We compiled data from Gartner, Stanford HAI, McKinsey, Deloitte, and independent enterprise surveys to examine the state of AI development cost in 2026. The findings show that 79% of enterprises experienced AI cost overruns in the past 12 months, with 85% systematically misestimating AI costs at the forecast stage. The largest gaps arise not from model licensing, but from data infrastructure, network access, and workforce readiness shortfalls that standard IT budgeting frameworks fail to capture.
This analysis covers global AI investment benchmarks, cost overrun rates, hidden cost drivers, variation by organisation size and financial governance maturity, talent cost factors, and ROI performance relative to development expenditure.
Enterprise AI Development Investment: Scale and Baseline Figures
AI development spending accelerated sharply in 2025, transitioning from selective experimentation to large-scale organisational commitment. The table below presents key investment benchmarks across global and enterprise dimensions.
| Metric | Figure | Source |
| Global corporate AI investment, 2025 | $581.7 billion (up 130% year-over-year) | Stanford HAI AI Index 2026 |
| Worldwide AI spending forecast, 2026 | $2.59 trillion | Gartner, 2026 |
| Worldwide generative AI spending, 2025 | $644 billion | Gartner, 2025 |
| Enterprise generative AI spend, 2025 | $37 billion (up 3.2× year-over-year) | Menlo Ventures, 2025* |
| Share of enterprise GenAI spend directed to AI applications | More than 50% | Menlo Ventures, 2025 |
| High-performing AI organisations allocating 20%+ of digital budget to AI | One in three | McKinsey State of AI 2025 |
The data reveals that AI development costs have shifted from a discretionary R&D expense to a core operational budget line. Investment has accelerated rapidly at both the global and enterprise level, with the generative AI segment growing faster than the broader market. The majority of enterprise AI spend is now directed at deployment and applications rather than infrastructure, meaning active use, not system buildout, is where most development costs are now concentrated.
The Prevalence of AI Cost Overruns in 2026
Despite the scale of investment, most enterprises are unable to accurately forecast what AI development costs. The following data captures the breadth of the cost overrun problem across large enterprise deployments.
| Metric | Figure | Source |
| Enterprises that experienced AI cost overruns in past 12 months | 79% | DoiT/Sapio Research, Feb. 2026† |
| Enterprises missing AI cost forecasts by more than 10% | 85% | Mavvrik/BenchmarkIT, 2025‡ |
| Enterprises missing AI infrastructure forecasts by more than 25% | 80% | Mavvrik/BenchmarkIT, 2025 |
| Enterprises missing AI forecasts by more than 50% | ~24% | Mavvrik/BenchmarkIT, 2025 |
| Enterprises able to forecast AI costs within 10% accuracy | 15% | Mavvrik/BenchmarkIT, 2025 |
| Enterprises reporting significant gross margin erosion from AI | 84% | Mavvrik/BenchmarkIT, 2025 |
The data indicates that AI cost overruns have become the default experience for enterprise organisations, not an outlier event. Most enterprises significantly miss their AI cost forecasts, and the gap between projected and actual spend is large enough to create material financial risk. The most common failure point is not overspent on model costs but on surrounding infrastructure, data operations, and integration layers, which are addressed in the following section.
The Hidden Cost Drivers Behind AI Budget Overruns
Standard IT budgeting typically prioritises model licensing and computing. The evidence shows these are not the primary source of AI development cost surprises.
| Unexpected AI Cost Category | Share of Enterprises Citing as Top Surprise | Source |
| Data platform and pipeline costs | 56% | Mavvrik/BenchmarkIT, 2025 |
| Network access and data transfer | 52% | Mavvrik/BenchmarkIT, 2025 |
| LLM API and model licensing costs | Ranked 5th of unexpected cost categories | Mavvrik/BenchmarkIT, 2025 |
| On-premises AI costs included in reporting | 35% (only) | Mavvrik/BenchmarkIT, 2025 |
| Organisations with mature cost tracking and attribution | 34% | Mavvrik/BenchmarkIT, 2025 |
| Organisations rating data management as highly prepared for AI | 40% | Deloitte State of AI 2026 |
The data shows that the most consequential AI development costs are infrastructure-adjacent rather than model-centric. Data platforms and network costs are the leading sources of budget surprises, while LLM API costs, the costs most anticipated in project scoping, rank fifth. Most organisations’ cost reporting does not capture on-premises AI infrastructure at all, meaning their forecast inputs are structurally incomplete from the outset.
AI Cost Overruns by Organisation Size and FinOps Maturity
Cost overrun rates vary by company size and the sophistication of the financial governance structures in place. The data reveals a pattern that runs counter to what most organisations expect.
| Organisation Segment | AI Cost Overrun Rate | Mean Overspend | Source |
| All enterprises surveyed | 79% | — | DoiT/Sapio Research, 2026 |
| Mid-size organisations (1,000–4,999 employees) | 81% | — | DoiT/Sapio Research, 2026 |
| Large enterprises | 76% | — | DoiT/Sapio Research, 2026 |
| Organisations with mature or leading-edge FinOps | 89% | 30.9% | DoiT/Sapio Research, 2026 |
| Organisations in early-stage FinOps development | 69% | 16.1% | DoiT/Sapio Research, 2026 |
The data displays that AI cost overruns are not a consequence of inadequate financial governance. Organisations with the most sophisticated financial tracking report on the highest overrun rates, most likely because they have the instrumentation to detect what less mature counterparts are not measuring. Organisational scale does not determine cost risk either: mid-size organisations report higher overrun rates than large enterprises despite smaller absolute AI budgets.
Talent and Workforce Readiness as a Structural AI Development Cost
Workforce capability is rarely included in AI development cost estimates, yet it functions as one of the most consequential inputs determining both delivery cost and ultimate return.
| Metric | Figure | Source |
| Biggest reported barrier to AI workflow integration | Insufficient worker skills | Deloitte State of AI 2026 |
| Organisations believing their workforce is truly AI-ready | 20% | Gartner, Dec. 2025 |
| Executives with a comprehensive AI strategy | 27% | Gartner, Dec. 2025 |
| Talent preparedness vs. strategy preparedness for AI | 20% vs. 42% | Deloitte State of AI 2026 |
| Workers in roles explicitly requiring AI fluency (2025 vs. 2023) | ~7 million vs. ~1 million | McKinsey, 2025 |
| Enterprises at risk of losing top AI talent by 2027 without a people strategy | 50% | Gartner, 2026 |
The data shows that talent is the most underprepared dimension of AI development cost, with a significant gap between how ready organisations feel strategically and how ready they are on actual workforce capability. Demand for AI-fluent workers has grown far faster than supply, and organisations that do not budget for workforce development alongside technical deployment face compounding secondary costs: poor system adoption, extended rework cycles, and talent attrition.
AI Development Cost Against Return: The Investment Gap
The final dimension of AI development cost is its relationship to measured return. The evidence indicates a significant disparity between potential return and the returns most organisations are currently realising.
| Metric | Figure | Source |
| Average return per $1 of generative AI investment | $3.70 | IDC/Microsoft [vendor-sponsored], 2025 |
| Median time to positive ROI | 14 months | IDC/Microsoft [vendor-sponsored], 2025 |
| Organisations achieving enterprise-wide financial impact from AI | ~6% | McKinsey State of AI 2025 |
| Enterprises able to calculate AI ROI without significant bottlenecks | 15% | DoiT/Sapio Research, 2026 |
| AI spend accountability: Technology vs. Finance leadership | 55% vs. 53% | DoiT/Sapio Research, 2026 |
| Organisations with mature cost attribution practices | 34% | Mavvrik/BenchmarkIT, 2025 |
The data reveals that the commercial case for AI investment is substantial, but most organisations are not yet realising returns at the level the data suggests is possible. A small minority have achieved enterprise-wide financial impact from AI, and most cannot accurately measure what their AI development spend is generating. Accountability for AI spend is split between Technology and Finance leadership in ways that tend to reduce forecasting rigour, compounding the cost visibility problem identified throughout this analysis.
To learn more or request an AI development cost assessment, please reach out here.
Sources
- Stanford HAI. AI Index Report 2026. Stanford University, April 2026. https://hai.stanford.edu/ai-index/2026-ai-index-report
- Gartner. “Gartner Forecasts Worldwide AI Spending to Grow 47% in 2026.” May 2026. https://www.gartner.com/en/newsroom/press-releases/2026-05-19-gartner-forecasts-worldwide-ai-spending-to-grow-47-percent-in-2026
- Gartner. “Gartner Forecasts Worldwide GenAI Spending to Reach $644 Billion in 2025.” March 2025. https://www.gartner.com/en/newsroom/press-releases/2025-03-31-gartner-forecasts-worldwide-genai-spending-to-reach-644-billion-in-2025
- Menlo Ventures. 2025: The State of Generative AI in the Enterprise. December 2025. https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/
- DoiT International / Sapio Research. “Why 79% of Enterprises Overspent on AI in 2026.” February 2026. https://www.doit.com/blog/ai-spending-survey
- Mavvrik / BenchmarkIT. 2025 State of AI Cost Management. September 2025. https://www.mavvrik.ai/news/2025-state-of-ai-cost-management-research-finds-85-of-companies-miss-ai-forecasts-by-10/
- Deloitte AI Institute. State of AI in the Enterprise 2026. 2026. https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html
- McKinsey & Company. “The State of AI: Global Survey 2025.” November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- Gartner. “Gartner Predicts by 2027, 50% of Enterprises Without a People-Centric AI Strategy Will Lose Their Top AI Talent.” May 2026. https://www.gartner.com/en/newsroom/press-releases/2026-05-13-gartner-predicts-by-2027-50-percent-of-enterprises-without-a-people-centric-ai-strategy-will-lose-their-top-ai-talent
- IDC / Microsoft. Business Opportunity of AI. 2025. [URL: to verify before publishing]
Client Note: The cost overrun statistics in Sections 2, 3, and 4 draw primarily on two commissioned survey studies: (1) DoiT/Sapio Research (February 2026, n=500 finance leaders), commissioned by DoiT International, a cloud management services company; and (2) the 2025 State of AI Cost Management, commissioned by Mavvrik, an AI cost governance platform (n=372 enterprise organisations). Both studies disclose their methodologies and sample compositions, and their findings are directionally consistent with each other and with broader enterprise AI reporting from Gartner and Deloitte. However, the commissioned context should be noted when citing these figures publicly. No tier-1 primary research institution has published equivalent granular data on AI cost overrun rates specifically, making these the most authoritative available sources for this dimension of the piece.
