2025年人工智能成本治理状况研究

2025 State ofAI Cost Governance IntroductionResearch OverviewTop Findings3 Participant Profile31 Table ofContents Conclusion: Preparing for 202634 Cloud & AI Infrastructure6CFO Takeaways AI Cost GovernancePractices and ProcessesThe Revenue Accountability EffectMaturity LevelsVisibility & AttributionCFO Takeaways11 AI Costs: Financial Management& MetricsMeasuring Financial ImpactForecast AccuracyGross Margin ImpactUsage Overage DetectionMonitoring ToolsCFO Takeaways21 AI Cost Crisis: Finance Leaders Face MarginErosion and Forecast Chaos Introduction AI infrastructure costs are already reshaping corporate profitability,but most finance teams lack the visibility and control to managethe impact. This research reveals four critical challenges thatdemand immediate CFO attention: Forecast Failure:85% of companies miss AI cost forecasts bymore than 10%, with nearly 25% missing by over 50%, creatingmassive gross margin risk as AI spending scales. Margin Hemorrhaging:84% of companies report AI costseroding gross margins by more than 6%, with over a quarterseeing hits of 16% or more. For example, a product at 80%gross margin could drop to 74% once AI costs are factored in. Visibility Breakdown:Only 35% include on-premise costs in AIreporting, and half of companies with AI-core products aren'ttracking their LLM API expenses—creating dangerous blindspots in cost-to-serve calculations. Infrastructure Complexity:61% operate hybrid AIenvironments spanning public cloud, private infrastructure, andthird-party services, fragmenting cost visibility and governanceacross multiple vendors and billing systems. The accountability gap is real: Companies charging for AIconsistently demonstrate 2-3x better cost discipline than thosegiving AI features away for free,suggesting that revenue pressuredrives the governance rigor most finance teams desperately need. ResearchOverview 85% of companies cannot forecast AI costswithin 10%. This research, conducted by Mavvrik in partnership withBenchmarkit, surveyed 372 companies to understand howorganizations are building, running, and financially governing AIworkloads. What we found reveals a market in transition: AI hasmoved from experimental budget line to material cost driver, butthe financial discipline hasn't caught up. For CFOs watching AI expenses balloon while gross margins shrink bydouble digits, this isn't just a forecasting problem, it's a strategiccrisis hiding in plain sight. The numbers tell a stark story: Across the full sample (N=372), 84%report AI costs eroding product gross margins by more than 6percentage points (600 bps), with over a quarter seeing hits of 16+points (1600 bps). The stakes couldn't be higher. As AI transforms from "nice to have"to "must have," the companies that master cost visibility andcontrol will protect their margins while competitors watch profitsdisappear into untracked infrastructure costs. Yet most finance leaders are flying blind: unable to predict nextquarter's AI spend, attribute costs to specific products or customers,or even see what's happening across their hybrid infrastructureenvironments. **Note: Gross margin impact findings reflect product delivery (COGS). About70% of respondents were SaaS and AI-native vendors, where inference, GPU,and API costs directly affect gross margin. For enterprises using AI internally,financial impact typically flows through OPEX and operating margin instead. Repatriation is becoming mainstream 67% of companies are actively planning to repatriate some AIworkloads to owned infrastructure, and another 19% are evaluatingthe move. The trend is most active in mid-market companies, whilelarge enterprises are more often in the evaluation stage. TopFindings The AI cost surface is broader than tokens Data platform usage is the #1 source of unexpected AI costs (56%),followed by network access to models (52%). LLM token costs rankfifth (37%). This diversity of cost drivers makes AI spend harder toforecast and control. Visibility and attribution gaps block action AI costs are already eroding gross margins Only ~35% of companies include on-prem components in AI costreporting, and about half include LLM API costs even when AI is acore product component. Teams say the #1 tactic to improve costmanagement is unified visibility across environments; clear costattribution is #2. 84% of companies report more than a 6% hit to gross margin fromAI costs. Within that, 58% see a 6–15% reduction and 26% report16%+ erosion. The financial impact is widespread and immediate,making cost visibility and control a strategic imperative for bothfinance and product leaders. Forecast accuracy is alarmingly low Charging for AI correlates with strongercost discipline Only 15% of companies forecast AI costs within ±10%. A majority(56%) miss by 11–25%, and nearly one in four (24%) miss by morethan 50%. For CFOs and budget owners, this level ofunpredictability makes it harder to protect gross profit targets as AIgrows as