AI token costs: what mid-market finance teams are missing

Token costs and the CFOs who didn’t see them coming

The average monthly AI spend by mid-sized firms grew 36% between 2024 and 2025, hitting $85,000 a month. Yet 61% are still tracking those costs in spreadsheets, according to CloudZero. A separate CloudZero report found 71% of companies exceeded their AI budgets in 2025.

Agentic adoption was always going to drive this kind of demand, but the finance function wasn’t watching. Uber’s CTO disclosed last month that 5,000 engineers burned through the company’s entire 2026 token budget in four months. ServiceNow hit the same wall. Deloitte found over half of CFOs now name cost management as a top-three internal risk.

aibl’s read: Mid-market finance functions were built around predictable fixed-seat contracts, and agentic workflows break that assumption at the infrastructure level. Consumption multiplies with use, varies with complexity, and compounds across every workflow running in parallel. Monitoring tools tell you what you spent. They don’t tell you what you’ll spend next.

Firms treating token costs as a technical detail rather than a financial one are already exposed. Getting costs under control means building the ability to act on them, not just to see them.

Has the iPhone moment arrived for AI services?

If you sold standalone GPS devices in 2005, the world was your oyster. By 2007, everyone had your product bundled in their smartphone, for ‘free’. Big moves by OpenAI and Anthropic are making people in the AI services world ask whether they’re next.

This week both companies announced rival efforts to integrate their own people and services into the AI services and deployment markets. Anthropic has seeded its semi-autonomous entity with $1.5B, backed by Goldman, Apollo, and Sequoia. OpenAI is in for $4B, with Bain and TPG among the investors. The detail is well covered by Axios.

This might speed adoption for mid-market firms, with Anthropic specifically saying its entity will focus there.

aibl’s read: These investments are a clear warning to AI services providers. Consulting firms that have so far led their clients’ AI approaches should pay attention. Like Apple making high-value apps native, the model providers could be on the verge of competing directly with the companies building on their infrastructure.

A European lens on AI’s productivity and employment impact

The working paper from the European Investment Bank is worth an hour of your time. If you don’t have one, here are the key findings.

For the research behind this, see aibl’s research on AI in UK finance and operations.

• AI causally lifts labour productivity by 4%. The study uses a method that isolates exogenous adoption, confirming the lift holds after controlling for the ‘good firms adopt AI’ selection problem.
• Jobs aren’t going anywhere, at least not yet. The data shows zero significant employment effect. AI is augmenting workers, not replacing them in the short run.
• Workers are capturing some of the gains. AI-adopting firms pay higher total wages and higher wages per employee, though whether that persists long-term is an open question.
• Size is destiny. Medium and large firms capture statistically significant productivity gains. Micro and small firms show nothing. Europe’s SME-heavy structure means most firms may be locked out of the upside.
• Training spend produces the strongest return. One extra percentage point of investment in employee training amplifies AI’s productivity effect by 5.9%. Software and data investment adds 2.4%. Hardware subsidies alone don’t change the result.
• Financial market development predicts AI adoption. Firms in financially underdeveloped EU countries lag badly. They stopped catching up after 2022, suggesting capital access is a binding constraint on diffusion.
• Aggregate macro projections are probably too optimistic. The 4% firm-level gain aligns with mid-range estimates, not ‘productivity boom’ scenarios. Adoption frictions and skill gaps absorb much of the potential before it reaches GDP.