Key Takeaways
- • AI token consumption is accelerating at 75% annually, with enterprise token volumes already exceeding 10 billion per month at scale, and projections showing a threefold increase in companies generating above 100 billion tokens monthly between 2025 and 2028 (Deloitte survey of 550 US enterprise leaders).
- • AI costs have surged from 31% of IT budgets in 2024 to 98% in 2026, making token economics a critical CFO accountability requiring dedicated budgets, dedicated compute allocation, and deliberate model portfolio management separate from legacy operational expenses.
- • Unit economics shift meaningfully across deployment models: cloud APIs offer flexibility at lower volumes; alternative "neocloud" solutions improve mid-scale economics; and self-hosted AI factories or on-premises models deliver lowest unit costs at sustained high volumes but require significant upfront capital investment.
- • Unmanaged token dynamics create four immediate CFO risks, including budget overruns, architectural misalignment, forecasting failures, and P&L governance gaps—necessitating CFOs establish economic thresholds, escalation points, and cost controls even when they do not select technical architectures.
- • Token economics predictability is achievable: economic inflection points become forecastable when token demand is understood early, enabling CFOs to optimize between cloud flexibility, hybrid neocloud approaches, and capital-intensive on-premises deployment models based on projected consumption trajectories.
Summary
Enterprise artificial intelligence spending has become a material balance sheet issue requiring sophisticated cost governance. AI token consumption—the fundamental billing mechanism for large language models and AI agents—is growing at approximately 75% annually, driven largely by AI agents that consume 30 to 60 times more tokens per task than earlier chatbot applications. According to a Deloitte AI Infrastructure 2028 survey of 550 US enterprise leaders, many companies today already generate above 10 billion tokens per month, with the proportion expecting to exceed 100 billion tokens monthly projected to triple between 2025 and 2028. This acceleration has compelled CFOs to treat token economics not as an operational footnote but as a core governance issue comparable to traditional capital expenditures.
The financial magnitude of this shift is stark. AI costs as a percentage of enterprise IT budgets jumped from 31% in 2024 to 98% in 2026, making AI spending management central to CFO accountability. The implication is that token costs can no longer be buried in legacy budget lines or treated as incidental cloud expenses. Instead, Deloitte's analysis and publisher guidance recommend establishing dedicated AI compute budgets, treating tokens as measurable units of value, and applying disciplined portfolio management to model deployment decisions. Once organizations account for tokens as discrete cost drivers, they gain visibility into spending patterns and can enforce meaningful cost controls.
Unit economics vary dramatically across deployment architectures, and the optimal choice depends on anticipated token volume and consumption trajectory. At lower volumes, cloud APIs provide superior flexibility despite higher per-token unit costs. As organizations scale to mid-range consumption levels, alternative cloud or "neocloud" hybrid solutions can improve unit economics relative to pure cloud approaches. At sustained high volumes, on-premises AI factories or self-hosted models deliver the lowest unit cost but require substantial upfront capital investment and operational complexity. The strategic implication for CFOs is that no single deployment model is universally "right"—rather, economic inflection points become predictable and actionable when token demand is understood early, enabling proactive transitions between cloud, hybrid, and on-premises architectures.
Unmanaged token dynamics present four immediate risks to CFOs: budget control failures, architectural-economic misalignment, forecasting inaccuracy, and P&L governance gaps. CFOs do not always architect technical infrastructure, but they routinely set economic constraints, escalation thresholds, and accountability frameworks. As token metering capabilities become standard across cloud providers and SaaS solutions, CFOs gain granular visibility into consumption. This visibility is both an opportunity and an obligation—the ability to track tokens enables disciplined cost management, but failure to establish governance frameworks can result in unchecked spending acceleration and margin compression. The practical response is to establish clear economic thresholds triggering escalation, implement portfolio-level model cost analysis, and align token economics decisions with broader enterprise value creation objectives.
The financial magnitude of this shift is stark. AI costs as a percentage of enterprise IT budgets jumped from 31% in 2024 to 98% in 2026, making AI spending management central to CFO accountability. The implication is that token costs can no longer be buried in legacy budget lines or treated as incidental cloud expenses. Instead, Deloitte's analysis and publisher guidance recommend establishing dedicated AI compute budgets, treating tokens as measurable units of value, and applying disciplined portfolio management to model deployment decisions. Once organizations account for tokens as discrete cost drivers, they gain visibility into spending patterns and can enforce meaningful cost controls.
Unit economics vary dramatically across deployment architectures, and the optimal choice depends on anticipated token volume and consumption trajectory. At lower volumes, cloud APIs provide superior flexibility despite higher per-token unit costs. As organizations scale to mid-range consumption levels, alternative cloud or "neocloud" hybrid solutions can improve unit economics relative to pure cloud approaches. At sustained high volumes, on-premises AI factories or self-hosted models deliver the lowest unit cost but require substantial upfront capital investment and operational complexity. The strategic implication for CFOs is that no single deployment model is universally "right"—rather, economic inflection points become predictable and actionable when token demand is understood early, enabling proactive transitions between cloud, hybrid, and on-premises architectures.
Unmanaged token dynamics present four immediate risks to CFOs: budget control failures, architectural-economic misalignment, forecasting inaccuracy, and P&L governance gaps. CFOs do not always architect technical infrastructure, but they routinely set economic constraints, escalation thresholds, and accountability frameworks. As token metering capabilities become standard across cloud providers and SaaS solutions, CFOs gain granular visibility into consumption. This visibility is both an opportunity and an obligation—the ability to track tokens enables disciplined cost management, but failure to establish governance frameworks can result in unchecked spending acceleration and margin compression. The practical response is to establish clear economic thresholds triggering escalation, implement portfolio-level model cost analysis, and align token economics decisions with broader enterprise value creation objectives.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.