The Per-Token Trap: Why AI Got Cheaper and Your Bill Didn't

The number that should be pinned to every CTO's whiteboard is stark: the price of a single AI token has fallen more than 90% since 2023. Yet, in that same window, total enterprise spending on large language models has doubled, according to Apollo chief economist Torsten Slok's analysis of the Silicon Data Token Expenditure Index. Ramp's billing data tells an even more dramatic story, showing that businesses' monthly AI spend roughly quadrupled between February 2025 and February 2026.

This is not a paradox waiting to happen; it is a reality that has already arrived. If your engineering team is still tracking "cost per million tokens" as the primary budget metric, you are optimizing the wrong number.

Understanding the Jevons Paradox in AI

In 1865, economist William Stanley Jevons observed something counterintuitive about coal: as steam engines became more efficient and the effective cost of coal fell, Britain did not burn less coal. It burned dramatically more. The efficiency gains made coal-powered industry viable for use cases that were previously too expensive, and total consumption exploded.

Torsten Slok labeled what is happening in AI spending "Jevons paradox in action." The mechanism is identical. As tokens get cheaper, companies do not spend less. Instead, they run more autonomous AI agents, automate complex workflows, generate more code, and attempt tasks that were previously cost-prohibitive.

The math is straightforward:

<blockquote> Cost per task = (tokens per task) × (cost per token) </blockquote>

While the cost per token is falling, the number of tokens consumed per task is rising much faster. That is the variable nobody budgeted for.

Bain & Company's analysis found that token costs halved from December 2024 to December 2025, but tokens consumed grew 450% over the same period. Bain attributes this surge to three main forces:

1. Enterprises upgrading to newer, more capable, and token-hungry models like DeepSeek-V3 and Claude 3.5 Sonnet.
2. Autonomous agents handling complex multi-step reasoning tasks.
3. Teams expanding AI usage across departments once they trust it with larger workloads.

The bill stays stubbornly high because it shifts from paying more per unit to paying for vastly more units.

The Agentic Shift and the Inference Paradox

The shift from simple chatbots to autonomous agents is the single largest driver of this cost explosion. A traditional chatbot query is a one-shot exchange: you send a prompt, you get a response, and you pay for a few hundred tokens. An AI agent operating inside coding tools, customer support pipelines, or enterprise automation operates entirely differently.

Gartner coined the term "Inference Paradox" to describe what happens when agents take over. Their prediction indicates that inference costs per agentic workflow will increase more than fivefold, even as per-token prices continue their downward trajectory.

Why does this happen? Because agents resend their full context window at every step. Each tool call, each reasoning iteration, and each self-correction loop multiplies the token count exponentially. A simple query uses a few hundred tokens, but multi-step reasoning and agentic coding tasks consume 5 to 30 times more. For complex autonomous workflows, consumption can easily hit 1000x the baseline.

The Uber Case Study: From Tokenmaxxing to FinOps

No company illustrates the per-token trap more vividly than Uber. In early 2026, Uber rolled out advanced coding assistants to thousands of engineers and built internal leaderboards ranking employees by token consumption—a practice the industry quickly dubbed "tokenmaxxing."

The result? Uber burned through its entire AI coding budget in the first four months of the year. Uber's leadership admitted that the direct link between raw AI spending and shipped product features was missing. However, their recovery plan serves as a blueprint for the industry:

• Improved Prompt Caching: Reducing redundant context re-reads across development sessions.
• Smart Model Routing: Automatically routing routine classification or summarization tasks to cheaper, highly efficient models.
• Real-Time Visibility: Giving developers immediate visibility into their token consumption and associated costs.
• Model Efficiency Testing: Evaluating newer model architectures for token economy rather than raw capability alone.

By implementing these guardrails, Uber managed to quadruple the number of engineers using frontier AI tools while simultaneously driving down the cost per token.

How n1n.ai Solves the Multi-Model Cost Crisis

Navigating the Jevons paradox requires more than just internal policy changes; it demands robust infrastructure. This is where n1n.ai comes in as the premier LLM API aggregator. By centralizing access to hundreds of frontier and open-weights models through a single, unified interface, n1n.ai empowers engineering teams to implement dynamic model routing and cost control without changing application code.

With n1n.ai, you can instantly route non-critical requests to cost-effective models while reserving high-cost frontier models strictly for complex, multi-step agentic workflows.

Practical Steps to Escape the Per-Token Trap

To keep your enterprise AI budget under control, your organization must transition from measuring raw consumption to measuring efficiency and return on investment.

1. Stop Measuring Cost per Token: Shift your primary metric to cost per successful outcome or feature shipped.
2. Build an Intelligent Routing Layer: Do not use a frontier model for tasks that smaller, highly optimized models can handle with identical accuracy.
3. Enforce Per-Team Budgets: Give every engineering squad real-time visibility into their spending so they understand the financial impact of every agentic loop.
4. Maximize Prompt Caching: Implement aggressive caching and context compression to minimize redundant input token re-reads.

Get a free API key at n1n.ai