The End of the AI Buffet

Anthropic's Platform Play — Why models are just the appetizer

By Nick Zervoudis Apr 23, 2026 Value from Data & AI

The Big Picture

TL;DR

It's just capitalism: First you subsidize, then you capture market share, then you extract. Anthropic is generating more revenue than OpenAI with roughly 5% of OpenAI's user base.

Revenue Comparison

Anthropic: $30B OpenAI: $24B

Anthropic generates more revenue with a fraction of users by targeting enterprise deals. 85% of Anthropic's revenue comes from API and enterprise, vs 73% consumer for OpenAI.

The Strategy

Models are commoditized. The real business is the platform & application layers. Anthropic is building Claude Code, Design, Co-Work, Marketplace, Managed Agents — all moves toward becoming where work happens.

Industry Analysis

Anthropic

Enterprise 85% Consumer 15%

AI safety company developing Claude. $30B revenue, targeting enterprise deals. Over 1,000 enterprise customers spend >$1M annually.

OpenAI

Consumer 73% Enterprise 27%

900M weekly active users, 90-95% free tier. $24B revenue, burns $17B annually. Introduced ads into ChatGPT.

The Pattern

Subsidize to capture, then extract — same playbook as Amazon, Uber, Facebook. The era of infinite subsidized compute is over.

Anthropic Product Suite

Claude Code

Low Lock-in

AI coding tool. Portable — your codebase lives in your repo, CLAUDE.md files are markdown. Developers love it.

Claude Design

High Lock-in

AI tool for visual work - prototypes, slides, interfaces. Sent Figma's stock down 5% on launch day.

Claude Co-Work

High Lock-in

Business collaboration tool. Deep context lock-in with business workflows.

Claude Managed Agents

High Lock-in

Cloud-hosted agent infrastructure. Launched 4 days after OpenClaw ban. Sandboxed execution, state management.

Marketplace

Medium Lock-in

Zero-commission model. Partners: GitLab, Snowflake, Harvey, Replit, Rogo, Lovable. Distribution channel control.

Claude Routines

High Lock-in

Saved configurations running on Anthropic infrastructure. Workflow lock-in — replaces your cron jobs and GitHub Actions.

Lock-in Analysis

Contract Lock-in

Multi-year agreements with spend commitments. Over 1,000 enterprises spend >$1M annually through cloud-channel agreements.

Data/Context Lock-in

Where your organization's history and integrations live. Claude for Enterprise: shared projects, conversation history, connectors, SSO, audit logging.

Model Lock-in

How easy to swap the underlying model. Foundation labs built abstraction layers making switching structurally easy.

Frequently Asked Questions

Because autonomous agents were consuming $1,000-$5,000/day of compute through subsidized subscriptions, bypassing prompt caching that reduces costs by 92%. It was about closing an exploit significantly more expensive to serve than same usage through Anthropic's own tools. [Answer]
Anthropic generates $30B with ~5% of OpenAI's user base by targeting enterprise deals. 85% of revenue comes from API and enterprise deployments, just 15% from consumer subscriptions. [Answer]
Models converge fast. Claude 3.5 Sonnet and Gemini matched GPT-4 within months. Open-source models like Llama 3 are good enough for many use cases. Foundation labs facilitate portability by building routing layers. [Answer]
Anthropic is building a single conversational interface where work itself happens - Claude Code for coding, Claude Design for visual work, Co-Work for business collaboration, creating an ecosystem hard to leave. [Answer]
First subsidize to attract users and capture market share, then tighten terms once users are hooked to extract profit. Same playbook as Amazon, Uber, Facebook - predates Silicon Valley by a century. [Answer]
Almost none - codebase lives in your repo, CLAUDE.md files are portable markdown. But it creates developer affection which creates procurement stickiness. [Answer]
Deep lock-in - shared projects, conversation history, company connectors, SSO, role-based access, audit logging all live inside Anthropic's application. Migration is ERP-scale. [Answer]
Contract lock-in (spend commitment duration), data/context lock-in (where org history lives), model lock-in (ease of swapping underlying AI model). [Answer]
Start by paying for Opus, prove value, move fast. Then evaluate migrating high-volume workloads to fine-tuned open models on own infrastructure. Prove value first, then optimize unit economics. [Answer]
Timing tells the story - Anthropic isn't just protecting margins, it's building a platform. The restriction was partly strategic, but safety concerns about unconstrained agents were also credible. [Answer]
More like Microsoft - bundling applications for enterprise buyers with consolidated billing. Google's experimental approach kills beloved but insufficiently popular products. Market will determine which wins. [Answer]
The era of infinite subsidized compute is over. Most dangerous: high data/context lock-in with unpredictable pricing. Negotiate contracts carefully, pay attention to where institutional knowledge lives. Plan accordingly. [Answer]

Glossary

AI system that autonomously performs tasks end-to-end without human intervention
AI tool that assists human operators but requires human direction and approval
Strategy of moving up the stack from infrastructure to platform layer to capture more value and create lock-in
Degree to which switching vendors is difficult due to data, contracts, or integration complexity
Business pattern of offering low prices to attract users, then raising prices once market share is gained
Percentage of revenue coming from enterprise vs consumer customers
When products become interchangeable with competitors, reducing differentiation and margins
AI infrastructure under national control to avoid US CLOUD Act jurisdiction
Technique reducing costs by 92% by caching repeated context instead of reprocessing
Intercom's domain-specific model saving $250K/month while outperforming frontier models

How to Evaluate AI Platform Strategy

1

Assess the lock-in types

For any AI tool, categorize lock-in as contract (commitment length), data/context (where org knowledge lives), or model (swap ease)

2

Calculate total cost of ownership

Include subscription, API usage, migration costs if switching. Consider the subsidy period ending.

3

Evaluate vendor captivity risk

Ask: what happens if provider deprecates model, raises prices, gets acquired, or is compromised?

4

Map data sovereignty requirements

For regulated industries, especially in Europe, consider CLOUD Act implications and sovereign AI options

5

Follow the Intercom path if building products

Pay for frontier models first to prove value, then optimize to fine-tuned open models later

6

Negotiate enterprise contracts carefully

Lock in pricing and terms now before subsidy era ends. Watch for multi-year commitments.

7

Watch for the extraction phase

When platforms become indispensable, expect terms to tighten. Plan for the inevitable.