Observed Signal · Apr 17, 2026 · Analysis · Source: CNBC Technology · Impact: 3/5 · Sentiment: Neutral
Anthropic Bets on Per‑Token Billing Amid Token Boom
A CNBC perspective argues current signals of explosive AI demand—measured by token consumption—may be overstated because volume-based metrics incentivize wasteful usage. Anthropic has shifted from flat-rate enterprise and consumer subscriptions to per-token billing for its latest models (charging $5 per million input tokens and $25 per million output tokens), and has cut off some third‑party tools that heavily consumed tokens. The piece contrasts Anthropic’s pricing discipline with broader industry practices (including unlimited or flat-rate plans), cites enterprise difficulty in proving ROI, and notes companies measuring adoption by token volume can encourage inefficient behavior. Ramp reports a 13x increase in AI spending among its customers; Salesforce is testing an
Impacts AI infrastructure economics, enterprise budgeting and revenue quality; shifts from flat-rate to per-token pricing affect vendor unit economics and will influence investor scrutiny as Anthropic and OpenAI pursue IPOs.
Track Anthropic Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Tokens are the basic usage unit for AI; agentic AI sessions can consume thousands to millions of tokens.
- Anthropic’s published rates for its latest model: $5 per one million input (prompt) tokens and $25 per one million output (response) tokens.
- Anthropic moved away from flat-rate consumer and enterprise pricing to per-token billing and on April 4 cut off some third-party tools that heavily consumed tokens; legacy flat-seat enterprise contracts are no longer available for new contracts.
- Industry executives (Databricks’ CEO Ali Ghodsi, Nvidia’s Jensen Huang) and enterprise leaders warn that measuring adoption by token volume can drive inefficient behaviour and obscure ROI.
- Ramp reports AI spending across its customers grew 13x over the past year; Salesforce is rolling out a metric called 'agentic work units' to measure work completed rather than tokens burned.
Connected Companies & Entities
6 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Enterprise AI Enters Consumption Era; Outcomes Over Tokens
Issue #500 of the What’s 🔥 in Enterprise IT/VC newsletter (published 2026-05-30) argues that enterprise AI is moving from a subsidy-driven 'tokenmaxxing' phase into a consumption-driven Phase 2 where vendors charge for inference and CFOs demand ROI. The author warns that measuring token consumption (or running token leaderboards) incentivizes wasted effort and highlights corporate examples: an Uber COO questioning AI spend, Amazon shutting an internal AI-use leaderboard, and Anthropic rejecting leaderboard ideas. The piece recommends intelligent model routing, use of open-weight models for non-frontier workloads, hybrid and on-prem deployments for sensitive data, and outcomes-based pricing. The newsletter cites supporting data and market signals (Goldman Sachs token-growth projection, Sierra’s outcomes pricing, Harvey’s work with Trajectory/NVIDIA) and frames this as an operational and commercial shift for enterprise AI buyers and vendors.
Token Becomes the Unit of Account in AI
This newsletter summarizes a week of AI industry developments emphasizing a shift toward token-based economics. Anthropic released Claude Opus 4.8 and disclosed rapid revenue growth while raising a $65B Series H at a ~$965B post-money valuation. Anthropic’s Opus 4.8 introduces agentic and governance features (effort control, dynamic workflows, improved self-checking) aimed at long-running agentic tasks. Infrastructure and routing firms are monetizing by token throughput: OpenRouter raised $113M at a $1.3B valuation as weekly token throughput rose to 25 trillion, and Cognition raised $1B (≈$26B valuation) reporting a $492M run-rate and heavy reliance on its coding agent Devin. Snowflake committed $6B to AWS and acquired Natoma to support agentic data access. The piece links these commercial trends to ethical concerns raised by Pope Leo XIV’s encyclical about technology disintermediating human judgment.
Industry Scrambles to Manage AI Token Costs
Enterprises are confronting rapidly rising AI inference costs as token consumption surges from agentic features and broad developer adoption. TechCrunch reports large organizations (including Uber, Microsoft and Priceline) exceeded or cut AI spending after unexpected bills and license pullbacks. In response, the Linux Foundation this week announced plans for the Tokenomics Foundation, a standards body to create canonical definitions, metrics and specs for AI token usage and billing; a formal launch is planned in July. Startups and established vendors (Pay-i, Paid, Jellyfish, Faros AI, Ramp, Datadog, New Relic and others) are building tooling for token-level observability, budgeting and optimization. Analysts and vendors warn companies must overhaul tooling and accounting to track trillions-of-rows token telemetry; Goldman Sachs projects global token usage could multiply ~24x by 2030.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.