Published Date: September 17, 2025

# CFO Dive: Most firms miss AI cost forecasts, survey finds

New research shows that businesses are struggling to forecast AI costs, with more than half missing targets by 11–25% and nearly one in four missing by over 50%.

CFO Dive’s Alexi Alexis reports on the risk of **significant cost overruns as AI pilots move into production in 2026**, with 84% of companies already seeing margin erosion linked to AI spend.

Click here to read the full story on Revenue Brew: [https://www.revenuebrew.com/stories/2025/09/16/companies-lack-forecasting-ai](https://www.revenuebrew.com/stories/2025/09/16/companies-lack-forecasting-ai)

### Recent Posts

## [How to Track AI Costs in 2026: From Usage Logs to Cost-to-Serve](/content/blog/how-to-track-ai-costs/index.html)

AI cost tracking in 2026 requires more than monitoring token spend or reviewing provider invoices. This guide explains how finance, FinOps, and engineering teams can track AI costs across workflows, customers, and environments using metrics like cost per inference, cost per workflow, and cost-to-serve.

## [FinOps for AI: Build vs Buy](/content/blog/finops-for-ai-build-vs-buy/index.html)

Building AI cost management internally sounds manageable until the integration and maintenance burden becomes clear. This article breaks down the cost, time, and visibility tradeoffs between building in-house and using a purpose-built platform.

## [Google Cloud Next 2026: What You Need to Know About AI Costs](/content/blog/google-cloud-next-2026-recap/index.html)

Google Cloud Next 2026 confirmed that AI is no longer experimental infrastructure. As agentic AI adoption accelerates, enterprises are facing new cost challenges tied to token usage, distributed services, cross-cloud architectures, and continuous inference workloads.