Companies Track 'Tokenomics' as AI Usage Costs Surge

A new financial pressure point has emerged for companies deploying generative AI: the rapidly escalating cost of tokens, the units that measure how much content AI models process and generate.

Roughly 300 companies discussed AI token costs during earnings calls in April and May 2025, compared to just 93 companies that mentioned tokens during the same period in 2024, according to a review of transcripts by WIRED. The shift reflects growing concern about managing expenses as AI adoption accelerates across organizations.

Why it matters

Token costs represent a fundamentally different economic model than traditional software licensing. Usage can spike unpredictably as employees find new applications, newer models cost significantly more than predecessors, and prices fluctuate across providers. This creates budget uncertainty that finance teams haven't encountered with conventional enterprise software.

The cost equation varies widely

At communications platform provider 8x8, approximately 1,800 employees use Anthropic's Claude for tasks including email drafting, customer feedback analysis, and code writing. The company estimates it has saved around $5 million annually by canceling subscriptions to dozens of software and educational tools that Claude now replaces, according to Joel Neeb, chief transformation and business operations officer. The company's annualized Claude bill remains "well below" those savings, though Neeb expects costs and savings to eventually converge as usage expands.

Other organizations report dramatic consumption increases. Royal Bank of Canada disclosed that token usage jumped 500 percent over six months. At Cisco, one-third of employees use an internal AI chatbot daily, creating what CEO Chuck Robbins described as "pretty, pretty crazy" token usage. Some senior engineers at analytics firm Amplitude spend thousands of dollars monthly on tokens, CEO Spenser Skates reported.

New management strategies emerge

Companies are developing systems to monitor token consumption and route prompts to the most cost-effective models. At 8x8, all employees can view a dashboard showing their Claude usage compared to colleagues—not as punishment, but to ensure no one falls behind in AI adoption, Neeb explained.

The company may soon require employees to justify using Claude's Opus 4.8 model, released in May and priced nearly 1.7 times higher than the February version. Access might depend on demonstrating that older models cannot accomplish the task.

Clothing brand Baseball Lifestyle 101 takes an aggressive approach, directing its top 50 managers to spend the equivalent of 20 percent of their salary on AI tokens monthly. Cofounder Bill Rom expects costs to exceed $100,000 per month by year-end but cites immediate returns, including Claude identifying inventory gaps that led to a $1 million order.

The productivity paradox

Neeb acknowledged the challenge of attributing business improvements directly to AI, but noted that 8x8 has seen rising customer satisfaction metrics and four consecutive quarters of revenue growth since expanding Claude access. The company monitors usage closely and has warned employees that refusing to develop AI skills will have consequences.

Neeb himself uses Claude to automate a daily email summarizing AI tips from industry sources. After noticing high token consumption, he asked Claude to optimize the automation—which reduced token usage by 80 percent for that task.

These details were first reported by WIRED.