Anthropic shifts to usage-based billing for firms amid compute crunch

TL;DR

Anthropic shifted Claude Enterprise pricing to charge businesses based on actual AI token usage rather than fixed subscriptions alone.
Customers now pay a $20 per user per month seat fee for access plus separate charges at standard API rates for all Claude, Claude Code, and Cowork usage.[[1]](https://www.theinformation.com/)[[2]](https://www.facebook.com/Techmeme/posts/anthropic-recently-changed-claude-enterprise-pricing-customers-now-pay-20month-p/1393186026177041)
Heavy users face much higher bills amid Anthropic's compute shortages from booming demand for coding and agent tools.[[3]](https://www.theinformation.com/articles/anthropic-changes-pricing-bill-firms-based-ai-use-amid-compute-crunch)

The story at a glance

Anthropic recently changed its Claude Enterprise pricing model in recent weeks to bill business customers based on the amount of AI they use, on top of a seat fee. The company made the shift, which has not been widely reported before, as demand surges for its Claude products. This comes as Anthropic grapples with compute constraints from popular coding and agent features. Businesses with heavy Claude usage are likely to see significantly higher costs.[[1]](https://www.theinformation.com/)

Key points

Previous enterprise plans included more bundled usage in seat fees; now the seat fee covers only platform access to web, desktop, mobile, Claude Code, and Cowork, with all tokens billed separately.[[4]](https://support.claude.com/en/articles/11526368-how-am-i-billed-for-my-enterprise-plan)[[5]](https://www.implicator.ai/anthropic-shifts-enterprise-billing-to-per-token-pricing-the-flat-fee-era-is-over)
Seat fee is $20 per user per month, confirmed by Anthropic spokesperson; usage charged at standard API token rates based on actual consumption.[[2]](https://www.facebook.com/Techmeme/posts/anthropic-recently-changed-claude-enterprise-pricing-customers-now-pay-20month-p/1393186026177041)
Change affects largest customers transitioning from legacy seat types at contract renewal; no new legacy plans available.[[4]](https://support.claude.com/en/articles/11526368-how-am-i-billed-for-my-enterprise-plan)
Driven by surging popularity of Claude's coding and agent products, which drive up compute costs for Anthropic.[[1]](https://www.theinformation.com/)
Team plans (up to 150 users) still offer more bundled usage per seat at higher fixed rates like $25-$150 per user; Enterprise for larger scale with metered billing.[[6]](https://www.reddit.com/r/ClaudeAI/comments/1sh2scb/claude_enterprise_pricing_am_i_missing_something)

Details and context

The shift reflects Anthropic's broader efforts to manage rapid growth, with annualized revenue reportedly topping $30 billion from over 1,000 business customers spending $1 million each yearly. Compute crunch stems from high-demand features like Claude Code and agents, which consume far more resources than standard chat—similar to recent moves blocking flat-rate subscriptions for third-party tools like OpenClaw.[[7]](https://www.anthropic.com/news/google-broadcom-partnership-compute)

Legacy enterprise contracts keep old terms until renewal, then move to the new model; light users may pay less overall, but heavy ones face sharp increases—for example, one estimate put mid-range consumption at hundreds of thousands yearly for large teams.[[6]](https://www.reddit.com/r/ClaudeAI/comments/1sh2scb/claude_enterprise_pricing_am_i_missing_something)

This pricing aligns enterprise more closely with API pay-as-you-go, giving admins tools like spend caps but exposing costs directly to usage patterns.

Key quotes

"The seat fee only covers access to the platform and doesn't include any usage. All usage across Claude, Claude Code, and Cowork is billed separately at standard API rates, based on what your team actually consumes." — Anthropic Enterprise help documentation.[[5]](https://www.implicator.ai/anthropic-shifts-enterprise-billing-to-per-token-pricing-the-flat-fee-era-is-over)

Why it matters

Anthropic's move highlights how AI providers are passing on soaring compute costs to enterprise users as demand outstrips supply. Heavy Claude users in coding and agents will see bills rise substantially, while light users might save, forcing businesses to track usage closely. Watch how competitors like OpenAI adjust pricing and if Anthropic adds more capacity through deals like its Google/Broadcom expansion.[[7]](https://www.anthropic.com/news/google-broadcom-partnership-compute)