As of November 15, 2026, Anthropic and the SEC provide public offering documents. This analysis follows document versions, accounting periods and metric definitions. Financial tables use USD billions unless shares, percentages or per-share prices are specified.
Anthropic roadshow launch announcement · SEC company submissions data
Establish the document and its date
The roadshow began on November 10. The SEC company index identifies the October 26 S-1, November 2 and November 10 S-1/A amendments, and five FWP free writing prospectuses. Six earlier DRS or DRS/A entries are historical draft submissions. A document’s date, index filing date and underlying event date can differ. This review covers the inventory and key sections below, emphasizing the latest US amendment. Regional materials establish their legal scope.
Anthropic roadshow launch announcement · Anthropic initial S-1 announcement · SEC original S-1 · SEC first S-1/A amendment · SEC second S-1/A amendment · SEC company submissions data
| Document family | Scope reviewed | Purpose |
|---|---|---|
| S-1 and two amendments | Business, MD&A, risks, governance, ownership and financial notes in the latest version | Check definitions, reconciliations and contractual conditions |
| IPO website FWP | Website captures, the 15-page fact sheet and founders’ letter | Compare the summary with original tables |
| Regional FWPs | Covers and relevant terms for the UK, Australia and Canada, EEA, and Japan | Identify jurisdiction and document differences |
| Official roadshow | Public player and complete captions, about 41 minutes 36 seconds | Locate the period attached to spoken metrics |
| Model disclosures | November 9 Fable/Mythos 5.2 release and key evaluations in the 264-page system card | Check capabilities, configurations and risks |
SEC second-amendment document index · SEC IPO website FWP · Anthropic IPO fact sheet · Anthropic founders’ letter · SEC UK materials FWP · SEC Australia and Canada FWP · SEC EEA FWP · SEC Japan amendment FWP · Anthropic system card index · Fable 5.2 and Mythos 5.2 announcement · Fable 5.2 and Mythos 5.2 system card · Anthropic official IPO website · Anthropic official roadshow captions
Media reporting has a different evidentiary role. Reuters’ November 13 account described demand based on unnamed sources. Keep expressions of interest, pricing and completed transactions at their respective stages. Public availability also has limits. The commercial presentation cannot provide undisclosed model weights, a complete training corpus inventory or future financial results.
Offering terms, secondary sales and proceeds
| Item | Disclosed amount or status | How to read it |
|---|---|---|
| Public offering | 825,000,000 Class A shares | 801,351,354 newly issued and 23,648,646 sold by existing holders |
| Indicative price | USD 70–78 per share | Pricing and effectiveness require subsequent steps |
| Underwriters’ option | Up to 123,750,000 shares for 30 days | Exercise changes proceeds and dilution |
| Concurrent private placement | 26.0 in Class N shares | A conditional transaction separate from the public offering |
| Estimated company net proceeds | 84.6 at USD 74 without the option | Includes the private placement after estimated costs. Secondary proceeds go to sellers |
Anthropic roadshow launch announcement · SEC second S-1/A amendment
Intended uses include operations, compute and infrastructure, personnel, and possible investments or acquisitions. The prospectus also estimates 3.8 for tax withholding and remittance associated with restricted stock unit settlement, using an assumed share price and tax rate. Transaction components, expenses and withholding determine resources available for deployment. Total fundraising cannot be inserted directly into a compute budget. Management retains broad allocation discretion.
Connect model performance to the business
The commercial path runs from models and APIs to Claude Chat, Claude Code and Cowork, then into organizations through direct sales, cloud partners and implementation. The roadshow’s proposition is that improved capabilities enable more tasks, valuable tasks expand deployments, and the resulting business supports further research and compute. Feedback, lower costs and higher usage interact along this path.
SEC second S-1/A amendment · Anthropic IPO fact sheet · Anthropic founders’ letter
An analytical framework follows from that proposition. Track task success, cost per accepted deliverable, continuing customer adoption and cash collection. A demonstration still has to become a deployed workflow with permissions, data access, integration, review and procurement. Each stage can take time or reduce the realized return. A benchmark score alone cannot identify which stage improved revenue.
Separate revenue from a run-rate snapshot
| Period or measurement date | Amount | Measure |
|---|---|---|
| Full year 2023 | 0.041 | Historical revenue |
| Full year 2024 | 0.386 | Historical revenue |
| Full year 2025 | 4.585 | Historical revenue |
| Q1 2026 | 4.731 | Quarterly revenue |
| Q2 2026 | 11.555 | Quarterly revenue |
| Q3 2026 | 16.956 | Quarterly revenue |
| First nine months of 2026 | 33.242 | Sum of three quarters |
| End of October 2026 | About 81 | Run-Rate Revenue |
The company annualizes consumption fees from the trailing 28 days by multiplying them by 13, then adds monthly subscription fees at the measurement date multiplied by 12. The measure changes with usage, pricing and active subscriptions. Keep quarterly revenue, year-to-date revenue and the dated run-rate together, then examine collection and subsequent retention.
The table implies approximately 46.7% sequential Q3 revenue growth and 1126.9% growth from the prior-year quarter’s 1.382. The base and observation window matter. Treating 81 as revenue already earned during the year mixes periods. Multiplying a quarter by four is a separate annualized scenario.
Check the customer denominator
At September-end, the company reported more than 11,900 organizations contributing at least USD 100,000 over the trailing twelve months. Its approximately 900% net revenue retention compares a qualified cohort’s current quarter with the equivalent quarter a year earlier. It excludes fixed-price Pro, Max and Ultra subscriptions. These boundaries affect how to interpret customer quality.
| Metric | Supported observation | Further evidence needed |
|---|---|---|
| Customers with at least USD 100,000 TTM revenue | More organizations reached the spending threshold | Timing, overlapping accounts and subsequent collections |
| Approximately 900% net revenue retention | Strong expansion in a specified base-period cohort | New-customer revenue, profit contribution and later cohort volatility |
| Enterprise case studies | Customers describe deployment and visible outcomes | Selection, comparison baseline and transferability |
Cases can help select a pilot. Enterprise deployments often include tailored systems, training and processes. Individuals need to test their own inputs and acceptance criteria. Keep customer counts, seats, active-user rates and actual time savings as different measurements to avoid equating adoption with productivity.
Compute efficiency and the spoken period switch
The fact sheet defines compute efficiency as revenue divided by compute and infrastructure expenses. Q3 gives 16.956 ÷ 9.863, approximately 1.719 times. Q2 was approximately 2.079 times. The company attributes part of Q3’s cost to additional capacity, including 5.0 under more expensive short-term arrangements. Delivery, utilization and pricing all affect a quarter’s efficiency.
The captions start with Q3 revenue at 30:36, then switch to year to date after 30:39 before stating 43% compute margin, 85% gross margin and 12% adjusted operating margin. Nine-month revenue of 33.242 and compute expense of 18.808 imply about 43.4% remaining after compute. Adjusted operating income of 4.081 implies about 12.3%. Preserve both the quarterly and cumulative periods.
Anthropic IPO fact sheet · Anthropic official roadshow captions
The latest financial statements classify expenses by nature. Their notes explain that cost of revenue and gross profit are no longer separate lines. This review did not find a complete bridge from the spoken 85% gross margin to that classification in the selected written materials. Retain it as management’s statement and obtain the cost-allocation definition before comparing peers. Revenue less all compute expenses is also affected by the split between training and serving.
SEC second S-1/A amendment · Anthropic official roadshow captions
Reconcile operating loss and adjusted profit
| Item | Amount | Meaning |
|---|---|---|
| GAAP operating loss | -0.567 | Operating result under the expense classification |
| Add stock-based compensation | 1.070 | An adjustment excluded by management |
| Add noncash charitable contributions | 1.960 | Another excluded adjustment |
| Adjusted operating income | 2.463 | Sum of the three amounts |
| Adjusted operating margin | About 15% | 2.463 ÷ 16.956 |
| Net loss | -23.291 | Also includes nonoperating items and taxes |
| Fair-value changes within the results | -21.936 | Convertible notes, derivative liabilities and SAFEs |
Adjusted operating income isolates performance after specified exclusions. Stock compensation still affects dilution, and charitable matching still allocates value. Noncash valuation changes strongly affect net loss, so examine cash flow and liability terms separately. Retaining GAAP, adjusted and cash measures for their respective purposes makes quarterly changes easier to follow.
revenue = 16.956
compute = 9.863
gaap_operating = -0.567
stock_compensation = 1.070
noncash_charity = 1.960
adjusted_operating = gaap_operating + stock_compensation + noncash_charity
print(round(revenue / compute, 3))
print(round((revenue - compute) / revenue * 100, 2))
print(round(adjusted_operating, 3))
print(round(adjusted_operating / revenue * 100, 2))What the cash flow statement establishes
| Measure | Period or date | Amount |
|---|---|---|
| Operating cash flow | Full year 2025 | -2.637 |
| Operating cash flow | First nine months of 2026 | 6.569 |
| Investing cash flow | First nine months of 2026 | -36.546 |
| Financing cash flow | First nine months of 2026 | 110.735 |
| Cash, cash equivalents and short-term investments | End of September 2026 | 131.845 |
Nine-month operating cash flow turned positive. Investing cash flow also includes purchases of short-term securities, prepayments and equipment. Classifying the whole investing outflow as research burn would combine different uses. Equating net loss with a cash decrease would do the same. Customer prepayments, supplier payment timing and working capital also affect operating cash flow.
This analysis does not calculate months of survival by dividing cash into total commitments. A funding model needs quarterly collections, payments, financing conditions, restricted uses and stress cases. Undelivered capacity’s payment dates and termination conditions can change the result. The disclosures identify potential pressure points without producing a stable countdown.
Build the 900.9 commitment total
| Component | Amount | Timing |
|---|---|---|
| Tabulated contractual obligations | 319.993 | At September-end, including purchases and commenced leases |
| Leases not yet commenced | 451.2 | Future undiscounted lease obligations |
| Short-term and similar arrangements | 17.4 | Related minimum payments at September-end |
| Later long-term and other commitments | 105.7 | Subsequent events |
| Later short-term arrangements | 6.6 | Subsequent events |
| Total | About 900.9 | Sum across categories and dates |
Approximately 451.6 is expected before the end of 2030, with substantially all the rest in 2031–2045. About 70.7 is expected in the twelve months after September-end. Arrangements without fixed dollar commitments add estimated payments of 150–175 outside the 900.9. About 186.2 involves pre-commencement termination rights or contingent conditions. Preserve the conditions, years and valuation basis.
These arrangements connect future supply with future demand. Examine annual maturities, delivery readiness, expected workloads and cost reductions. Termination clauses require contract-level reading. They cannot make every arrangement freely cancelable. Undiscounted multi-year payments also have a different time basis from one quarter’s expense.
Shareholders, suppliers and concentration
Concurrent placement participants include AMD, Broadcom, Google, Micron, Microsoft, NVIDIA and TSMC Partners. The company also buys capacity across chips and clouds. A participant can play capital, supply, distribution or competitive roles at once, with rights determined by separate contracts. The narrative of interchangeable capacity needs to be checked against hardware availability, software portability and payment conditions.
The prospectus says the five largest customers contributed approximately 25% of nine-month revenue, with no single customer above 10% in that period. Commitment length, the ability to reduce usage and cloud distribution affect durability. Related-person disclosures also cover Spark-linked financings and founder-related arrangements. Read the voting ownership table together with Class N economic rights before inferring strategic partners’ total economic holdings.
PBC, the LTBT and special voting rights
| Class or structure | Disclosed arrangement | Analytical question |
|---|---|---|
| Public benefit corporation | Directors balance shareholders, affected stakeholders and the public purpose | How commercial and mission trade-offs are reported |
| LTBT and Class T | Generally elects and removes a board majority. Class T has no economic rights | Board election mechanism and specified-seat procedures |
| Founder LLC and Class F | Fixed 50.1% of voting power on matters generally submitted to shareholders | Founder influence and continuing conditions |
| Class A | One vote per share, within the Class F structure | Relationship between public economic and voting rights |
| Class N and Class E | Class N generally has nonvoting economic rights. Class E employees vote on limited matters | Scope, conversion and employee procedures |
The LTBT selects directors. The board and management direct daily operations and strategy. One specified Class T seat involves the nominating committee and, in certain circumstances, an employee vote. The mission is built into governance. Its effects need to be observed through later reports, board actions and risk handling. Ownership and capital documents identify decision rights without establishing that every conflict will be resolved.
Model disclosures and access
| Configuration | Public positioning | Verify before use |
|---|---|---|
| Fable 5.2 | General-use configuration with additional limits on high-risk dual-use work | Surface, plan eligibility, version and safeguards |
| Mythos 5.2 | More permissive safeguards in dual-use domains, with direct access subject to vetting | Approved scope and relevant verification program |
| Claude Security | Enterprise product powered by model capabilities including Mythos | Product permissions and direct model access separately |
Anthropic system card index · Fable 5.2 and Mythos 5.2 announcement · Fable 5.2 and Mythos 5.2 system card
A system card documents capability, safety testing and deployment judgments. Service commitments, retention, tool permissions and organizational approval need separate checks. A model family’s appearance in the roadshow cannot establish access to every configuration for every subscriber. Record version, qualification and surface in one pilot worksheet.
Read evaluations in their experimental setting
| Evaluation setup | Reported result | Boundary |
|---|---|---|
| OSWorld 2.1 offline subset, 82 tasks | Fable 5.2 partial score 89.3%, strict pass rate 54.6% | The two metrics define success differently |
| Opus 5.5 rerun with the same configuration | Partial 87.2%, strict pass 53.2% | Preserve tools and reasoning settings |
| Five independent attempts, 1080p, up to 500 steps | Pass@1 averaged across attempts | This is not success after giving a task five chances |
These are vendor-reported evaluations on pages 234–237. Offline execution limits networking and the task mix. Partial checkpoints cannot replace end-to-end business acceptance. A team can use them to choose pilot tasks, then track failures, edits, tool costs and elapsed time. Changes to the benchmark, snapshot, budget or grading require a new comparable baseline.
The card contains extended safety sections, limitations and regressions. It helps identify risks and controls. Customer stories, internal code contribution and benchmark scores measure different objects. Internal code output still needs review, defect, rollback and delivery-value measures to establish productive outcomes.
SEC second S-1/A amendment · Fable 5.2 and Mythos 5.2 system card
Risks and remaining information gaps
| Risk source | Disclosure signal | Next observable |
|---|---|---|
| Capacity and demand mismatch | Long-term capacity and conditional commitments | Delivery, utilization and cost per task |
| Customer and channel concentration | Dependence on large customers and clouds | Continued use, collection and concentration |
| Law, intellectual property and safety | Litigation, access restrictions and misuse | New cases, disclosures and control evidence |
| Competition and migration | Changing prices and capabilities | Quality and cost on the same acceptance set |
| Governance and financing | Special rights, noncash valuations and financing conditions | Capital changes, board procedures and ability to pay |
This mapping from risks to observables is this article’s analytical framework. Public materials do not establish the full training recipe, all contract economics, future cash flows, repeatable ROI in every sector or the final offering outcome. Preserve gaps for withheld, redacted or summarized information instead of filling them with media estimates.
Build a reader’s review workbook
Fix the document version and currency first. For each claim, record the period, definition, page, calculation and status. Add timestamps for spoken material and snapshots and configuration for evaluations. S-1 versions, FWPs and syndicated reporting repeat information. Treat common-origin repetition as one evidence chain.
claim_id,source_version,period,unit,metric,reported_value,calculation,status
C01,S1A2_2026-11-10,Q3_2026,USD_billion,revenue,16.956,,reported
C02,S1A2_2026-11-10,Q3_2026,percent,adjusted_operating_margin,15,2.463/16.956*100,rounded
C03,system_card_2026-11-09,offline_set,percent,strict_pass_rate,54.6,,vendor_evaluation- Compare tables and notes across S-1 versions and record the periods covered by new data.
- Record the reconciliation and cash flow alongside every profit measure.
- Mark long-term targets as management expectations and identify observations that can update key assumptions.
- Pilot low-risk tasks on owned or authorized material, defining review, cost and failure-stop conditions first.
- When new documents arrive, update the version and differences while retaining the earlier judgment and its basis.
Revisit the same evidence chain
Business and technical judgments can change with new disclosures and your acceptance results. A useful review identifies which definition, period or condition changed the conclusion.
- Identify the document and version
- Check period and definition
- Reconcile financials and contracts
- Verify model and task conditions
- Record judgment, gaps and the next evidence
This is a selected reading and calculation of materials available as of November 15, 2026. Check later pricing, effectiveness, financial and model updates through the same document chain, retaining the version and period.