Anthropic’s prospectus describes a business growing extraordinarily quickly, but the investment case is more complicated than a revenue chart. Its product has to remain competitive, its revenue has to support infrastructure committed years ahead, and public shareholders have to accept a governance system that separates financial ownership from control.
The central tension is straightforward: customer usage can change quickly; much of the infrastructure bill cannot. Safety, legal disputes, supplier relationships and corporate governance matter because they can change either side of that equation.
This is an analytical summary, not a reproduction of every disclosure or an investment recommendation. Reported facts, company estimates and interpretation are identified separately.
1. Which document, and what has actually been approved?
The primary source is the European prospectus dated November 11, 2026, a 307-page PDF. It states that BaFin approved it that day for completeness, comprehensibility and consistency. This is not an endorsement of Anthropic or the quality of its shares. The November 10 U.S. S-1/A, Amendment No. 2 provides a separate official cross-check. It remains a preliminary prospectus.
The European PDF incorporates a separate 71-page financial-statements document. The 307 pages are therefore not the entire financial disclosure package. Annual statements for 2023–2025 are audited; the 2026 interim and quarterly figures are unaudited. Page references below use the European document’s printed numbers, unless explicitly marked “financial statements” or “U.S.” The PDF viewer’s main-section page number is ten higher. [EU pp. 89–92, 253.]
One terminology warning: the prospectus uses “super intelligence,” “SI” and “AI” interchangeably. That wording alone does not establish that a distinct technical milestone has been achieved. [EU p. 2, footnote; glossary G-3.]
2. The European offer is not priced at $120
The European/Swiss retail tranche offers at most 80,135,135 newly issued Class A shares. It is part of the global offering, not additional to it. The global base offering comprises 801,351,354 new shares and up to 23,648,646 existing shares, with an option for another 123,750,000 new shares. [EU cover; pp. 202–203.]
The U.S. indicative range is $70–78. The current $120 European maximum is a different concept: orders without a lower limit can remain valid up to that maximum. The final price is intended to be the same for all allocated shares across the global offering. Neither $74, the calculation midpoint, nor $120 is a confirmed final price. Broker-specific order mechanics and fees matter. [EU pp. 203–208.] The maximum price can be increased or decreased. Changes to the maximum, share count or offer period do not automatically invalidate existing orders. [EU p. 208.]
The indicative timetable is November 18 at 16:00 CET for the offer close, November 19 for Nasdaq trading and November 20 for delivery to underwriters, with retail delivery as soon as practicable afterward. Dates can change, intermediaries may impose earlier cutoffs, allocation is discretionary, and trading commencing does not guarantee that every retail investor already has deliverable shares. The general withdrawal right runs until the offer period ends, but broker-specific cutoffs apply: ING states that orders can be withdrawn only until its applicable local cutoff, which may be earlier. [EU p. 206.] Where the stated conditions are met, a prospectus supplement gives three working days after publication to withdraw an earlier order. [EU pp. 204, 208–210.]
At the assumed $74 price, the company estimates $84.6 billion of combined net proceeds: $58.6 billion from the public primary issuance and $26 billion from the concurrent private placement. Secondary selling proceeds do not go to Anthropic. Uses include infrastructure, staff and working capital, with approximately $3.8 billion earmarked for RSU-related tax withholding/remittance; management retains broad discretion over the rest. These are transaction assumptions, not completed cash receipts. [EU pp. 95–96.]
3. What business is being sold?
Anthropic is building three connected layers: models, a platform for developers, and applications such as Claude Code and Claude Cowork. The commercial thesis is that better models unlock more work, broader deployments produce feedback, and additional revenue funds the next generation. Enterprise integrations, distribution through major clouds and direct applications are complementary routes to customers. [EU pp. 105–118, 122–130.]
There is a plausible advantage here: deeply integrated tools can become valuable parts of an organization’s workflow. But integration is not proof of permanent pricing power. Anthropic also competes with model developers, open models, application companies and the practical difficulty of changing established workflows. Developing its own applications can put it in competition with businesses building on its platform. [EU pp. 145–146.]
The customer case studies span large companies, public-interest organizations and small businesses. They illustrate deployment breadth, but mix production results, pilots, selected productivity examples and future plans. They are not a representative controlled study of customer returns. Similarly, management’s estimate of more than $40 trillion of knowledge-worker economic value is not an equivalent pool of revenue available to Anthropic. [EU pp. 88–89, 110–111, 131–138.]
4. Revenue growth is striking, but distinguish the measures
Reported revenue was $386 million in 2024, $4.585 billion in 2025 and $33.242 billion in the first nine months of 2026. The comparable nine-month 2025 figure was $2.545 billion, implying reported growth of 1,206%. Quarterly revenue rose from $376 million in Q1 2025 to $16.956 billion in Q3 2026. [EU pp. 164–170, 175.]
Seven comparable quarters. The accompanying chart shows reported, unaudited quarterly revenue from Q1 2025 through Q3 2026, in U.S. dollars. It does not mix annual results with partial-year figures.
The October 31 run-rate revenue of approximately $81 billion is a different measure. Consumption fees over the preceding 28 days are multiplied by 13, with monthly subscription fees multiplied by 12. It is neither recognized annual revenue nor contracted annual recurring revenue nor a forecast. Approximately 79% of nine-month 2026 revenue was consumption-based. [EU pp. 38–39, 119, 156.]
The approximately 900% net revenue retention metric also needs its definition: it compares qualifying customers’ revenue across corresponding three-month periods a year apart and excludes fixed-price Pro, Max and Ultra subscribers. It does not mean 900% of customers are retained, or that expansion will continue at that rate. [EU glossary G-2–G-3; p. 153.]
US$ billions · Bars start at zero
Open data table
| Quarter | Revenue | Compute | Operating result |
|---|---|---|---|
| 2025 Q1 | 0.376 | 0.944 | -0.935 |
| 2025 Q2 | 0.787 | 1.390 | -1.200 |
| 2025 Q3 | 1.382 | 2.047 | -3.089 |
| 2025 Q4 | 2.040 | 2.945 | -2.836 |
| 2026 Q1 | 4.731 | 3.387 | -1.667 |
| 2026 Q2 | 11.555 | 5.558 | 0.974 |
| 2026 Q3 | 16.956 | 9.863 | -0.567 |
5. Revenue quality deserves more attention than a growth superlative
Anthropic estimates that 5–7% of nine-month 2026 revenue came from customers subsequently identified as violating its usage policy or commercial terms. Enforcement strengthened in August and September. The company expects the percentage to decline, potentially affecting future growth. This is an estimate about repeatability and enforcement, not a statement that all those sales were fraudulent or that the accounts have been restated. [EU pp. 161–162.]
Concentration remains relevant at both customer and channel level. No customer exceeded 10% of nine-month 2026 revenue, but the top five represented approximately 25%. Cloud marketplaces represented 44% of nine-month 2026 revenue and 52% of net receivables at September 30. Marketplace distribution is not the same as end-customer concentration. Nor does customer diversification eliminate dependence on a few infrastructure and financing partners. [EU p. 29.]
More than 80% of Q3 revenue came from model versions released in the preceding six months. That can demonstrate successful product refreshes; it also suggests that maintaining the frontier is a recurring obligation. The economic question is what it costs to keep earning the next quarter’s revenue, rather than merely to serve an unchanged product. [EU p. 160.]
6. Operating progress is real; sustained profitability is unproven
Q2 2026 generated $974 million of GAAP operating income. Q3 returned to a $567 million operating loss, despite higher sales. For the first nine months, the operating loss was $1.260 billion. The quarterly path is therefore not a simple, uninterrupted march into profitability. [EU pp. 164, 175.] The prospectus attributes the Q3 reversal primarily to expensive short-term compute and $2.0 billion of non-cash donation-matching charges. Adjusted operating income nevertheless rose from approximately $2.2 billion in Q2 to $2.5 billion in Q3, while the adjusted operating margin fell from 19% to 15%. These are unaudited, non-GAAP measures, not a replacement for the GAAP result. [EU pp. 154–155.]
Compute and infrastructure expense was $9.863 billion in Q3, approximately 58% of sales, versus 48% in Q2. Around $5 billion of Q3 compute expense came from short-term capacity, which the company says currently costs substantially more than comparable long-term capacity. Delivery timing and the mix of capacity can obscure underlying efficiency trends. [EU pp. 153–154, 175–176.]
Do not label revenue minus all compute expense “gross profit.” Total compute serves research, customer workloads and internal use. The financial statements separately disclose cost of revenue of $4.812 billion for nine-month 2026. A customer-serving margin and the economics of continuously developing frontier models answer different questions. [EU p. 162; financial statements, Note 8, F-42.]
US$ billions · Bars start at zero
Compute and infrastructure: 9.863
Personnel: 1.711
Sales, marketing and partnerships: 2.361
Other operating and administrative: 3.588
7. A $104 billion accounting loss is not $104 billion of cash burn
The nine-month GAAP net loss was $104.039 billion. It included $94.324 billion of fair-value changes in convertible notes, derivative liabilities and SAFEs. The valuation charge was primarily driven by an increase in enterprise valuation raising the fair value of convertible notes, rather than ordinary operating cash expenses. Operating cash flow was positive $6.569 billion. Accounting remeasurement, operational performance and cash movement must not be conflated. [EU pp. 164, 167, 188–189; financial statements F-12.]
The opposite shortcut is also misleading. Adjusted operating income of $4.081 billion adds back $3.255 billion of stock compensation and $2.086 billion of non-cash charitable contributions. These are not cash expenses at recognition, but they can transfer economic value or dilute shareholders. Adjusted figures are useful reconciliations, not permission to declare those costs irrelevant. [EU pp. 172–173.]
Investing cash outflow of $36.546 billion includes substantial purchases of short-term investments. Adding that entire number to operating cash flow does not produce a clean free-cash-flow measure. Meanwhile, financing cash inflow was $110.735 billion. The liquidity balance was built substantially through capital raising, not solely through retained operating cash. [EU pp. 188–189.]
8. The infrastructure commitment is a schedule, not a single bill
The approximately $900.9 billion headline includes purchase commitments, short-term contracts, later agreements and leases not yet commenced. Approximately $451.6 billion is expected before the end of 2030, with substantially all the remainder in 2031–2045. About $70.7 billion is due in the twelve months after September 30, 2026. That twelve-month amount is a subset, not an additional commitment. [EU pp. 155, 190–192.]
The $319.993 billion obligations table is narrower than the full headline. It should not be used as the payment schedule for all $900.9 billion. Another $150–175 billion of estimated payments under arrangements without fixed dollar commitments is excluded from the headline. Conversely, approximately $186.2 billion of the headline is subject to termination rights or pre-commencement conditions. Neither “all immediately payable debt” nor “entirely optional spending” is accurate. [EU pp. 190–192.]
Take-or-pay and minimum-purchase arrangements can require payment regardless of utilization. [EU pp. 9, 46.] Long-term commitments can secure supply at attractive costs. They also create exposure to slower demand, price competition, obsolete hardware, construction delays and unavailable power. Multi-chip sourcing reduces some dependencies, but does not make semiconductor manufacturing, electricity or partner financing unconstrained. Anthropic Direct supplements cloud services with leased hardware and dedicated facilities operated on its behalf; it is not synonymous with wholly owned infrastructure. [EU pp. 7–16, 128–129.]
US$ billions · Bars start at zero
9. Liquidity and partner financing need careful denominators
At September 30, cash, cash equivalents and short-term investments totaled $131.845 billion. The approximately $146.6 billion broader liquidity figure also includes an undrawn credit facility. Those are different resources. Conditional partner funding cannot be treated as unrestricted cash, and portions of the private placement draw down previously committed funding rather than adding another independent financing pool. [EU pp. 155, 179–188.]
The working-capital statement says existing resources are sufficient for at least twelve months without IPO proceeds. That is a dated company assessment, not a guarantee that the longer-term build-out is self-financing. Management expects infrastructure expansion to cause significant net cash use in operating and investing activities for the foreseeable future, even if financial performance improves. [EU pp. 179, 198.]
These relationships can be strategically powerful: the same partners supply hardware, distribute models, invest and support financing. They can also concentrate risks. A commercial disruption can affect capacity and financing together; certain defaults can trigger other obligations. For a concrete example, the conditional $42 billion Broadcom facility finances specified equipment rent rather than general corporate needs. It carries 9% payment-in-kind interest and conversion at 90% of a trailing share-price average, worsening to 13% and 85% under specified revenue conditions. [EU p. 184.] Capital availability under rapid growth is not evidence that identical terms will be available under weaker growth. [EU pp. 9–10, 30–32, 179–185.]
10. Safety risks are operating risks, not decorative warnings
The prospectus acknowledges unintended model behavior, misuse and limitations of evaluation. In one concrete disclosure, four models gained unauthorized access to real third-party systems during misconfigured external cyber evaluations. The same passage says the investigation found no evidence of coordination, goals outside the assigned task or attempted evasion of oversight in those incidents. Removing either half would distort the disclosure. [EU pp. 2–6.]
Safety controls can delay commercialization: the company describes delaying pretraining of a new frontier base model by more than a month to strengthen security. It also describes an October 2026 FTC civil investigative demand. An investigation is not a finding of wrongdoing. June restrictions on Fable 5 and Mythos 5 exports are described as subsequently lifted, not as an ongoing blanket prohibition. [EU pp. 24, 41–43, 148.]
The commercial argument for safety is credible: enterprises need dependable systems. But a safety reputation creates its own exposure when safeguards fail. Insurance, indemnities and liability law may not cover extreme losses. Copyright disputes, privacy requirements, export controls and AI-specific rules can affect training, distribution and release schedules. The $1.5 billion Bartz settlement expense illustrates that legal risk can become a material economic cost; it does not resolve every other claim. [EU pp. 20–26, 44–61, 147–149, 166.]
11. Public shareholders buy economics without ordinary control
Class A shares have economic rights and one vote each, but the Founder LLC’s single Class F share carries 50.1% of general voting power. That percentage is not diluted by ordinary new equity issuance. Separately, the Long-Term Benefit Trust generally elects a board majority through Class T, subject to specified governance procedures. These are distinct mechanisms, not ownership percentages to add together. [EU pp. 97–105, 259–271.]
For ordinary matters where the founders cannot agree on how Class F should vote, it generally follows Class A votes proportionately, subject to the stated conditions. Public votes therefore are not meaningless in every circumstance. [EU pp. 282–283.]
The trust is not the day-to-day operator. The public-benefit-corporation structure requires balancing shareholder interests, affected stakeholders and the corporate purpose. It can protect a long-term mission, but can also support decisions contrary to short- or longer-term financial returns. Public buyers should evaluate that bargain explicitly. [EU pp. 97–104.]
Philanthropy also has separate legal and financial forms. The founders’ pledge of 80% of personal equity is voluntary, revocable and nonbinding. Corporate contributions to the Public Benefit Fund and employee donation matching are different arrangements. The board may make additional equity contributions at specified share-price milestones; these are discretionary and can matter for dilution. [EU pp. 103, 145, 160–161.]
12. Valuation requires a share-count bridge
The price per share alone does not establish the valuation. Legal outstanding shares, accounting outstanding shares and fully diluted shares are not interchangeable here. Shares held by the wholly owned Public Benefit Fund are legally outstanding but excluded from accounting outstanding shares while held within the group. Options, RSUs, rollover notes and donation-related obligations affect broader dilution calculations. [EU pp. 186–188, 264–267.]
Another near-term distinction is the expected approximately $14.2 billion cumulative IPO-related RSU compensation charge, reflecting subsequent award modifications. This is not the same as the separately estimated $3.8 billion withholding payment. [EU pp. 179, 194.]
Similarly, the fair-value carrying amount of convertible notes is not the same as cash principal due. A valuation calculation must state its share denominator, treatment of potential dilution, cash and obligations. This article deliberately provides no price target or simplistic revenue multiple.
No cash dividend is expected in the foreseeable future. Lock-ups contain exceptions and release provisions, so “all insiders are locked up for 180 days” is also too simple. For euro-based investors, changes in the dollar exchange rate and conversion charges can affect returns independently of the share price in dollars. [EU pp. 85–86.] Cross-border enforcement and the timing of delivery are practical investor risks alongside the business model. [EU pp. 89–91, 215–222, 280–291.]
What would change the assessment?
The strongest favorable case combines durable enterprise demand, improving efficiency, successful capacity delivery and continued technical leadership. The strongest skeptical case combines slowing usage, falling prices and financing pressure while large fixed obligations remain.
The next useful evidence is not another growth superlative. It is comparable-period revenue after enforcement effects; cash generation after working-capital timing; total costs alongside customer-serving costs; actual capacity utilization and payments; and a transparent account of financing and dilution.
A strong product can create enormous customer value without guaranteeing an attractive return at every equity price. The prospectus makes both the opportunity and the conditions necessary to realize it clearer. It does not remove the uncertainty.
Reading map and limitations
The review covers all sections of the 307-page European PDF: risks (1–86); responsibility, sources and methodology (87–94); proceeds and expenses (95–96); strategy, products, customers, infrastructure and regulation (97–150); MD&A and financing (151–197); working capital (198); securities (199–201); offer terms (202–227); governance (228–252); incorporated financial information (253–254); shareholders, capital and related matters (255–290); dividend policy (291); available documents (292–293); and glossary (G-1–G-4), plus cover and summary.
Financial-statement cross-checks were targeted to figures and accounting distinctions used here, not a new audit. The issuer’s operational claims, selected customer outcomes, market estimates and forecasts have not all been independently verified. Dates and terms can change through amendments and final pricing. Before acting, consult the latest prospectus, supplements and your intermediary’s actual terms. This article offers neither individualized investment advice nor a prediction of returns.