Anthropic S-1/Aamended IPO prospectus

Executive Summary

Anthropic sells Claude artificial-intelligence models and applications to businesses, developers and consumers through usage fees and subscriptions. It is expanding from conversational assistance into coding and multi-step business workflows, supported by frontier-model research and dedicated computing infrastructure. Revenue grew 1,206% Y/Y in 9M 2026, while operating losses narrowed; the much larger net loss chiefly reflects noncash increases in convertible-financing valuations. The IPO combines company fundraising with existing-holder sales but remains unpriced, preventing assessment of valuation or buyer ownership. Investors must weigh exceptional adoption against enormous infrastructure obligations and governance that prioritizes safety and public benefit alongside shareholder returns.

Business Model and Growth Drivers

Usage expansion drives the business: consumption revenue reached $26.3B in 9M 2026, vs $2.13B a year earlier; subscriptions contributed $6.94B, vs $420M. Customers generating at least $100,000 of trailing-year revenue exceeded 11,900, vs 950, and supplied 74% of trailing-year revenue. Net revenue retention was ≈900%: revenue from qualifying customers over three months was nine times their comparable year-earlier spending, excluding fixed-price consumer plans.

Claude Platform lets developers embed models and agents in products; Claude Code handles software development, while Claude Cowork extends agentic work beyond coding. Anthropic began combining Cowork and Claude.ai in 9/2026 to connect conversations with multi-step tasks. Management says Opus 5.5 reduced customer cost per task ≈40% while preserving margins. Recently released models generated over 80% of Q3 revenue, illustrating how continued model improvement supports demand—and how quickly products turn over.

Distribution spans direct channels, AWS, Google Cloud and Microsoft Azure. Anthropic retains direct customer contracts and pricing control while paying partner fees. Cloud marketplaces supplied 44% of 9M revenue; the top five customers supplied ≈25%. Those cloud companies are simultaneously infrastructure suppliers, investors and competitors.

Management is moving toward “Anthropic Direct,” combining cloud capacity with dedicated facilities and multiple chip architectures. Its first directly provisioned facility opened in 2026. This can improve efficiency but adds construction, power, financing and utilization exposure.

The Offering

• Size and valuation: share counts, price range and proceeds remain blank; implied market capitalization, revenue multiple, buyer stake and post-IPO fully diluted shares cannot yet be calculated. The historical 6/30/2026 public-benefit issuance implies 16.9B fully diluted shares, assuming full Amazon-note conversion—not an IPO share count.
• Structure and proceeds: Anthropic and selling stockholders will sell Class A shares; only primary proceeds fund the company. Planned uses include compute, infrastructure, staffing and employee-equity tax withholding. Goldman Sachs and Morgan Stanley lead the syndicate; listing is sought on Nasdaq under ANT, with a 30-day over-allotment option.
• Capitalization and dilution: preferred stock and convertible notes will convert, employee restricted stock units will settle, and IPO proceeds will be added net of expenses and withholding payments. After-IPO cash, equity and tangible assets per share remain unquantified, so the gap between the purchase price and tangible backing cannot yet be assessed.
• Resale supply: lock-ups generally run up to 180 days, with founders and larger holders restricted up to 365 days, subject to staged releases and waivers. Some non-founder employees and officers may sell part of their holdings immediately after listing; later releases track earnings announcements.

Key Financial Highlights

Amounts in $B; parentheses indicate losses or outflows.

• Earnings bridge: $94.3B of noncash financing revaluation losses largely reflected higher enterprise valuations; other expense added $10.4B. Adjusted operating income was $4.08B, excluding stock-based compensation and noncash charitable contributions, among other items.
• Compute economics: expense reached $18.8B but fell to 57% of revenue. Q3 nevertheless required $5.0B of expensive short-term capacity, showing that demand growth can precede economical supply.
• Equity deficit: $158B of convertible preferred stock is outside stockholders’ equity, while accumulated losses total $156B; the deficit is not equivalent to borrowing. Outstanding debt principal was $6.20B.
• Cash conversion: noncash losses and $3.26B of stock compensation explain much of the divergence from net loss; receivables absorbed $7.08B.
• Commitments: ≈$895B of purchase and lease obligations include ≈$69.8B due over the following year and ≈$448B before year-end 2030—far beyond current operating cash generation.

Governance and Control

• Founder LLC’s non-economic Class F share controls 50.1% of general voting power; the Long-Term Benefit Trust separately elects a board majority. Public Class A holders cannot override either mechanism.
• Anthropic is a controlled public benefit corporation and will use the governance-committee independence exemption. Safety and societal objectives may justify delaying commercially attractive releases.
• Outstanding options and restricted stock units total 2.90B shares, equivalent to 17% of the earlier June fully diluted benchmark. Options carried $119B of intrinsic value; IPO effectiveness would trigger $8.2B of cumulative compensation expense.
• The Public Benefit Fund received 5% of June fully diluted shares; discretionary issuances at share-price milestones and expanding employee-plan reserves create further dilution.

Risk Assessment

• Auditor transition: Ernst & Young replaced Grant Thornton for FY2025; no disagreements were reported, but internal-control effectiveness was not audited.
• Infrastructure mismatch: long-term minimum payments can precede usable capacity or customer revenue, requiring additional financing.
• Demand durability: customers can reduce consumption quickly, while competition and cheaper models may weaken extraordinary expansion rates.
• Partner dependence: cloud partners can influence capacity, distribution and pricing while competing with Claude.
• Safety and legal exposure: misuse incidents, export restrictions, unresolved government-access disputes and copyright claims can interrupt commercialization or create liabilities.
• Founder dependence: losing founders or scarce technical leaders could disrupt frontier research, execution and customer confidence.
• Minority-holder economics: founder and trust control, public-benefit issuances and staged resale releases limit investor influence and can dilute returns.

What to Watch

• Enterprise expansion: the $100,000-plus customer cohort and retention should show broader workflow adoption; weaker spending would challenge the expansion thesis.
• Product efficiency: Opus 5.5’s reported task-cost reduction must translate into sustained usage and economical serving, rather than dependence on costly emergency capacity.
• Infrastructure funding: later filings should show committed capacity entering service and generating cash; delays or underutilization would expose the funding mismatch.
• IPO economics: completed pricing and capitalization tables must reveal buyer ownership, tangible backing and dilution before valuation can be judged.