> **来源：[研报客](https://pc.yanbaoke.cn)** # 2025 State of AI Cost Governance Summary ## Core Content The **2025 State of AI Cost Governance** report by Mavvrik and Benchmarkit highlights the growing financial challenges associated with AI adoption across organizations. As AI becomes a core component of business operations, the lack of visibility, control, and accurate forecasting is leading to significant margin erosion and financial uncertainty. The report underscores that AI is no longer an experimental cost line item, but a material driver of profitability and a strategic concern for CFOs. ## Main Findings ### AI Cost Impact on Margins - **84%** of companies report AI costs eroding product gross margins by **more than 6 percentage points (600 bps)**. - **58%** see a **6–15% reduction** in margins. - **26%** report **$16\%+$ erosion** in gross margins. - Example: A product with **80% gross margin** could drop to **74%** once AI costs are factored in. ### Forecast Accuracy Issues - Only **15%** of companies forecast AI costs within **±10%**. - **56%** miss by **11–25%**. - **24%** miss by **more than 50%**. - This level of inaccuracy threatens **gross profit targets** and **financial predictability**. ### Hybrid and Multi-Cloud Complexity - **61%** of companies operate in **hybrid environments**, combining public cloud, private infrastructure, and third-party services. - **Multi-cloud** is now the **standard**, with **AWS** (77%) leading overall usage, but **Azure** (82%) dominates among companies with **> $250M revenue**. - **Hybrid complexity** increases **cost visibility challenges** and **billing fragmentation**. ### Repatriation Trends - **67%** of companies are **actively planning** to move AI workloads to **owned infrastructure**. - Another **19%** are **evaluating** the move. - **Mid-market companies** are more likely to **act**, while **large enterprises** are often in **evaluation**. ### Cost Drivers Beyond Tokens - **Data platform usage** is the **#1 source** of unexpected AI costs (**56%**). - **Network access to models** is the **#2** source (**52%**). - **LLM token costs** are **only the 5th** most common cost driver (**37%**). ### Visibility and Attribution Gaps - **Only 35%** of companies include **on-premise costs** in AI reporting. - **About 50%** include **LLM API costs** even when AI is a core product. - **Unified visibility** is the most cited tactic for improving AI cost management (**33%**), followed by **clear cost attribution** (**22%**). ### Revenue Accountability Effect - Companies that **charge for AI** show **2–3x better cost discipline**. - **70%** of charging companies can track **cost-to-serve precisely**, compared to **29%** of free providers. - **71%** of charging companies use **real-time usage alerts** for overages. - **Charging for AI** drives **strategic cost management**, including **P&L ownership**, **customer-specific cost attribution**, and **pricing decisions**. ## Key Takeaways for CFOs - **AI cost governance** is not optional—it's a **strategic imperative**. - **Budgets exist**, but **attribution lags**, with only **35%** tracking on-prem costs and **50%** reporting LLM API usage. - **Monetization** is linked to **stronger cost discipline** and **higher governance maturity**. - **Visibility** is the **foundation** of effective AI cost governance. - **Hybrid environments** are becoming the **new norm**, requiring **unified reporting** and **cross-platform cost tracking**. - **Forecasting accuracy** is **low** across all company sizes, with **85%** missing forecasts by **more than 10%**. ## Financial Management & Metrics - **59%** of companies measure AI costs as a **percentage of revenue**. - Only **29%** measure AI costs against **COGS**, which is the **most relevant metric** for gross margin. - **Charging for AI** leads to **more precise profitability tracking**. - **Real-time monitoring** is **not universal**, with many companies only detecting overages **after invoices arrive**. - **CSP tools** and **internal dashboards** are **common**, but **specialized platforms** are underutilized. ## Maturity Levels - **Only 34%** of companies have an **advanced** AI cost management program. - **Early stage (30%)** and **developing (36%)** are more common. - **Industry** has a greater impact on maturity than **company size**. - **Manufacturing** leads in maturity (**50% advanced**), while **Financial Services** and **Agentic AI companies** lag (**40% and 38% early stage**). - **Monetized AI products** are more likely to have **precise cost tracking** and **governance policies** in place. ## Conclusion As AI moves from an experimental tool to a **core business driver**, the need for **cost visibility**, **forecast accuracy**, and **governance maturity** becomes **critical**. The report emphasizes that **CFOs must treat AI costs as part of COGS**, not just as innovation expenses. With **85% of companies missing forecasts** and **84% experiencing margin erosion**, the financial discipline required to manage AI is no longer a luxury—it's a **necessity**. Companies that **charge for AI** are more likely to **achieve better governance**, **track costs precisely**, and **make strategic decisions** based on real-time data. To thrive in **2026**, CFOs must **embed cost governance** into every AI initiative and **invest in unified visibility and control systems**.