AI has become one of the fastest-growing categories of enterprise technology spending, yet most organizations have no clear view of where their tokens go or whether that spending delivers business value. That is the central argument of a new analysis by Ameya Kanitkar, co-founder and CTO of Larridin, a Bay Area-based startup building an AI-powered organizational platform. The piece, published Aug 12, 2026, 08:15am EDT as a Forbes Council Post, draws on proprietary data from Larridin's enterprise customers to show how quickly AI costs can spiral out of control.
The problem is not that companies are unwilling to spend. It is that AI spending changes daily, making it nearly impossible to budget with the same confidence applied to traditional technology purchases. Kanitkar argues that AI introduces a variable operating expense that changes minute by minute, and that costs often increase sharply over a few weeks or months.
Why AI Defies Traditional IT Budgeting
IT organizations spent decades developing mature practices for managing cloud infrastructure, software licenses, and hardware investments. Software costs can be calculated per user and budgeted annually. Cloud computing grew more slowly than AI, giving organizations years to create governance and FinOps practices. AI adoption is happening much faster than cloud computing did, creating a fast-growing and poorly understood expense.
Traditional technology budgets rely on predictable patterns. AI usage varies dramatically across employees, teams, and workflows. A budget built at the beginning of the year can become outdated in a few months or even weeks. That unpredictability makes AI fundamentally different from the line items finance teams have managed for years.
Token spending is also scattered. It is spread across cloud model providers, AI platform subscriptions, browser plugins, desktop agents, custom connectors, and API gateways. Each of those sources has its own billing model, reporting system, and usage dashboard. Finance teams receive invoices from different vendors, while IT monitors the tools they manage. The result is a fragmented picture that no single department fully controls.
The Hidden Cost of Shadow AI
The analysis also highlights the role of unsanctioned "shadow" AI. Employees adopt AI tools without formal approval, and those costs land on invoices that finance teams may not recognize. Leaders see productivity upticks but often lack understanding of the reasons or costs behind them.
A common assumption is that AI spending will mainly come from a handful of enterprise subscriptions, but in reality, token consumption is scattered. That scattering makes it easy for costs to grow unnoticed. Some organizations have slowed AI adoption because costs appear to be rising too quickly. Others increase budgets without knowing if additional token spend produces meaningful outcomes.
The data from Larridin's enterprise customers illustrates the scale of the problem. Over a 10-week period, Claude's active users increased by 41% and sessions increased by 76%. That kind of growth, while promising for adoption, makes cost forecasting extremely difficult.
Model Choice Drives Cost Discrepancies
The choice of AI model has a direct and measurable impact on spending. According to proprietary data from Larridin, Anthropic's Claude LLM Sonnet model receives 70% of API prompts across Larridin's enterprise customers but accounts for only 51% of API spend. Claude's Opus model handles 27% of prompts but consumes 48% of the spend.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
The reason is simple: each prompt for Claude's Opus model costs 3-4 times more than for Sonnet. The average cost per prompt for Claude Sonnet 4.5 is $0.06, while the average cost for Claude Opus 4.1 is $0.22. That difference compounds quickly at enterprise scale.
Kanitkar notes that the higher cost of models like Opus doesn't necessarily represent waste. Teams may intentionally reserve them for complex work. But without visibility into which model is being used for which task, finance teams cannot distinguish between intentional premium usage and runaway spending.
Cost Alone Tells Very Little About Value
Users don't see costs directly when choosing a model or entering a prompt. That lack of visibility creates a disconnect between spending and value. Cost alone tells very little about value. A team might spend heavily on premium models for routine tasks, while another team delivers strong results using cheaper models.
The analysis proposes a new playbook for managing AI spend. Managing token spend starts with a complete inventory of where AI costs originate. Organizations should connect token spend to the work being performed. Companies should establish ROI guardrails to identify the point where additional AI investment stops improving productivity, software quality, or business outcomes.
The ROI threshold for AI investment will be different for every organization. There is no universal formula. But the principle is the same: every token should be traceable to a business outcome.
The Stakes Are Rising Fast
The urgency of this issue is underscored by a prediction from Gartner. Gartner researchers predict that by 2028, the cost of AI coding tokens alone will surpass the salary of an average software developer. That single metric shows why AI spending can no longer be treated as an afterthought.
Token spending is becoming a permanent line item in enterprise budgets. Managing AI spending successfully means understanding where every token is going, why it was used, and what value it created. That requires new financial management practices, not just new tools.
The article is part of the Forbes Technology Council, an invitation-only community for world-class CIOs, CTOs, and technology executives. The opinions expressed are those of the author, Ameya Kanitkar. The article includes a note that the voice experience is generated by AI.
For organizations still treating AI as a small experimental budget, the data suggests a different reality. AI is one of the fastest-growing categories of enterprise technology spending, and it is growing faster than the governance structures meant to manage it. The question is not whether AI spending will become a major line item. It already has. The question is whether finance and IT teams can catch up before the costs outpace the value.