Perspective: AI Demand Is Inflated, and only Anthropic Is Being Realistic

Why It Matters

Usage‑based pricing aligns revenue with real customer value, reducing the risk of a market correction and offering investors clearer insight into sustainable AI growth.

Key Takeaways

• Tokens measure compute spend, not business outcomes.
• Anthropic shifted from flat‑rate to per‑token pricing.
• Per‑token billing aligns revenue with actual usage.
• Flat‑rate plans can mask inflated AI consumption.
• Industry peers, like OpenAI, face pressure to adopt usage‑based models.

Pulse Analysis

Token counts have become the headline metric for AI adoption, but they tell only part of the story. Companies that track how many tokens engineers consume are effectively measuring spend, not output, which can drive wasteful behavior and obscure true ROI. As enterprises grapple with budgeting for AI, the focus is shifting from raw token volume to the tangible business results those models enable.

Anthropic’s decision to replace flat‑rate enterprise subscriptions with per‑token billing marks a strategic pivot toward transparency. By charging $5 per million input tokens and $25 per million output tokens, the firm captures revenue that mirrors actual workload, discouraging the “burn‑and‑earn” loops some firms have adopted. This model also provides granular data for product development and pricing optimization, positioning Anthropic as a potential benchmark for investors wary of inflated demand forecasts.

The broader market is feeling the pressure. OpenAI’s unlimited plans, while popular, risk over‑promising on usage without delivering measurable value, prompting rivals and customers alike to reconsider pricing structures. New metrics such as Salesforce’s “agentic work units” aim to quantify output rather than input. As AI spending surges—evidenced by a 13‑fold increase among Ramp’s clients—companies that align billing with real work are likely to attract capital and retain customers when the hype cycle settles.