How Enterprises Can Stop AI From Bleeding Their Tech Budgets

For years, enterprises treated artificial intelligence (AI) as an experiment. Pilot projects were greenlit with generous budgets, measured more by enthusiasm than by outcomes. The reasoning was that a failed use case would limit costs, while success would lead to scale later.

As enterprises accelerate AI deployment, the spotlight is turning to managing costs more effectively. Companies across various industries are facing AI expenses that outpace their budgets, leading to increased emphasis on governance, spending discipline, and solid business cases.

According to analysts, consulting firms, and industry practitioners AIM spoke to, token consumption, model access fees, and AI tool subscriptions are emerging as significant cost centres for enterprises, underscoring the need to apply the same financial rigour to other major technology investments.

This includes building explicit business cases, selecting the right tool for each task, and putting governance guardrails in place before costs spiral out of control.

"There should be more discipline in terms of what exactly you are using AI for and how much you are paying for it," says Siddhartha Tipnis, Partner and Technology Sector Leader at Deloitte India. 

He explained that when making a personal decision, people consider the benefits of their expenditure. The same logic applies in a corporate context, known as the business case. If a use case promises business benefits—such as increased revenue, reduced costs, faster time to market, or improved customer experience—then the decision to deploy is clear.

Choosing the Right Tool

The most consistent recommendation from experts is that not every problem requires generative AI, and deploying frontier models where simpler tools would suffice is one of the fastest routes to an unmanageable bill.

"Not everything needs generative AI. Many use cases could be automated. One needs to be very clear about where intelligence and generative AI are needed, and where basic engineering, data engineering, piping, or automation are needed," adds Tipnis.

Routing tasks that do not require frontier model capabilities, such as routine summarisation, standard queries, and structured data extraction, to smaller, cheaper models can significantly reduce per-query costs.

Gaurav Vasu, founder and CEO of research firm UnearthInsight, frames the challenge in terms of organisational maturity and argues that most enterprises are still a significant distance from where they need to be.

"As enterprises move from experimentation to scaled deployment, AI cost management is becoming an increasingly important topic of discussion," He says. 

Vasu adds that most enterprises operate within fixed technology budget frameworks and cannot continuously expand technology spending without impacting profitability and shareholder expectations. As AI adoption grows, organisations are increasingly recognising the need for stronger governance around usage, consumption, and business outcomes.

Experts believe enterprises need to build frameworks that connect AI spending directly to business outcomes, rather than tracking activity metrics such as adoption rates or volumes of AI-generated output.

"We have to get a new key performance indicator (KPI) for this. The new KPI that I call out is the cost per outcome. Most organisations right now are tracking token consumption, API costs, and cloud spend. These are all technical metrics," says Ashish Chandra, Partner and Global Head of AI at KPMG.

He adds that enterprises should instead evaluate AI based on business outcomes such as cost per loan approval, claims processed, audits completed, or customer issues resolved. According to Chandra, AI spending should be judged by the value it generates rather than the underlying technology costs.

Why the Rising Urgency?

These prescriptions carry urgency because the cost problem is already materialising, and visibly.

According to the State of FinOps 2026 report by the FinOps Foundation, AI cost management is the primary skill set that teams need to develop, and 98% of enterprises surveyed are now managing their AI spend, up from 31% two years ago. The report also noted that many organisations report being asked to self-fund AI investments through optimisation savings, tying traditional FinOps work directly to strategic technology enablement.

The case that has drawn the most attention recently is Uber. The company's chief technology officer disclosed earlier this year that Uber had exhausted its entire 2026 AI budget by April, following an enthusiastic enterprise-wide rollout of AI coding tools. 

Vijay Gopalakrishnan, a Partner at Deloitte India, says the challenge is real but addressable. AI solutions, including generative and agentic AI, are already moving successfully into production at scale, he notes, and the companies doing it well have followed a clear sequence.

"The first key step is to select the right use cases, supported by a strong functional understanding of the requirements that AI must meet," Gopalakrishnan says. Then enterprises need to choose an appropriate technology stack and design and deploy the solution efficiently, for example, by using the right caching approaches to keep API calls to external systems optimal, minimising unnecessary token usage, and controlling inference costs.

But technical optimisation, Gopalakrishnan argues, is only part of the answer. "Establishing strong AIOps, LLMOps, and AgentOps capabilities to monitor and manage both performance and cost helps ensure deployed solutions remain sustainable over time." 

This means real-time financial dashboards, departmental budgets for AI spend, model access controls that route teams to cost-appropriate models, and automated alerts when consumption approaches budget thresholds.