As companies race to integrate artificial intelligence into their operations, many are discovering that adoption without discipline comes with hidden costs. A key economic concept driving this is 'tokenomics'—the billing model for large language models (LLMs) based on token consumption. According to Nishanth Gopalakrishnan's article in The Hindu Business Line, understanding tokens is essential for enterprises managing AI spending.
What Are Tokens?
A token is the smallest unit of language that a large language model (LLM) uses. According to the article, a token can be a set of characters, part of a word, a whole word, or even a short multi-word phrase. The code that converts language to tokens is called a tokeniser. A good rule of thumb is that an English word is represented by approximately 1.5 tokens—so a 100-word prompt would involve about 150 tokens. However, this rule does not apply to image-, audio-, or video-based inputs. After input is tokenised, the model converts tokens into vectors (long lists of numbers) for computation.
Tokenomics: How LLM Billing Works
When it comes to LLMs, model providers such as OpenAI and Anthropic bill customers by tokens, directly relating to the amount of processing their GPUs perform. This billing is more relevant for enterprise customers using APIs than for individual consumers. Just as electricity bills vary by consumer type (industrial vs. residential), LLM pricing depends on model complexity—a larger, more robust model costs more per token.
| Billing Aspect | Input Tokens | Output Tokens |
|---|---|---|
| Cost per million tokens | Lower | Higher |
| Processing required | Reading input | Generating output |
Typically, users are billed per million tokens, with input tokens costing less than output tokens because generating output requires more processing than reading input.
The Rise of Tokenmaxxing
As AI adoption accelerates, some companies have pushed their workforce to integrate AI into daily workflows—often without capping token consumption. According to the article, employees were tracked and ranked based on token usage as a proxy for productivity and evaluated to see if they had become 'AI-native'. This led to a phenomenon called 'tokenmaxxing', where employees used larger or reasoning models even when the problem didn't warrant them, simply to boost token consumption. Importantly, reasoning steps also consume tokens even if they don't appear in the output.
Over the past year or so, companies have pushed their workforce to adopt AI in their workflows, often without capping token consumption. Employees were even tracked and ranked based on token usage, as a proxy to measure productivity and to evaluate whether they have become 'AI-native'.
This behavior mirrors the incentive misalignment seen in early cloud computing adoption, where uncontrolled usage led to cost overruns. For CTOs and technology procurement leaders, tokenmaxxing underscores the need for governance policies that align AI usage with actual business needs, rather than raw consumption metrics.
Implications for Enterprise Technology Leaders
For decision-makers evaluating AI investments, tokenomics provides a clear cost framework: each query carries a variable price based on model size and output length. Without consumption caps or usage policies, enterprises risk budget inflation from unnecessary token consumption—especially when employees default to powerful reasoning models for simple tasks. The article suggests that disciplined adoption requires understanding token economics and implementing monitoring tools to track and optimise token usage across the organisation.
Sources:
