Anthropic
Anthropic, PBC is a Delaware public benefit corporation that develops and maintains advanced artificial intelligence models and provides access to these models through the Claude platform and applications. The company sells AI “intelligence” primarily via consumption-based usage (metered in tokens) and also offers subscription tiers, serving enterprises, AI-native builders, governments, research institutions, and individual prosumers. Anthropic is headquartered in San Francisco, California and distributes Claude both directly and through major cloud service provider marketplaces.
Fiscal year 2025 · change from previous year
Development 2023–2025
Business
Share of revenueUsage-based access to Claude models via APIs and apps, metered in tokens.
Fixed-fee plans for Claude applications with bundled usage and capabilities.
Customers
Anthropic sells to enterprises embedding Claude into production workflows (including regulated industries), as well as AI-native companies building products on Claude. Customers can buy directly from Anthropic or via CSP marketplaces that handle billing and collections, and systems integrators and software/data platforms broaden deployment. A smaller portion of revenue comes from government and research institutions and from individual prosumers subscribing to Claude applications.
Consume Claude via direct contracts and/or CSP marketplaces to automate and augment core workflows (e.g., underwriting, clinical reporting, cybersecurity, investment analysis).
Build and monetize software products (e.g., coding tools, productivity apps) using Claude Platform APIs and model capabilities.
Subscribe to Claude applications for higher usage limits and advanced features, with optional incremental consumption.
Procure Claude for specific missions and programs, often with additional compliance, procurement, and contractual requirements.
Apply Claude to research and societal-impact use cases (e.g., materials science, energy security, public benefits delivery).
Strategy
Anthropic’s strategy emphasizes embedding Claude into enterprise-critical workflows and enabling third parties to build products on the Claude Platform, with distribution partnerships placing Claude inside existing customer technology stacks. The company prioritizes broad availability through major cloud platforms while maintaining direct customer relationships, and it invests heavily in securing and scaling compute capacity to train and serve frontier models. As a public benefit corporation, it also allocates resources to safety and public benefit initiatives intended to support responsible deployment.
Mission-critical deployments can expand usage over time under token-based pricing.
CSP marketplaces extend reach and simplify procurement while Anthropic retains direct customer contracts.
Frontier model training and inference require large-scale compute that must be secured in advance.
Customer adoption in regulated and high-trust settings depends on perceptions of safety and reliability.
Geography
Anthropic is headquartered in San Francisco, California and sells globally, with revenue reported by customer billing address as United States and International. The company distributes Claude through direct sales as well as through major cloud platforms (AWS, GCP, and Azure), which can shape customer access and purchasing channels across regions. International operations introduce multi-currency exposure, though the company reports the majority of revenue denominated in U.S. dollars.
Risks
Anthropic’s business model depends on sustained access to large-scale compute and on distributing models through a small number of cloud platforms, creating concentration risk in suppliers and channels. Demand is usage-driven (token consumption), so revenue can vary with customer workloads and large-customer spend patterns, and a limited number of customers can represent a meaningful share of revenue in a given period. As a safety-positioned AI lab, the company faces elevated reputational, legal, and regulatory risk from harmful outputs or perceived inconsistencies between stated safety commitments and practices.
- Compute availability, cost, and geopolitical constraints on chipsImpact: CriticalLikelihood: High
Frontier model development requires scarce compute and chips subject to supply constraints and export controls; conflicts of interest can arise with vendors that also compete in AI models.
Exposure: Hybrid sourcing model includes cloud agreements and direct chip leasing/hosting.
- Customer concentration and variability in large-customer spendImpact: HighLikelihood: High
Top customers may not be on long-term contracts and can reduce usage quickly under consumption pricing.
Exposure: Top five customers ~25% of revenue for nine months ended Sep 30, 2026; no single customer >10%.
- Competition from AI labs and platform providersImpact: HighLikelihood: High
As model capability commoditizes, differentiation may shift toward compute scale and distribution, where CSPs are both partners and competitors.
Exposure: CSP partners also offer competing AI models and services.
- Dependence on CSP marketplaces as distribution and billing channelsImpact: HighLikelihood: Medium
A significant share of revenue and receivables flows through CSP marketplaces; changes or termination could disrupt customer access and increase migration costs.
Exposure - Harmful use or harmful outputs and AI safety governance trade-offsImpact: HighLikelihood: Medium
The company may not detect or prevent all harmful outputs; safety-focused release decisions can slow launches and increase scrutiny relative to competitors.
Exposure: Use of external reviewers and adversarial testers can also raise confidentiality risks.
- Public-sector procurement riskImpact: MediumLikelihood: Medium
Government sales involve fiscal/procurement policy changes and specialized contracting requirements that can delay or constrain revenue.
Exposure: Government agency contracts represented ~1% of revenue in each period presented.
Accounting
Judgemental itemsUsage is metered and recognized as customers consume model capacity; period-to-period usage can drive volatility.
Impact: Revenue timing and comparability across quarters/periods
Subscription fees are recognized over the term, with upsell bridges to incremental consumption.
Impact: Shift in mix changes deferred revenue and recognized revenue patterns
CSPs bill end customers and remit net of fees while Anthropic often maintains separate end-customer contracts.
Impact: Affects reported revenue presentation and receivables concentration
Remeasurement can cause large non-cash gains/losses tied to valuation and assumptions.
Impact: Net income (loss) volatility unrelated to operating performance
Equity awards require valuation assumptions and drive non-cash expense.
Impact: Operating expense level and diluted equity value interpretation
Compute build-out includes lease obligations and capex incurred ahead of operational availability.
Impact: Balance sheet leverage indicators and operating vs investing cash flows
Similar listed companies
Publicly listed companies with a comparable business focus.
C3.ai, Inc.
AI · Services-Prepackaged Software
Global-E Online Ltd.
GLBE · Services-Prepackaged Software
Leet INC.
LTESF · Services-Prepackaged Software
Tyler Technologies, Inc
TYL · Services-Prepackaged Software
The Descartes Systems Group Inc
DSGX · Services-Prepackaged Software
ServiceNow, Inc.
NOW · Services-Prepackaged Software
Last updated: 26/10/2026