amplitude CEO Spenser Skates said some top engineers spend "thousands of dollars a month or more on tokens." Box CEO Aaron Levine said "the token budgeting conversation has absolutely taken over as one of the most important" topics his team is having (WIRED).
In other words, the bill is real and growing. What varies is whether companies are offsetting it by trimming other categories of software, training, and services spend.
Baseball Lifestyle 101, a clothing brand expecting roughly $250 million in 2026 sales, has taken a different approach. Cofounder and chief strategy officer Bill Rom told about 50 top managers to spend around 20% of their salary on AI tokens each month. The company expects AI costs to exceed $100,000 a month by year-end, and Rom credits Claude with helping land a $1 million order (WIRED).
That kind of top-down mandate, treating AI tokens as a per-employee budget line, is one of the more visible signs that finance teams are absorbing AI into the planning cycle rather than treating it as an ad hoc experiment.
8x8 is also starting to manage cost inside the AI line itself. Sales and finance, which together make up 28% of headcount, account for only 15% of token consumption. The company is running an internal hackathon aimed at the finance team. Neeb has used Claude to optimize its own YouTube summary automation, cutting token use by 80%.
The next squeeze is already visible. 8x8 is weighing caps on Claude Opus 4.8, a model released last month that costs roughly 1.7 times Anthropic's February release. Future access to the most expensive tier may require showing that older models cannot do the job (WIRED).
That is the part of the curve that will define the next year of enterprise AI budgeting. The early wins came from canceling obvious substitutes, the kind of subscriptions that exist because no general-purpose tool was available. As companies push AI into harder work, where the model replaces a vendor contract but the vendor was already cheap, the math gets tighter.
Meta, Uber, and Salesforce have publicly flagged AI cost concerns and begun introducing usage caps. Workers at Amazon and Meta have alternately rubber-stamped AI use or pulled back, drawing internal criticism about waste (WIRED).
The pattern that is forming in 2026 is not that AI is cheap, and not that it is unaffordable. It is that the unit economics of enterprise software are being rewritten in real time, and the CFOs who learn that math first will set the template for the next decade of corporate IT spend.
The question worth watching is whether 8x8's $5 million figure stays an outlier or becomes the rule.