AI cost chaos stalls enterprise ambitions

A pervasive lack of cost visibility severely hampers CEOs and business leaders in managing AI budgets, with many admitting only a limited understanding of their AI expenditures. According to KPMG's Steve Chase, the challenge is less about rigid cost control and more about achieving clear visibility into where AI dollars are spent and the outcomes they generate, underscoring cost transparency as a critical leadership priority amid rising AI bills.

The transition from predictable, flat-rate SaaS pricing to complex, consumption-based AI billing models has introduced significant financial unpredictability. CFO Greg Henry of 1Password highlights how token-based pricing disrupts traditional budgeting by causing unpredictable cost spikes, a sentiment echoed by GitHub's shift of Copilot to usage-based billing, which led to unexpectedly large invoices and prompted some enterprises to seek cheaper alternatives.

Fragmented billing systems and limited integration of AI usage data into financial reporting tools further complicate cost tracking and forecasting. Much of the detailed consumption data remains locked within vendor dashboards rather than accessible financial systems, making it difficult for finance teams to monitor expenses effectively, as noted by BankInfoSecurity and industry experts like Flexprice CTO Nikhil Mishra, who criticizes vendors for failing to set clear expectations about real consumption costs.

The absence of clear ownership and enforcement of AI cost governance exacerbates budgeting challenges, with over half of organizations lacking a designated AI cost owner and less than half fully enforcing cost policies. This governance gap, combined with inconsistent billing metrics across multiple providers and limited cost awareness among engineers—as only 45% understand the costs of what they build—contributes to financial unpredictability and stalled AI deployments when costs outweigh expected value.

The transition from traditional flat-rate licensing to usage-based and tokenomics pricing models has fundamentally disrupted cost management for enterprises, introducing complex, non-linear variables that traditional IT budgeting frameworks cannot handle. As Steve Chase of KPMG emphasizes, the focus has shifted from rigid cost control to achieving comprehensive cost visibility, understanding where AI spend is directed and the outcomes it produces. However, this new pricing landscape has led nearly a quarter of enterprises to struggle with managing these costs, with 42% reporting only partial visibility, and nearly half admitting to stalling AI deployments due to unexpected budget shocks that often exceed initial projections.

Major AI vendors like Anthropic, OpenAI, and GitHub have moved away from predictable subscription fees toward usage-based billing, significantly complicating financial forecasting and inflating software budgets. Bain & Company projects AI datacenter build costs to reach $2 trillion by 2030, with 80% of decision-makers anticipating rising data and software expenses that frequently surpass human labor costs. This shift has forced CFOs such as Matthias Steinberg of MindBridge and Patrick Villanova of BlackLine to grapple with the unpredictability of token consumption metrics, which obscure the translation of usage into clear business value and necessitate new internal controls like routing layers to optimize cost-efficiency.

The intricacies of tokenomics pricing are further compounded by the operational realities of agentic AI workflows, which can generate exponentially higher costs than traditional copilots due to multiple model calls, retries, and orchestration steps. Experts like Elmer Morales highlight the need to model failure paths and embed real-time architectural controls—such as token caps and retry limits—to prevent runaway spending, as retrospective FinOps dashboards fail to provide timely cost containment. Gartner estimates that autonomous AI agents consume five to 30 times more tokens per task than standard chatbot interactions, accelerating cost growth beyond the mitigating effect of declining token prices.

Budgeting under usage-based AI pricing remains exceptionally challenging due to the organic and unpredictable nature of consumption, which can spike unexpectedly as employees discover new applications across organizations. CFO Greg Henry of 1Password warns that traditional budgets are ill-equipped to absorb these fluctuations, while the lack of pricing transparency and detailed consumption data from vendors—criticized by Flexprice CTO Nikhil Mishra—further obscures forecasting efforts. This opacity risks incentivizing expensive token consumption disconnected from business outcomes, echoing past pitfalls of output-based metrics like lines of code, and has led some organizations to freeze spending or delay AI initiatives despite the technology's strategic importance.

Rising and often unexpected AI costs have forced many organizations to take immediate operational actions such as spending freezes, project delays, and tighter governance. According to Mavvrik’s 2026 report, 62% of enterprises experienced surprise AI expenses that materially influenced business decisions, with 33% implementing emergency spending freezes and 25% delaying or canceling AI initiatives. This reactive stance is echoed by companies like Atlassian and Amazon, which have restricted employee access to premium AI tools and imposed token caps to curb ballooning bills, illustrating how financial pressures directly slow AI adoption and complicate strategic planning.

A pervasive lack of cost visibility and weak financial controls exacerbate these operational challenges, undermining organizations’ ability to manage AI spending effectively. KPMG’s analysis highlights that firms with full transparency into AI operating costs realize returns on investment at five times the rate of those without such insight. Yet, as Aidana Zhakupbekova of Rydoo points out, most businesses still cannot precisely quantify their AI expenses, leading to stalled initiatives and heightened scrutiny from finance teams and boards. This opacity extends to engineering teams as well, with only 45% understanding the costs of what they build, fueling inefficiencies and wasted spend.

Finance leaders emphasize that treating AI spending with the same rigor as other strategic investments is critical to unlocking sustainable value, yet many organizations fall short. Aidana Zhakupbekova stresses the need for early financial oversight linking AI expenditures to business outcomes to avoid AI becoming a mere operational cost. The EY survey reveals that while 98% of senior leaders acknowledge escalating AI token costs have prompted strategic reassessments, only 64% actively monitor usage with clear budget guardrails. This gap often leads to spending freezes and cautious scaling, as seen in the ad industry where firms like Dollar Shave Club and PMG implement token caps and usage triage to manage costs during critical periods like Black Friday.

The financial strain from unpredictable AI expenses is driving a strategic pivot toward in-house AI development and more stringent governance to regain control and reduce vendor dependency. EY’s research shows 91% of senior leaders now view custom AI-built applications as essential, with 87% piloting or deploying such solutions, signaling a shift away from traditional enterprise software. However, the absence of clear AI cost ownership and evolving governance policies continues to hinder effective cost management, as Brogan advises establishing dedicated cost owners and iterative governance frameworks involving engineers and business partners to avoid surprises and enable better strategic planning.

Enterprises are increasingly adopting AI infrastructure strategies that emphasize financial transparency and accountability to tame unpredictable usage-based costs. By combining centralized AI platforms with localized deployments, organizations maintain operational consistency while adapting to regional resource and regulatory demands, as highlighted in the 2026 ISG report where Sonam Chawla noted that providers enabling resilient models with predictable economics are best positioned to support AI as a core business capability. This disciplined approach reduces idle capacity and inter-team friction, ensuring cost control is embedded alongside technical performance.

CIOs are shifting toward FinOps-style real-time management of AI consumption, recognizing that traditional budgeting falls short against AI's complex, consumption-driven pricing models. Michael Corrigan of World Insurance Associates stresses the need to actively govern AI usage, while Elmer Morales of koder.com advises mapping AI workflows—not just headcount—to understand cost drivers, including failure and retry behaviors that can inflate expenses. Moreover, W.W. Grainger’s Pavan Madduri underscores embedding hard cost controls like token caps and runtime limits directly into system architecture to prevent bill shock rather than relying on retrospective dashboards.

To further rein in AI spending, CFOs and leaders are defining clear outcome-based cost boundaries before vendor negotiations and educating users on the expensive nature of AI tools, as Matthias Steinberg recommends promoting judicious use of less costly models like Copilot. Companies like MindBridge are pioneering internal cost control infrastructures, such as routing layers that automatically select the most cost-effective AI models, while also leveraging open-source and smaller AI models to reduce reliance on pricey large providers without sacrificing much performance.

Despite widespread concern—82% of senior leaders express anxiety over AI token costs—only 64% actively monitor usage with budget guardrails, prompting a strategic pivot from rapid AI adoption to value-focused prioritization, according to EY’s 2026 survey. Interestingly, 37% of organizations have expanded AI rollout in response to cost scrutiny, while 29% accelerate deployment, reflecting a nuanced balance between fiscal discipline and competitive urgency. This financial pressure is also driving 91% of leaders to prioritize in-house AI software development, aiming to gain tighter cost control and reduce dependency on traditional SaaS pricing models.