Companies are grappling with surging AI token costs, but early adopters show AI investment can pay off if managed strategically.

What happened: AI usage is exploding across companies—Royal Bank of Canada saw token usage jump 500 percent over six months, and Cisco's CEO said token usage is "getting pretty, pretty crazy." About 300 companies mentioned tokens in earnings calls or analyst discussions in April or May, compared with just 93 a year earlier. Meanwhile, 8x8 has saved about $5 million（約8億円） in annual costs by replacing dozens of software tools with Claude, and Baseball Lifestyle 101 is spending the equivalent of about 20 percent of top managers' salaries on AI tokens monthly, which the cofounder says has already driven a $1 million（約1.6億円） order.

Why it matters: Companies are now treating token costs (the price of AI analyzing and generating content) as a major budget concern and competitive lever. Some firms like Meta, Uber, and Salesforce have begun introducing usage caps due to cost pressure, while others are banking on AI-driven efficiency to offset the expense. The tension is real: AI can automate work and cut overhead, but unbounded usage can become expensive, and measuring whether productivity gains outweigh token costs remains difficult.

What to watch: 8x8 is considering whether to require employees to prove that older, cheaper AI models won't work before they can use Anthropic's Opus model, which costs nearly 1.7 times more than an offering released in February. The company is also monitoring token usage dashboards across teams and holding AI hackathons to spur adoption in lower-using departments like finance and sales, which together account for 28 percent of employees but just 15 percent of token consumption.