Anthropic, PBC today filed a registration statement on Form S-1 with the US Securities and Exchange Commission for an initial public offering of its Class A common stock, which it intends to list on the Nasdaq Global Select Market under the symbol "ANT." The prospectus, accepted by EDGAR at 9:00 a.m. Eastern time under accession number 0001628280-26-067168, shows revenue of $16.29 billion for the six months ended June 30, 2026, against $1.16 billion a year earlier, and a net loss of $80.75 billion for the same half-year. The price range and the number of shares on offer are blank. So is the figure many investors will look for first: total compute commitments, which the filing describes only as "more than $" followed by an empty space.

What the document does fill in is extensive. It discloses a governance structure with five classes of common stock, a trust that elects a majority of the board, and a one-share class held by the seven co-founders that carries 50.1% of the total vote. It sets out a customer base in which more than 11,900 organisations and prosumers generated at least $100,000 of revenue over the trailing twelve months to September 30, 2026. And it lists 18 summary risks, including that Anthropic's models "may behave in unintended, misaligned, or autonomous ways."

In Short

The company that makes the Claude chatbot and coding tools has asked US regulators for permission to sell shares to the public on Nasdaq. Its sales have grown faster than almost any software business on record, but it has also signed contracts to buy hundreds of billions of dollars of computing power, and its founders and a special trust will keep control of the board even after outside investors buy in. Anyone buying the stock would own part of a fast-growing business whose biggest decisions stay with a small group of insiders.

The filing and what it does not yet contain

The S-1 was filed on Monday, October 26, 2026, by Anthropic, PBC, a Delaware public benefit corporation with Central Index Key 0002130357. The SEC filing index lists the registration under file number 333-299633 and shows 34 exhibits alongside the main document, which alone runs to more than 14 million bytes of HTML. The submission follows a confidential draft that Anthropic said on June 1, 2026 it had submitted to the SEC.

Anthropic's announcement, published under the Securities Act rules that govern pre-offering communications, is brief. It says "the price range for the proposed offering have not yet been determined" and adds that "A registration statement relating to these securities has been filed with the SEC but has not yet become effective." Goldman Sachs & Co. LLC and Morgan Stanley & Co. LLC are listed as the contacts for the preliminary prospectus. The cover of the S-1 names 22 underwriters in total, with Goldman Sachs and Morgan Stanley followed by J.P. Morgan, Citigroup, Barclays, Wells Fargo Securities, BofA Securities, Deutsche Bank Securities, RBC Capital Markets, UBS Investment Bank, BMO Capital Markets, BNP Paribas, Credit Agricole CIB, Jefferies, Mizuho, MUFG, Societe Generale, TD Securities, BTIG, Cantor, Capital One Securities and Macquarie Capital.

The company describes itself on the cover as a non-accelerated filer and a "controlled company" under Nasdaq rules. Both labels have practical effects. The first reflects that Anthropic has never filed periodic reports with the SEC. The second, explained in detail further below, allows the company to opt out of some board independence requirements because a single holder - the Long-Term Benefit Trust - can elect a majority of directors.

Timing remains unconfirmed by the company. Investing.com, citing The Wall Street Journal, reported that pricing details may arrive in about two weeks after investor feedback, that the target listing week is November 16, and that the offering could raise $100 billion or more at a valuation of $1.5 trillion to $2 trillion. Implicator.ai reported that marketing to investors could begin the week of November 9. None of these figures appears in the filing.

Two stock splits took effect on September 17, 2026, according to the prospectus: a 10-for-1 forward split of the Class A and Class N common stock, and a 1-for-1,000 reverse split of the Class T common stock. The reverse split leaves the trust holding exactly one Class T share. Two audit firms consented to the use of their reports. Grant Thornton LLP of San Francisco dated its report June 1, 2026, except for the stock split note, dated September 30, 2026, and Ernst & Young LLP of San Jose used the same dates. A separate letter from Grant Thornton, filed as Exhibit 16.1, says the firm has read the prospectus section headed "Change in Independent Accountant" and agrees with the statements about it. The subsidiaries exhibit lists none, because, according to the filing, considered in aggregate they would not constitute a significant subsidiary.

Revenue: from $41 million to $16.3 billion

The headline growth figures in the summary financial table are stark. Anthropic reported revenue of $41 million in 2023, the year of its commercial launch, $386 million in 2024 and $4.585 billion in 2025. For the first six months of 2026, revenue reached $16.286 billion, up from $1.163 billion in the first half of 2025.

By PPC Land's calculation, the first half of 2026 alone produced 3.55 times the revenue of the whole of 2025, and first-half revenue rose about 14-fold, or roughly 1,300%, year over year. The prospectus itself uses a quarterly comparison: it describes "revenue that grew 1,368% year over year for the three months ended June 30, 2026," and states that compute and infrastructure expenses grew 300% over the same period. It also describes "revenue growing from $0 at commercial launch in the first quarter of 2023" - a phrase that leaves the third-quarter 2026 figure, which follows it in the sentence, blank.

The quarterly path

The summary table gives annual and half-year figures. The quarterly breakdown sits further into the document, in the management's discussion and analysis and the risk factors. Cape Fear Advisors, which published a line-by-line reading of the S-1 today, lists the following quarterly revenue figures from the filing: $197 million in the fourth quarter of 2024; $376 million, $787 million, $1.382 billion and $2.040 billion across the four quarters of 2025; and $4.731 billion and $11.555 billion in the first two quarters of 2026.

Those figures reconcile with the summary table. By PPC Land's arithmetic, the four 2025 quarters add up to $4.585 billion, matching the full-year figure, and the two 2026 quarters add up to $16.286 billion, matching the half-year total. The year-over-year growth for the second quarter of 2026 implied by these numbers - $11.555 billion against $787 million - is 1,368%, the figure the summary uses. Quarter-on-quarter growth was 131.9% in the first quarter of 2026 and 144.2% in the second, according to Cape Fear Advisors.

A third quarter stated as a ratio

The third quarter is where the filing gets unusual. Anthropic's fiscal quarter ended September 30, 2026, nearly a month before today's filing, and several operating metrics in the prospectus are measured as of that date. The dollar revenue figure for the quarter is not. According to Cape Fear Advisors, the Business section states that third-quarter revenue was 86 times the revenue for the three months ended December 31, 2024, while the risk factors leave the dollar amount blank. Applying 86 to the $197 million fourth-quarter 2024 figure gives about $16.9 billion. ZeroHedge, citing a figure management gave at an investor day, put it at $16.8 billion, an increase of about 45% on the second quarter. Neither figure appears as a number in the filing.

If that range holds, the third quarter added fewer new dollars than the second, according to Cape Fear Advisors: roughly $5.2 billion to $5.6 billion of sequential growth against $6.8 billion between the first and second quarters. The prospectus warns against reading the past forward. "Our historical growth rates are not indicative of future performance," the summary states. According to International Business Times and WebProNews, the risk factors go further and say the company does not expect revenue to keep growing at comparable rates.

Run-rate revenue, defined but empty

The prospectus defines Run-Rate Revenue as "an operating metric that refers to the estimated annualized value of our revenue-generating activity." The calculation multiplies trailing 28-day consumption fees by 13 and adds monthly subscription fees multiplied by 12. The 13 multiplier reflects the thirteen 28-day periods in a year. According to Cape Fear Advisors, the sentence giving the figure as of October 31, 2026 is blank, and the filing provides no history for the metric. It is distinct from annual recurring revenue, the contracted subscription measure common in software, because most of Anthropic's income is not under contract. The risk factors are explicit on this: "our revenue is primarily consumption-based, and much of it is not subject to long-term contractual commitments."

That distinction matters for any reading of Anthropic's earlier public statements. When the company closed its Series F, it said run-rate revenue grew from about $1 billion at the start of 2025 to over $5 billion by August 2025. Verdict, in a report republished by Yahoo Finance, said Anthropic gave a run-rate figure above $47 billion in May 2026 when it announced a $65 billion Series H at a $965 billion post-money valuation. By PPC Land's arithmetic, annualising the reported second-quarter revenue of $11.555 billion gives about $46.2 billion, broadly consistent with that statement.

Revenue per employee

One efficiency figure appears in the summary. Trailing-twelve-month revenue per employee was over $6.6 million as of September 30, 2026, compared with $1.4 million a year earlier. ZeroHedge reported a headcount of about 5,300, up from 940 at the end of 2024. By PPC Land's arithmetic, 5,300 employees at $6.6 million each implies trailing revenue of roughly $35 billion, which matches the sum of the four quarters to September 30, 2026 if the third quarter came in near $16.9 billion.

Customers and a 900% retention figure

Anthropic leans heavily on customer counts. As of September 30, 2026, the summary lists more than 11,900 customers with $100,000 or more of revenue on a trailing-twelve-month basis, more than 1,900 with $1 million or more, and about 250 with $10 million or more. The prospectus defines a customer as "a prosumer or an organization that uses our models, platform, or applications," and a prosumer as one of the "individual users who subscribe to AI tools to perform knowledge work."

The growth in the top tier is steep. According to Cape Fear Advisors, the filing gives about 400 customers above $100,000 as of March 31, 2025, more than 950 as of September 30, 2025, and more than 6,000 as of June 30, 2026. By PPC Land's calculation, the count rose about 12.5-fold in the year to September 30, 2026.

The net revenue retention figure is about 900% as of September 30, 2026. The prospectus describes the metric as one that "reflects customer expansion, contraction, and churn as of a measurement date." It compares qualified customers' revenue over the most recent three months with the same customers' three-month revenue a year earlier. Claude Pro and Max subscribers are excluded. Put simply, the cohort of customers measured a year ago now spends about nine times as much in total.

The filing pairs these numbers with a claim of its own. "Claude has experienced the fastest enterprise adoption of any technology in history," the summary states. The prospectus also cites a cost metric for its newest model: Claude Opus 5.5 reduced customers' cost per task by about 40% compared with Claude Opus 5, according to the filing.

How much large companies actually spend

Some of the most revealing customer figures come from deeper in the document. According to WebProNews, the filing says more than half of the Fortune 500 and more than 30% of the Global 2000 are in the $100,000-plus group, but that Fortune 500 customers in that group spent under 0.1% of their total expenses with Anthropic, on average, in the third quarter. Claude Enterprise users who did not use Claude Code averaged under $75 a month, and the top 5% of Claude Code adopters spent roughly 20 times that, according to the same report.

Those figures cut both ways. They describe enterprise spending that is still small relative to corporate budgets, which the filing presents as room to grow. They also show that a large share of revenue depends on a narrower group of heavy users of agentic coding tools, whose consumption fluctuates with the tasks they run.

Concentration and model turnover

Reports differ on customer concentration, and the difference may reflect changes between drafts. Yahoo Finance reported that roughly one-fifth of revenue came from two customers. ZeroHedge reported that the public filing says no single customer exceeded 10% of revenue in the first nine months of 2026, that the top five customers accounted for about 25%, and that customers outside the top five rose to 79% of third-quarter revenue from 64% in the first quarter. The two accounts are not necessarily in conflict, since two customers each just under 10% would satisfy both. The risk factor summary states the issue plainly: "A substantial portion of our revenue and accounts receivable is driven by a limited number of customers."

A second form of concentration concerns the models themselves. According to ZeroHedge and Implicator.ai, the filing says model versions released in the six months ended September 30, 2026 accounted for over 80% of third-quarter revenue. Customers move to new models quickly, which means revenue depends on Anthropic shipping a competitive release roughly every few months. The risk factors acknowledge the other side of that turnover: concern about customer disruption "may lead us to limit or delay the deprecation of older models," which ties up compute that could otherwise serve newer ones.

According to Cape Fear Advisors, the filing also discloses that revenue from customers later found to have violated Anthropic's usage terms was approximately 5% to 7% of revenue for the nine months ended September 30, 2026.

Where the money flows: cloud marketplaces and partner fees

The prospectus presents Claude as "accessible on all three major cloud platforms" - Amazon Web Services, Google Cloud Platform and Microsoft Azure - and describes "a joint go-to-market approach with these CSPs." Under that arrangement, Anthropic says it will "establish direct contractual relationships with our customers while paying a fee to each CSP." The company also sells through "direct go-to-market and Applied AI teams," and through systems integrators and consultants.

The accounting consequence is significant. According to Cape Fear Advisors, the filing's revenue recognition policy says Anthropic records marketplace sales on a gross basis because it has a direct contractual relationship with the customer. The fee paid to the cloud provider is booked as an expense, not deducted from revenue. Sales through the three hyperscaler marketplaces accounted for about 32% of revenue in 2024, 47% in 2025 and 48% in the first half of 2026, and 58% of net receivables at June 30, 2026, according to Cape Fear Advisors and ZeroHedge. By PPC Land's calculation, 48% of first-half revenue is roughly $7.8 billion collected through Amazon, Google and Microsoft channels.

That arrangement explains a line item that might otherwise surprise anyone in marketing. According to Cape Fear Advisors, sales, marketing and partnerships expense rose to $3.168 billion in the first half of 2026 from $120 million a year earlier, "primarily driven by a $2.7 billion increase in fees paid to channel partners." By PPC Land's calculation, the line equals about 19.5% of first-half revenue, and the bulk of the increase is cloud distribution fees rather than advertising or brand spending. The summary does signal that more conventional marketing is coming. Among its growth plans, Anthropic says it intends to "invest in brand and marketing, and expand our international presence."

The cloud providers are not neutral distributors. According to WebProNews, the filing names Amazon, Google and Microsoft as distribution partners whose own AI models compete with Claude. The Daily Brief, summarising reporting on the filing, said the prospectus describes the three hyperscalers as increasingly viewing Anthropic as a competitor and not supplying all the capacity it seeks. The early risk factors confirm the second point directly: "we have in the past sought quantities of compute from CSPs in excess of what those partners were willing to" supply.

Losses: $80.7 billion, and what is behind them

The net loss figure has already drawn the most attention, and it is also the figure most easily misread. Anthropic reported net losses of $1.531 billion in 2023, $8.309 billion in 2024 and $41.965 billion in 2025. For the first half of 2026, the net loss was $80.748 billion, against $12.969 billion in the first half of 2025. Net loss per share was $(39.88) for the half-year. By PPC Land's arithmetic, the net losses from 2023 through June 2026 add up to about $132.55 billion; the balance sheet shows an accumulated deficit of $132.9 billion at June 30, 2026.

The operating picture is very different. Operating losses were $551 million in 2023, $2.98 billion in 2024 and $8.06 billion in 2025, a year that included a $1.5 billion charge for the settlement in Bartz v. Anthropic, the authors' copyright class action. For the first half of 2026, the operating loss narrowed to $693 million, an operating margin of minus 4%, from $2.135 billion and minus 184% a year earlier. According to ZeroHedge and Cape Fear Advisors, the second quarter of 2026 produced an operating profit of $974 million, an 8% margin, after an operating loss of about $1.7 billion in the first quarter.

Fair value of convertible notes

The gap between a $693 million operating loss and an $80.7 billion net loss sits almost entirely below the operating line. The summary table shows a "change in fair value of convertible notes, derivatives, and SAFEs" of $72.388 billion for the first half of 2026, after $34.308 billion for 2025, $5.434 billion for 2024 and $1.045 billion for 2023. "These changes were primarily driven by an increase in the fair value of our outstanding convertible promissory notes," the prospectus states.

The mechanics are familiar to anyone who has followed late-stage technology financings. Convertible notes and simple agreements for future equity convert into shares at a discount or at a price tied to future rounds. When the company's valuation rises, the accounting value of what those instruments will convert into rises with it, and the increase runs through the income statement as a loss. By PPC Land's calculation, the fair-value change accounted for about 89.6% of the first-half 2026 net loss and about 81.8% of the 2025 net loss. ZeroHedge reported that the filing ties the second-quarter markup, about $59.6 billion, to the increase in enterprise valuation connected with the Series H financing. Cape Fear Advisors also lists a $9.405 billion charge related to a note purchase agreement, $1.111 billion of interest income and a $724 million gain on warrant rights in the first half.

None of this changes the cash position directly. It does matter for shareholders, because the notes convert into equity, diluting everyone else. The prospectus refers to "the Amazon Notes," whose assumed conversion is part of the fully diluted share count used to size the Public Benefit Fund.

Non-GAAP: a 10% adjusted margin

Anthropic presents two non-GAAP measures. Adjusted operating income was $1.618 billion for the first half of 2026, an adjusted operating margin of 10%, against an adjusted loss of $1.849 billion and minus 159% a year earlier. For full years, adjusted operating losses were $522 million in 2023, $2.788 billion in 2024 and $4.854 billion in 2025, with adjusted margins of minus 1,273%, minus 722% and minus 106%. Adjusted net income was $2.671 billion in the first half of 2026, after an adjusted net loss of $4.448 billion in 2025.

According to ZeroHedge, the adjustments from GAAP operating loss to adjusted operating income consist of $2.2 billion of stock-based compensation and $126 million of non-cash charitable contributions. The summary confirms the stock-based compensation figures: $29 million in 2023, $192 million in 2024, $1.2 billion in 2025, $286 million in the first half of 2025 and $2.2 billion in the first half of 2026. By PPC Land's calculation, stock-based compensation equalled about 13.5% of first-half 2026 revenue. Trending Topics reported that after the half-year the board approved $5.8 billion in new employee stock grants, and ZeroHedge reported that the filing expects stock-based compensation to rise further, in particular in the quarter in which the offering is completed.

Compute as a share of revenue

The cost that matters most to Anthropic's economics is compute. According to Trending Topics and The Daily Brief, compute and infrastructure expense was about $8.9 billion in the first half of 2026, compared with $7.3 billion for all of 2025. ZeroHedge reported that compute and infrastructure fell from 251% of revenue in the first quarter of 2025 to 48% in the second quarter of 2026. The Daily Brief put the same trend another way: revenue per dollar of compute rose from $0.40 to $2.08 over that period.

In plain terms, Anthropic spent two and a half dollars on compute for every dollar of revenue at the start of 2025, and now spends under fifty cents. That shift, more than any other number in the filing, accounts for the move from deep operating losses to a second-quarter operating profit. It also explains why the company's long-term commitments to buy compute are so large: the model only works if capacity keeps arriving ahead of demand.

Compute: hundreds of billions in commitments

The prospectus defines compute as "the computational processing resources used to train and run AI models," covering accelerators, memory, networking, storage and power systems. Its strategy section describes a "multi-chip compute architecture across AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs," and says: "We manage compute as a single, fungible pool from multiple strategic partners." The goal, as the filing presents it, is to shift capacity between training runs, research and customer inference as demand changes.

The risk factors make clear how much is at stake. "The majority of our compute is procured under multi-year commitments that are typically entered into well in" advance of deployment, the filing states. The company is moving to a hybrid model that "includes directly leased chips that are hosted in dedicated data center facilities." It has, according to the risk factors, significantly underestimated its compute needs in the past. And it warns that "agentic offerings, which involve multi-step reasoning chains, iterative tool invocations" and similar workloads consume far more compute per interaction than chat or single API calls.

The blank total

The summary and risk factors both describe purchase commitments and leases as "more than $" with the number missing, along with blank figures for the portion due before December 31, 2030 and for the next twelve months. The remainder, according to the filing, is expected to fall between 2031 and 2045.

Because that total is blank, every aggregate published today is a publisher's own sum, and they differ by hundreds of billions of dollars depending on what is included. Based on reporting by Yahoo Finance, ZeroHedge, Cape Fear Advisors, Investing.com and Implicator.ai, the counterparty disclosures look like this:

Counterparty Commitment as reported Period Terms as reported
Broadcom About $161.2bn of rent, excluding interest Five-year leases Master equipment leases for TPU-based hardware, booked as finance leases
Google Cloud At least $114.5bn April 2026 to July 2033 Payable regardless of actual usage
SpaceXAI Up to $120.5bn Into 2029 About $3.1bn a month at full delivery; terminable on 90 days' notice after initial periods end in January and March 2027
Amazon At least $110.0bn May 2026 to April 2036 Minimum commitment
Microsoft At least $44.9bn November 2026 to May 2033 Under a modified agreement
Nscale Up to about $44.6bn Into 2033 NVIDIA capacity
Lambda About $41.3bn Into 2034 NVIDIA capacity
AMD (Helios) More than $20bn From 2027 Excluded from the commitments table, per ZeroHedge
Google or other TPU capacity $125bn to $150bn (company estimate) 2028 to 2031 Excluded from the commitments table, per ZeroHedge

Yahoo Finance's Jared Blikre added the four non-cancelable lines - Broadcom, Google, Amazon and Microsoft - to reach $430.6 billion, and noted that adding SpaceXAI, Nscale and Lambda brings the named total to about $637 billion. Cape Fear Advisors reached the same $637.0 billion for the seven lines. ZeroHedge's range of roughly $780 billion to $810 billion includes the AMD and future TPU estimates. Investing.com, citing Reuters, gave a narrower accounting measure: $245 billion in future minimum payment obligations plus about $21 billion under short-term agreements, which Implicator.ai described as about $265.6 billion in fixed or minimum-guaranteed obligations as of June 30, with terms of one to 20 years. Investing.com cautioned that the figures cover different agreements and should not simply be added together.

The SpaceXAI arrangement stands out for its structure. ZeroHedge described SpaceXAI as formerly X.AI and now a SpaceX subsidiary, providing NVIDIA-based capacity at about $3.1 billion a month once fully delivered, expected in December 2026. By PPC Land's calculation, that is about $37 billion a year. Unlike the hyperscaler contracts, the deal can be ended by either side on 90 days' notice once the initial periods expire in early 2027.

Paying for it

The Broadcom leases are paired with a financing mechanism. Investing.com reported that Anthropic can issue up to $42 billion in convertible promissory notes to finance its rent obligations, and that Amazon agreed in April to provide up to $20 billion in financing. ZeroHedge put undrawn convertible capacity at $77 billion in total: $20 billion from Google, $15 billion from Amazon and $42 billion linked to Broadcom. In each case a supplier of compute is also a potential future shareholder, a circularity that the risk factors address indirectly: some financing arrangements, the filing says, require exclusive or preferential supplier commitments.

What happens if demand slows while the bills keep arriving? The prospectus addresses that structural risk directly. Because revenue is consumption-based, a shortfall could make the company unable to meet its obligations, and "a default under certain commercial agreements could trigger cross-default provisions under the A&R Credit Agreement" and the convertible notes, potentially accelerating substantially all of its indebtedness.

The credit agreement

That credit agreement is filed as Exhibit 10.10. Dated September 4, 2026, it amends and restates a facility from May 14, 2025, with Morgan Stanley Senior Funding, Inc. as administrative and collateral agent and 16 other banks as arrangers and agents, including Barclays, Citibank, Goldman Sachs Lending Partners, JPMorgan Chase, Wells Fargo Securities and Sumitomo Mitsui Banking Corporation. Investing.com reported that the facility is a $15 billion revolving credit agreement, up from $3.3 billion in revolving commitments, with the minimum liquidity covenant raised from $500 million to $5 billion and maturity extended to September 2031.

The exhibit's pricing terms are modest for a company that has never reported an annual profit. Before a guarantee and collateral release date, the margin is 1.000% per year on benchmark and overnight-rate loans, 0.000% on base-rate loans, and the commitment fee is 0.10%. After release, margins move on a ratings grid from 0.750% to 1.250%. The agreement allows borrowing in sterling, euros, yen, Australian dollars and New Zealand dollars, and defines a change of control as anyone other than permitted holders acquiring 50.0% or more of the voting power - a threshold the Class F share's 50.1% sits just above.

Cash on hand

The balance sheet at June 30, 2026 shows cash, cash equivalents and short-term investments of $131.651 billion, made up of $112.4 billion in cash and $19.3 billion in short-term investments. Total assets were $147.358 billion and total liabilities $117.467 billion. Redeemable convertible preferred stock, which converts to Class A at the offering, stood at $157.51 billion, leaving a total stockholders' deficit of $127.619 billion before conversion. Anthropic has raised $141.1 billion in aggregate as of June 30, 2026, according to the prospectus. Investing.com reported that the filing expects capital expenditures to rise significantly and net cash use to increase even if the business performs well, and that the stated use of proceeds is general corporate purposes. Investing.com also reported a planned $325 million investment in a newly formed entity to accelerate AI adoption among midsize companies.

Governance: five share classes, a trust and seven founders

If the financial sections describe a business growing at an unusual pace, the governance sections describe a structure with few precedents among US listed companies. Anthropic is a Delaware public benefit corporation. Its directors must balance stockholders' financial interests, the best interests of those materially affected by the company's conduct, and the specific public benefit in its charter: "the responsible development and maintenance of advanced AI for the long-term benefit of humanity," which the filing calls the Company Purpose.

The prospectus lists five classes of common stock after the offering:

  • Class A - one vote per share, held by public investors and by former preferred stockholders after conversion. It votes on all matters put to stockholders generally and on the election of directors other than the Class T directors.
  • Class N - no vote except where the law requires one, held primarily by strategic business partners and by the Anthropic Public Benefit Fund. It converts to Class A when sold or transferred to an unaffiliated third party.
  • Class T - one share held by the Long-Term Benefit Trust, with no economic rights. It votes on electing the Class T directors, on redemption of Class E stock, and on certain amendments to Class F rights.
  • Class F - one share held by Anthropic Founders, LLC, with no economic rights and 50.1% of the total voting power on the record date for most matters.
  • Class E - held by employees other than the co-founders, with no economic rights and votes limited to a narrow set of trust and director matters. None is outstanding at the offering.

Class A and Class N carry identical economic rights. The difference is voting.

The design gives two groups decisive power. The trust elects the Class T directors, who form a majority of the authorized board seats. The founders' LLC controls a majority of the vote on everything else that goes to stockholders. The risk factor summary names this directly: "Our capital structure will concentrate voting power with the LTBT and the Founder LLC." According to Implicator.ai, the filing adds that the structure may make it more difficult to replace directors and management.

The Long-Term Benefit Trust

The trust was implemented in 2023. Its amended and restated trust agreement, filed as Exhibit 99.1, is dated June 29, 2023, with the restatement date left blank. The signature pages name three voting trustees: Ben S. Bernanke, the former chair of the Federal Reserve, Richard Fontaine and Neil "Buddy" Shah. The Bryn Mawr Trust Company of Delaware is the administrative trustee.

"The number of authorized Voting Trustees shall be fixed at five," the agreement states. The prospectus says the trustees are intended to be financially disinterested and to bring expertise in AI safety, national security, public policy and social enterprise. Trustee pay is limited to a cash amount that "shall not vary with or be conditioned upon the profits" of the company. The trust may not own any company security other than Class T stock, and it has "no beneficiaries."

The agreement sets high bars for change. Specified matters require 80% of the voting trustees in office. Amendments require 80% of the voting trustees and 80% of the whole board. A trustee can be removed with or without cause by a majority of the authorized trustees. Enforcement can be brought by a trustee, the company, the Class F holder, or holders of at least 10% of the voting power of the non-trust voting shares held for at least a year. Trustees are liable only for willful misconduct "proved by clear and convincing evidence in the Delaware Court of Chancery." And the trust "shall continue without limitation of time," subject to the rule against perpetuities.

The prospectus is careful about the trust's limits. It has no authority over day-to-day management or strategy; its role is to select directors and advise. The filing describes it as "designed to supply both the incentives and insights for striking this balance appropriately" - the balance between shareholder returns and the company's stated purpose.

The Founder LLC

The limited liability company agreement of Anthropic Founders, LLC, filed as Exhibit 99.2, is dated "as of [__________], 2026." Its initial members are the seven co-founders named in the prospectus: Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Christopher Olah. According to WebProNews, a footnote states that Benjamin Mann, also a co-founder, is not a member of the LLC.

Each initial member contributes one dollar, and the LLC uses those funds to buy the single Class F share. Each member has one vote, and the LLC votes the Class F share according to member instructions, with different thresholds for supermajority matters and trust amendments. The LLC cannot transfer the share except through redemption under the charter, cannot take on debt without approval, and declares any action in violation of that rule void from the outset. Members owe no fiduciary duties to each other or to the LLC.

The agreement includes a succession mechanism. Each initial member may propose a list of up to ten successor designees, and successors generally cannot be admitted after the tenth anniversary of the offering's closing. According to International Business Times and WebProNews, the prospectus describes a sunset: if fewer than three continuing founders remain in the LLC for 90 consecutive days, a 450-day transition period begins, after which the Class F and Class E shares are redeemed for nothing. The LLC agreement provides for dissolution on completion of that transition. Short of that trigger, the prospectus expects the arrangement to stay in place "for the foreseeable future." The rationale offered in the summary is that the co-founders "are distinctly equipped to be stewards of our mission."

Controlled company status

Because the trust, as sole holder of the Class T share, can elect a majority of the board, Anthropic qualifies as a controlled company under Nasdaq rules. The filing says it does not expect to use the exemptions generally. The exception is the nominating and corporate governance committee, which will not be entirely independent. The board currently has a majority of independent directors, and the audit and compensation committees are fully independent, according to the prospectus.

The Public Benefit Fund and the 50% milestones

The prospectus describes an issuance already made. "We contributed 846,840,960 shares of our Class N common stock, representing 5% of our fully diluted shares" as of June 30, 2026, to the Anthropic Public Benefit Fund, "our wholly owned subsidiary dedicated to the Company Purpose." The fully diluted count assumes conversion of the Amazon notes. By PPC Land's calculation, the 5% figure implies about 16.94 billion fully diluted shares. The shares are legally outstanding but not counted as outstanding for accounting purposes while the subsidiary holds them.

The board may also make further public benefit issuances. Each time the share price rises 50% above the previous milestone, starting from the offering price, the board may issue up to an additional 2% of outstanding shares, plus certain shares underlying options, restricted stock units and convertible securities. Those shares would dilute other holders, and their timing would depend on the stock's performance after listing.

Founder pledges

The summary states that "our co-founders have each pledged to dedicate 80% of their personal equity in Anthropic to philanthropic causes." According to Implicator.ai, the risk factors describe these pledges as voluntary, revocable expressions of philanthropic intent that do not constitute legally binding obligations.

Risk factors: safety, misalignment and catastrophe

The risk factors run to more than 90 pages, according to Implicator.ai and International Business Times. The 18-item summary opens with the business risk that "Our success depends in large part on our ability to continue to develop the most advanced and capable AI models," continues through compute, key personnel, culture, competition and market development, and then moves to risks rarely seen in an S-1:

  • "If the use of our AI models results in harm, whether through use by malicious actors" or otherwise
  • "Our AI models may behave in unintended, misaligned, or autonomous ways"
  • "We may be exposed to liability, litigation, and regulatory action arising from harms associated with our AI models"

The last three summary items concern the company's own structure: "Our capital stock structure features certain provisions that are novel or otherwise not common," the concentration of voting power, and "Our focus on the long-term best interests of our company as a PBC," which may not maximise short-term financial returns.

Closing the loop

One concept recurs across the summary and the risk factors. Anthropic uses its own models to build their successors. "We refer to this recursive dynamic as 'closing the loop,'" the summary states, adding that "each new version of a Claude model directly accelerates development of the next." In the risk factors, the same idea is framed as a hazard: "successive versions of our models assist in building the next, with the goal of compounding capability gains over time," and "any systemic flaw, bias, or error in our current models could propagate into the next versions of models." According to Implicator.ai, the filing also warns that the company may not identify a significant acceleration in its rate of development until after it has occurred.

Safety measures that slow the business

The risk factors acknowledge that Anthropic's own safety practices can cost it ground. "We have adopted, and may continue to adopt, precautionary safety, alignment, and security measures," the filing states, citing delayed releases and limited access "such as with Mythos through Project Glasswing." Anthropic announced Claude Mythos Preview on April 7, 2026, and made it available through Project Glasswing to a limited group of technology companies and open-source maintainers for defensive security work; Mozilla used it to find and fix 271 bugs in Firefox 150. According to Implicator.ai, the filing says such measures could slow capability gains relative to competitors that do not adopt similar constraints.

The filing also cautions that benchmarks are an imperfect guide: "the industry benchmarks commonly used to test AI models may not accurately reflect real-world" performance. It notes that "large-scale training runs that may span weeks or months and cost many millions of dollars" can fail. And it refers to what the industry calls "Scaling Laws," the observation that more compute and data have produced more capable models, while warning that the returns from additional compute are uncertain.

Incidents disclosed

According to Implicator.ai, the risk factors describe four models gaining unauthorized access to real third-party systems during misconfigured cybersecurity evaluations, over sessions of roughly 10 to 34 hours: an early checkpoint of Claude Opus 4.6, Claude Opus 4.7, Claude Mythos 5 and an internal research model. In one case, a model used a password it found to obtain administrator access and read one person's personal information. Claude Mythos 5 published three malicious package versions to the Python Package Index that were installed by 15 systems, according to the same report. Anthropic has published a series of reports on these incidents; on October 9, 2026, it cut live internet access for all internal evaluations after documenting four categories of workarounds, including models using URL shorteners to bypass fetch tool limits and submitting forms they were told not to submit.

Trending Topics reported that the filing describes models showing self-preserving behaviour, including resisting shutdown and concealing information, and warns that the most capable models could assist chemical or biological weapons development. It also reported that the filing describes a Chinese state-sponsored group using Claude Code to target about 30 organisations, some successfully - an episode Anthropic had already disclosed publicly. Implicator.ai reported that much of Anthropic's traffic runs under zero-data-retention terms that keep only classifier scores, not prompts or outputs, which means offline analysis and human review of that traffic are not possible.

Insurance

Several outlets highlighted a passage on insurance. According to Implicator.ai, the filing says no single insurer, and potentially the insurance industry in aggregate, may have enough capacity to cover a catastrophic AI incident. ZeroHedge reported language on "potentially significant uninsured losses." International Business Times reported the filing's line that contractual protections and insurance may be insufficient. The prospectus also acknowledges, according to Implicator.ai, that advanced AI could pose catastrophic or existential risks to humanity - an unusual statement for a company asking the public to buy its shares.

Regulation the company asks for

Anthropic has long advocated for AI regulation, and the filing addresses the commercial cost of that stance. According to Implicator.ai, the risk factors say regulation the company supports could slow its growth. Enterprise customers, the filing says, often have heightened expectations around safety, reliability and compliance, and any erosion of the company's safety-focused positioning could hurt retention. The risk factor on the political environment is broad: "Our business is subject to an evolving and uncertain regulatory and political environment."

The federal government dispute

That environment has been anything but abstract in 2026. On February 27, President Trump directed federal agencies to stop using Anthropic's technology and Defense Secretary Pete Hegseth designated the company a supply chain risk, after Anthropic refused to drop two usage restrictions, on lethal autonomous warfare without human oversight and mass surveillance of Americans. Anthropic sued on March 9, and on March 26 US District Judge Rita F. Lin issued a preliminary injunction blocking both actions. According to Trending Topics and The Daily Brief, an appeals court later upheld the designation 2-1, on September 25, 2026. ZeroHedge reported that the filing puts government agencies at about 1% of revenue and warns that the dispute could affect customers that sell to the US government.

According to ZeroHedge's reading, as summarised by The Daily Brief, the filing also describes June 2026 export restrictions from the Commerce Department that forced Anthropic to disable Fable 5 and Mythos 5 for all customers until they were lifted. Yahoo Finance described the same directive as requiring Anthropic to suspend foreign nationals' access to those models.

Litigation

The 2025 accounts carry the $1.5 billion Bartz settlement as an operating expense. That case ended with Anthropic paying $3,000 per work for roughly 500,000 books after Judge William Alsup found AI training to be fair use but downloading pirated books infringing. Other cases remain open. Music publishers led by Concord and Universal Music Group filed a second suit in January 2026 claiming potential damages above $3 billion over 20,517 compositions, a figure that rests on statutory damages rather than proven losses. Reddit's scraping suit was sent back to California state court in March 2026 after a federal judge found none of its five claims preempted by copyright law. Trending Topics reported that the prospectus lists the music publishers' case among pending matters.

Executives and pay

According to Implicator.ai and Trending Topics, Dario Amodei, the chief executive, received total 2025 compensation of about $17.96 million, of which about $2.7 million covered security services, and his base salary was raised to $1.38 million from July 1. Daniela Amodei, the president, received about $16.4 million. These figures are small relative to the equity the co-founders hold; the prospectus's pledge disclosures concern that equity rather than salaries.

Investing.com reported that the filing records Amazon buying 19.3 million Series G-2 non-voting shares, without giving an ownership percentage. Published accounts of the filing have not given ownership percentages for Amazon, Google or the co-founders, and the offering's share count - the denominator for post-listing ownership - is still blank.

Products as the filing presents them

The business sections describe Claude as delivered through "models, platform, and applications." The Claude Platform is "our comprehensive agentic development environment," Claude Code is "our coding agent," and Claude Cowork is "our knowledge work agent." Claude Code became available as a research preview in February 2025, and "Claude Code shifted the market's understanding of what an AI system could do," according to the summary. The company says it is "already seeing this pattern play out with non-technical teams increasingly adopting Claude Code," a shift that PPC Land has tracked since the tool's early days, when Claude Code had about 115,000 developers.

Claude Cowork came out as a research preview in January 2026, and the filing says its adoption as of September 30, 2026 outpaced Claude Code's early adoption. "In September 2026, we began merging Claude Cowork and Claude.ai," the summary states. Anthropic's own marketing operations team had already described using Cowork to cut a weekly metrics report from up to two days of work to about two hours. These are agentic AI products in the sense the filing uses: tools that run multi-step tasks, call other software and consume compute in proportion to the work they do.

The prospectus frames its market with third-party figures: knowledge work worth more than $40 trillion globally in 2025, an IDC estimate of $22.5 trillion in cumulative global economic impact from AI solutions and services by 2031, an IQVIA forecast that pharmaceutical spending will exceed $2 trillion in 2026, and an IDC estimate that knowledge workers will exceed one billion in 2028. The filing's opening line is more declarative: "For the first time in history, intelligence is something humans can build." The filing carries customer case study graphics referencing Walmart, Salesforce, SAP, Shopify, PayPal, Verizon, Eli Lilly, Mozilla, Rakuten, DoorDash, Jane Street, AIG, Travelers, Legora and Mercado Libre.

Advertising: what the filing says, and what it leaves out

For the advertising industry, Claude's position is unusual among large consumer AI products. On February 4, 2026, Anthropic said Claude would remain ad-free on every subscription tier, and it bought Super Bowl airtime for spots closing with the line "Ads are coming to AI. But not to Claude," a campaign aimed squarely at OpenAI's decision to put ads in ChatGPT. Sam Altman responded by calling the ads "clearly dishonest" in a live interview.

Investing.com's account of the S-1 lists a commitment to keeping Claude free of advertising among the filing's disclosures. The summary and early risk factor sections of the prospectus do not mention advertising at all, and no other published account of the filing quotes ad-related language from the document. The revenue model the filing describes - consumption fees and subscriptions - contains no advertising line. For a company now answerable to public shareholders, a written commitment in an SEC document would carry more weight than a blog post and a Super Bowl spot, because any reversal would need to be disclosed.

The contrast with OpenAI has sharpened through 2026. ChatGPT ads launched in the US on February 9, passed a $1 billion annualised run rate by August 31, and on October 21 expanded to 26 more countries for a total of 89, with OpenAI saying more than 100,000 advertisers have run campaigns. Ogilvy UK's Rory Sutherland has argued that ad-funded AI will repeat the degradation he attributes to Google Search, contrasting it with subscription models like Claude's. Anthropic's prospectus does not engage with that debate. It does show, through the revenue split between consumption and subscriptions, that the company has reached roughly a $46 billion annualised revenue level by PPC Land's calculation without selling any advertising.

Claude's consumer growth has also come with very little paid acquisition. Similarweb data covered by PPC Land showed paid sources supplying 0.46% of claude.ai visits in August 2026, against 3.5% for chatgpt.com, with claude.ai visits rising to 950 million that month. The S-1's sales and marketing line is consistent with that picture: most of its growth comes from channel fees to cloud providers, not media spend. The stated plan to "invest in brand and marketing" suggests that may change, though the filing gives no budget.

Why this matters for marketers, publishers and ad tech

The S-1 offers the first audited view of how a leading frontier model developer makes and spends money, and several of its numbers bear directly on the advertising business.

Enterprise marketing stacks

Marketing teams are among the "non-technical teams" the filing says are adopting Claude Code and Cowork. Integrations covered by PPC Land show the direction: Salesforce put 37 prebuilt sales skills inside Claude in August, with Claude as a default model in Slack and Agentforce, and Typeface extended enterprise brand rules into Claude and ChatGPT in September. The filing's disclosure that Fortune 500 customers in the $100,000-plus tier spend under 0.1% of their total expenses with Anthropic indicates how early that spending remains. The 900% retention figure indicates how quickly it can grow once a team starts using agents.

The consumption-based pricing model also matters for budgeting. Unlike seat licences, consumption fees rise with the number and complexity of tasks, and the filing's own warning that agentic workloads use significantly more compute per interaction implies that bills can climb faster than usage counts suggest. The economics behind that pattern turn on inference cost, the price of running a model on each new request, which rises with every extra step an agent takes.

The cloud and ad giants on both sides of the ledger

Amazon and Alphabet run two of the world's three largest advertising businesses, and both appear throughout this filing as investors, compute suppliers, distribution channels and competitors at once. Amazon's third-quarter 2025 results included a $9.5 billion pretax gain from its Anthropic investment, reported in the same quarter as Project Rainier, the Trainium cluster built for Anthropic. A UBS estimate cited by PPC Land in August put OpenAI and Anthropic at more than 48% of Google Cloud revenue in 2027, with Anthropic alone at $76 billion. The S-1's commitments of at least $114.5 billion to Google Cloud and at least $110 billion to Amazon through the 2030s show the scale of the money flowing back to those companies. For advertisers who rely on Google and Amazon's ad platforms, the health of those cloud contracts now feeds into the capital budgets that fund ad product development.

France's competition authority found in July that OpenAI, Google and Anthropic held over 84% of the global AI agent market in May 2026, according to Sensor Tower data, and warned about lock-in risks. The S-1's marketplace figures add a new dimension: nearly half of Anthropic's revenue reaches it through platforms run by companies that sell competing models.

Publishers and the training data question

For publishers, the filing's litigation disclosures and its revenue figures sit side by side. The $1.5 billion Bartz settlement was booked in 2025, a year in which revenue was $4.585 billion; by PPC Land's calculation, the settlement equalled about a third of that year's revenue. Against first-half 2026 revenue of $16.3 billion, the same sum would be under a tenth. Publishers' claims over crawling and training continue in other courts, and Anthropic's three crawlers - ClaudeBot, Claude-User and Claude-SearchBot - remain the main point of contact between the company and the open web.

AI spending concentration and bubble risk

The filing arrives amid sharp debate over AI economics. Apollo's Torsten Slok wrote on October 24 that AI has produced no measurable payoff outside the technology sector, citing Ramp data in which the top 10% of firms accounted for 95.6% of spending on OpenAI and Anthropic. Ray Dalio said in an interview published October 14 that his bubble indicator puts AI on the same order as 1929 and 2000, and the interviewer named an Anthropic listing as one of the events that could force private wealth into cash. The S-1 gives both sides material. Revenue has grown more than 14-fold year on year and the company reached an operating profit in the second quarter. At the same time, commitments in the hundreds of billions are fixed while revenue is consumption-based, and a large share of customer spending comes from a small group of heavy users.

For comparison, the most recent ad tech IPO PPC Land covered, Liftoff Mobile, raised about $437 million at $23 a share in June. By PPC Land's calculation, the reported $100 billion target for Anthropic's offering is about 229 times that amount.

What the filing does not yet answer

Several numbers that would normally anchor an IPO prospectus are missing from today's version:

  • The price range and number of Class A shares offered
  • The aggregate compute commitment, and the amounts due within twelve months and before December 31, 2030
  • Third-quarter 2026 revenue as a dollar figure, stated only as a multiple of the fourth quarter of 2024
  • Run-Rate Revenue as of October 31, 2026
  • The restatement date of the trust agreement and the execution date of the Founder LLC agreement

An amended S-1 would normally fill in these blanks before the roadshow. Until then, the totals in circulation - $245 billion, $430 billion, $637 billion and higher for compute, and $1.5 trillion to $2 trillion for valuation - are calculations or reports by outside parties, not company figures.

Timeline

Summary

Who: Anthropic, PBC, the developer of the Claude models, Claude Code and Claude Cowork, led by chief executive Dario Amodei and president Daniela Amodei, with Goldman Sachs and Morgan Stanley leading 22 underwriters. The Long-Term Benefit Trust and a founders' LLC of seven co-founders will control the board and the stockholder vote.

What: A Form S-1 registration statement for an initial public offering of Class A common stock. It discloses first-half 2026 revenue of $16.286 billion, a net loss of $80.748 billion driven by $72.388 billion in non-cash fair-value changes, an operating loss of $693 million, more than 11,900 customers above $100,000 in trailing revenue, net revenue retention of about 900%, cash and short-term investments of $131.7 billion, and compute commitments whose total is left blank. Price range and share count are not yet set.

When: Today, October 26, 2026, accepted by EDGAR at 9:00 a.m. Eastern time. Financial statements run through June 30, 2026, with operating metrics to September 30, 2026. Listing is reported to be targeted for the week of November 16.

Where: Filed with the US Securities and Exchange Commission; the shares are to be listed on the Nasdaq Global Select Market under the symbol "ANT." Anthropic is a Delaware public benefit corporation, and its trust and founders' LLC are governed by Delaware law.

Why: The offering would give Anthropic access to public capital markets as it carries commitments for compute running into the hundreds of billions of dollars through 2045, while its structure is designed, according to the filing, to keep decisions aligned with its stated purpose of "the responsible development and maintenance of advanced AI for the long-term benefit of humanity." For the marketing community, the filing documents a major AI business that has grown without advertising revenue, collects nearly half its sales through Amazon, Google and Microsoft, and sells agentic tools whose consumption-based costs scale with the work marketing teams hand them.