AI costs are shifting toward token-based billing, but a lack of standardization and independent verification creates financial risks for global CFOs and CIOs.

For decades, the global economy has relied on a fundamental principle of commercial fairness: the measurement of a commodity must be independent of the seller. Whether it is a fuel pump calibrated by a weights and measures office, an electricity meter with a CE mark, or a commercial scale verified by a notified body, the integrity of the transaction rests on a certified standard. This ensures that the buyer is not simply taking the seller's word for how much of a resource was consumed.

However, as enterprises and public administrations integrate artificial intelligence into their core operations, they have entered a financial arrangement that defies this historical logic. The primary unit of cost for AI—the token—is currently a black box. In the current market, the token is defined by the provider, produced by the provider, counted by the provider, and billed by the provider. For the corporate buyer, there is no independent mechanism to verify the telemetry of their consumption.

The invisible meter of the AI economy

The token serves as the variable kilo of the AI era, yet it lacks a universal definition. Because different providers define and count tokens differently, comparing offers across the market has become a fragile exercise. This lack of transparency creates a structural imbalance where the provider has every incentive to ensure the count leans in their favor, while the client lacks the tools to audit the bill.

This opacity is not merely a technical nuisance; it is a financial risk. Without independent standards or verifiable telemetry, businesses are essentially signing blank checks based on proprietary metrics. As AI spending scales from experimental pilots to systemic infrastructure, the absence of a standardized measurement system for tokens makes it nearly impossible to maintain strict budgetary control or ensure the loyalty of commercial transactions.

Why the Federal Reserve is watching token prices

The implications of this pricing model have reached the highest levels of financial oversight in the United States. During a recent Jackson Hole speech, Federal Reserve Chairman Kevin Warsh highlighted token prices as a critical window into the broader economics of AI. Warsh suggested that these prices could indicate whether AI is truly acting as a new factor of production or if the industry is heading toward a bubble.

The Fed's interest lies in the signal that token pricing sends about productivity. If the cost of tokens for advanced models remains high while older models drop toward marginal cost, it suggests that customers are willing to pay a premium for superior intelligence—a sign of genuine value creation. Conversely, if prices plummet across the board because models have become interchangeable, it signals commoditization. In such a scenario, the massive capital flowing into AI might not generate the expected returns, raising concerns about the sustainability of the current investment cycle.

From token maxxing to task-based ROI

Within the corporate C-suite, the conversation is evolving from simple consumption metrics to outcome-based analysis. The era of token maxxing—the attempt to either maximize output or minimize consumption for the sake of the metric itself—is proving to be a dead end. The emerging consensus among CIOs is that cost per token and cost per task are not competing metrics, but two sides of the same coin.

The reality is that different business problems require different levels of intelligence. A simple data extraction task can be handled by a small, efficient model, while complex strategic reasoning requires a sophisticated model with a longer context window. Spending more tokens on a complex problem is often the smarter investment if it reduces the number of iterations required to reach a correct result. The goal is no longer to minimize the token, but to optimize the intelligence used for the specific task.

The real optimization is not about maximizing or minimizing tokens. It is about using the right intelligence for the right task.

The productivity paradox and the failure rate

Despite the rush to adopt these tools, the gap between expenditure and results remains wide. Data indicates a sobering trend: an MIT study revealed that 95% of AI models implemented in 2025 failed to deliver their expected results. This suggests that buying an AI license or consuming millions of tokens does not automatically translate into organizational efficiency.

For CFOs, the critical question has shifted. The focus is no longer just on the input cost—the token price—but on whether that cost is generating measurable productivity gains, improving margins, or opening new revenue streams. As noted in recent industry analysis, the danger for many organizations is becoming locked into a single provider's ecosystem before they have fully understood the cost-to-value ratio of their specific use cases.

Strategic imperatives for the modern CIO

To navigate this environment of uncertainty, technology leaders are encouraged to maintain a strategy of continuous experimentation. Because the technology is still relatively immature, the cost of testing different models and team structures is lower than it will be once the market reaches full maturity. Organizations that explore multiple alternatives now will be better positioned to switch to more accessible or transparent solutions in the future.

The transition toward an agentic world—where AI agents perform multi-step tasks autonomously—will only exacerbate the need for transparent billing. When multiple agents collaborate, token consumption can spike unpredictably. Without a verifiable way to track these interactions, the monthly AI bill can quickly transform from a manageable operating expense into a financial liability.

Global implications for US and UK enterprises

For businesses operating in the USA and UK, the current token-based economy presents a unique set of challenges. Unlike the European Union, where the AI Act introduces stringent frameworks for transparency and risk management that may eventually influence how AI services are audited, the US and UK markets remain more driven by contractual agreements between private entities.

In the US, the focus is heavily on the macroeconomic signals. As the Federal Reserve monitors token pricing, US companies should expect increased scrutiny on how AI investments are impacting their balance sheets. The lack of a regulatory body to certify token counts means that US and UK firms must rely on internal telemetry and rigorous vendor SLAs to protect themselves.

The primary risk for these enterprises is vendor lock-in. When a company builds its entire workflow around a proprietary tokenization method, switching providers becomes a costly architectural overhaul. To mitigate this, international firms should prioritize model-agnostic layers and demand greater transparency in how tokens are calculated and billed. The move toward a task-based cost model is not just a financial preference; it is a strategic necessity for any company aiming for a sustainable AI ROI.

FAQ

---

Sources: Agendadigitale ·