Who Owns Anthropic? Shareholders, Stakes and Control in the 2026 S-1

The filing is on EDGAR. Next I’ll open the S-1 itself and pull the principal stockholders section.
The EDGAR HTML didn’t surface the table in a text search, so I’m opening the prospectus text and secondary transcriptions that quote the ownership footnotes.
The prospectus HTML is truncated before the ownership table. I’ll pull the section from EDGAR anchors and from transcriptions that quote the footnotes.
The cover is confirmed on EDGAR. I’ll pull the table of contents and any transcribed ownership rows so the percentages can be checked against the filing itself.
The October 26, 2026 S-1 is on EDGAR and is real. The ownership grid itself could not be read cell by cell from the SEC HTML, and the percentage columns in that preliminary prospectus are blank.

The filing is verified; the table is not fully machine-readable

Anthropic, PBC filed a Form S-1 on October 26, 2026, accepted at 9:00:11 a.m. Eastern. EDGAR identifies it as accession 0001628280-26-067168, file number 333-299633, CIK 0002130357, registrant Anthropic, PBC, Delaware, SIC 7372. The primary document is anthropicpbc-sx1.htm. EDGAR’s full-text index returns that same file as the only S-1 hit for the phrase “Principal and Selling Stockholders” on that date. The cover, read directly from the SEC HTML, is a preliminary prospectus: price range, shares offered by the company, and shares offered by selling stockholders are all blank. Class A is to list on Nasdaq Global Select Market as ANT. [1] [2] [3] [4]

The section is titled Principal and Selling Stockholders, not “Principal Stockholders.” A transcription of the supplied PDF places it on physical PDF pages 226–229, with the ownership narrative and footnotes on pages 227–228 and the next section, Description of Capital Stock, starting on page 229. The SEC HTML fetch confirmed the cover and then truncated well before that section, so the numeric grid was not extracted from EDGAR itself. What follows is reconstructed from that section’s footnotes in the transcription, checked against a secondary reading of the same table. Where those two agree, the share totals are reported. Where the filing leaves a blank, the blank is left blank. [5] [5] [6]

Percentages before the offering are not in the document

The section says beneficial ownership is as of September 30, 2026, and that the pre-offering percentage is based on outstanding Class A, outstanding Class N, and one Class T share, after giving effect to the preferred-stock conversion, convertible-notes conversion, and RSU settlement. Those Class A and Class N denominators are blank in the text. The same is true of the post-offering denominators. A reading of the table states that the percentage columns are blank, so ownership percentages of each class, and of total shares, cannot be taken from this preliminary S-1. [5] [6]

The only voting figure that is a stated number, rather than a blank percentage column, is structural: the single Class F share, issued to Founder LLC immediately before the offering closes, is entitled to 50.1% of the vote on matters submitted to stockholders generally. All Class A votes together are capped at 49.9%. Class T, held by the Long-Term Benefit Trust, elects a board majority and is excluded from that general-vote denominator. Class N has no vote except as required by law. Class F and Class T have no economic rights. [3] [5]

Reconstructed holder table, before the offering

Counts below are Class A beneficial ownership as of September 30, 2026, unless another class is named. They include options exercisable within 60 days and RSUs whose service condition is satisfied within 60 days, as the footnotes define them. Totals are the sum of those footnote components. They match the table totals reported by a secondary reading of the same S-1. The “% of class” and “% of total shares” columns are not stated in the filing. Shares to be sold are blank on the cover and are not in the recovered footnotes.

Named holder	Class	Shares beneficially owned before the offering (Sept. 30, 2026)	% of that class	% of total shares	Citation
Dario Amodei (economic; footnote 1)	Class A	240,637,980	blank	blank	PDF pp. 227–228, Principal and Selling Stockholders, fn. (1)
Shares subject to Dario Amodei’s voting proxies (footnote 2)	Class A	4,621,120	blank	blank	Same section; count is the residual in fn. (9)(i) and is stated in secondary readings of the table
Dario Amodei total, including proxies	Class A	245,259,100	blank	blank	Same section
Daniela Amodei (includes spouse options and RSUs)	Class A	241,408,940	blank	blank	fn. (3)
Krishna Rao	Class A	9,497,900	blank	blank	fn. (4)
Reed Hastings	Class A	9,410	blank (legend treats under 1% as *)	blank	fn. (5)
Christopher Liddell	Class A	9,410	blank	blank	fn. (6)
Vas Narasimhan (includes Bodhi LLC)	Class A	176,050	blank	blank	fn. (7)
Yasmin Razavi	Class A	not stated; Spark Capital shares are excluded and disclaimed	blank	blank	fn. (8)
All executive officers and directors as a group (7 persons)	Class A	496,360,810	blank	blank	fn. (9)
Tom Brown	Class A	244,173,040	blank	blank	fn. (12)
Jack Clark	Class A	242,271,840	blank	blank	fn. (13)
Jared Kaplan	Class A	242,351,170	blank	blank	fn. (14)
Sam McCandlish	Class A	231,809,780	blank	blank	fn. (15)
Christopher Olah	Class A	242,882,420	blank	blank	fn. (16)
Long-Term Benefit Trust	Class T	1	100% of the authorized/outstanding Class T share described in the section	not an economic share; general-vote % not applicable	fn. (10)
Founder LLC	Class F	1, issued immediately prior to closing, not part of the Sept. 30 Class A/N/T percentage base	1 of 1 Class F	50.1% of general stockholder voting power; no economic %	fn. (11); cover

Footnote components, which are what the transcription actually prints:

• Dario, excluding proxies: 237,813,080 record or trust shares + 2,185,400 options + 639,500 RSUs = 240,637,980. [5]
• Daniela: 238,319,540 record or trust + 2,185,400 options + 130,550 spouse options + 639,500 RSUs + 133,950 spouse RSUs = 241,408,940. [5]
• Rao: 153,960 record + 7,843,940 options + 1,500,000 options in a trust he controls = 9,497,900. [5]
• Hastings and Liddell: 9,410 RSUs each. Narasimhan: 169,770 at Bodhi LLC + 6,280 RSUs = 176,050. [5]
• Group line: 481,077,470 already-beneficial Class A + 13,845,290 options + 1,438,050 RSUs = 496,360,810. The 481,077,470 figure equals the identified directors’ and officers’ record shares plus 4,621,120. That residual is the proxy block. A secondary reading of the table states the same 4,621,120 and the same 245,259,100 Dario total. [5] [6]
• Brown 241,348,140 + 2,185,400 + 639,500 = 244,173,040. Clark 239,446,940 + the same option and RSU pair = 242,271,840. Kaplan 239,526,270 + the same pair = 242,351,170. McCandlish 228,984,880 + the same pair = 231,809,780. Olah 240,057,520 + the same pair = 242,882,420. [5]

The seven-person group is the three named executive officers plus four non-employee directors (Hastings, Liddell, Narasimhan, Razavi). Brown, Clark, Kaplan, McCandlish, and Olah are in the selling-stockholder footnotes, not in that group total. Their Class F vote is not included in any co-founder’s personal voting-power percentage. Each continuing founder has one vote inside Founder LLC on how the Class F share is cast, and each can vote personal Class A shares independently. [5] [7]

Who is named, and who is not given a countable row

Amazon and Google are not given share counts in the recovered principal-stockholder footnotes. Elsewhere the prospectus is reported as saying that, after preferred stock and certain notes convert, each will hold greater than 5% of outstanding capital stock, in non-voting Class N, and that the exact counts are blank in this filing. That is a capital-stock statement, not a filled row in the ownership table, and it is not a percentage of a voting class. Class N converts to Class A only on a sale or transfer to an unaffiliated party. [6] [8]

Do not treat outside estimates as table entries. They disagree. PitchBook’s reading of the prospectus puts Google at about 7% and, on PitchBook’s own calculation, Amazon at roughly one-fifth if notes convert in full, subject to a 12% cap. Other commentary still cites an older roughly 14% Google figure and a 14.99% Google cap. Those are not cells in the September 30 table. [9] [8]

Spark Capital is named only in Yasmin Razavi’s disclaimer. The footnote lists Spark Capital Growth Funds IV, V, and VI and related entities, address 200 Clarendon Street, Floor 59, Boston, and says she disclaims beneficial ownership of their shares. No Spark share count was recovered from the footnotes. PitchBook identifies Spark as the only pure-play venture firm named in the filing and says it invested about $777 million; that is not a table percentage. [5] [9]

The seven co-founders are identified as selling stockholders. How many shares each will sell is blank, as is the size of the primary offering. The Wall Street Journal’s reading is consistent with the table at a rounded level: each of the seven holds more than 200 million shares. [10] [9]

What this means if you are trying to underwrite control versus economics

Economic ownership and control are different instruments. The Class A counts above are the economic stakes the filing actually numbers. Control of ordinary stockholder votes is one Class F share at a fixed 50.1%, undiluted by new Class A issuance. Control of a board majority is one Class T share at the Long-Term Benefit Trust. Amazon and Google can be greater-than-5% economic holders and still have no vote unless they sell Class N to an unaffiliated buyer. Public Class A buyers are buying a claim on economics inside a 49.9% aggregate vote cap, with the percentage of the company each founder owns still unstated because the company left the share denominator blank. [3] [6]

A final amendment that fills the price range should also fill the outstanding-share counts and the percentage columns. Until that amendment, any percentage of Anthropic attributed to a named holder other than the Class F 50.1% vote and the single Class T share is an estimate, not a figure from this S-1 table.

---

Recent Findings Supplement (October 2026)

The Anthropic, PBC Form S-1 (registration statement for its IPO) was filed with the SEC on October 26, 2026 (accession no. 0001628280-26-067168; CIK 0002130357) and is publicly available on EDGAR.[1][2]

The main document (anthropicpbc-sx1.htm) is a large HTML filing (~14 MB) containing the prospectus. The “Principal and Selling Stockholders” section appears in the prospectus (corresponding to physical PDF pages 226–229 in the filing).[3][3]

The section discloses beneficial ownership of capital stock as of September 30, 2026 (pre-offering, after giving effect to conversions, RSU settlements, and a 10-for-1 forward split of Class A and Class N shares effected September 17, 2026). It covers named executive officers (NEOs), directors, the group of all directors and executive officers, >5% beneficial owners of any voting class, and selling stockholders. Percentages are calculated based on outstanding shares of the relevant classes (Class A, Class N, Class T) before the offering, with footnotes for options/RSUs exercisable or settling within 60 days, voting proxies, trusts, and the Founder LLC’s Class F share (which carries 50.1% of total voting power on most matters).[3]

Exact full table extraction from the primary SEC HTML document was not feasible via available tools due to its size and structure; secondary sources that reviewed the filing provide consistent details on the named co-founders and key holders (primarily Class A common stock beneficial ownership, including options/RSUs within 60 days and certain proxies). These align with the governance structure described (seven co-founders via Founder LLC holding the single Class F share for 50.1% voting control; Long-Term Benefit Trust holding the Class T share for majority board election rights).[4][5]

Best available secondary sources report the following pre-offering beneficial ownership (Class A common stock as of Sept. 30, 2026; figures include exercisable options/RSUs within 60 days and noted proxies; exact table footnotes and percentages of classes/total shares are in the primary filing):[4]

• Dario Amodei (CEO, director): ~245,259,100 Class A shares (including ~237.8M held directly/trusts + options/RSUs + voting proxies over certain DAF shares). Member of Founder LLC.
• Tom Brown (co-founder): ~244,173,040 Class A shares. Member of Founder LLC.
• Christopher Olah (co-founder): ~242,882,420 Class A shares. Member of Founder LLC.
• Jared Kaplan (co-founder): ~242,351,170 Class A shares. Member of Founder LLC.
• Jack Clark (co-founder): ~242,271,840 Class A shares. Member of Founder LLC.
• Daniela Amodei (President, Chairperson): ~241,408,940 Class A shares (including spouse holdings/options/RSUs). Member of Founder LLC.
• Sam McCandlish (co-founder): ~231,809,780 Class A shares. Member of Founder LLC.
• Krishna Rao (CFO): ~9,497,900 Class A shares (options/RSUs).
• All current directors and executive officers as a group (7 persons): ~496,360,810 Class A shares (plus the Founder LLC’s Class F share).
• Long-Term Benefit Trust (LTBT): Holds the single Class T share (director election rights; limited other voting).
• Founder LLC (seven co-founders): Holds the single Class F share (50.1% aggregate voting power on most stockholder matters; no economic rights; subject to sunset if fewer than three continuing founders remain for 90 days).

Additional >5% holders or selling stockholders (if any beyond the above) and precise percentages of each class and total voting power (pre-offering) are detailed in the primary filing’s table with footnotes. Class N shares (non-voting, held by strategic partners) and other classes are also addressed.[4][5]

Note that the filing blanks the exact number of shares offered and price range (typical for a preliminary S-1); post-offering adjustments assume no underwriter option exercise. The seven co-founders (Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, Christopher Olah) are explicitly named; Benjamin Mann is noted as not part of the Founder LLC.[5]

For the verbatim table, page/section citations, and full footnotes (including exact share breakdowns, option/RSU details, and voting adjustments), consult the official EDGAR filing directly (S-1 document, Principal and Selling Stockholders section). Secondary summaries above are consistent across recent reports but derive from the primary source. No discrepancies were noted in coverage from October 26–27, 2026.

Anthropic’s October 26, 2026 S-1 treats Amazon and Google as large economic owners of non-voting stock, not as governors. The mechanism is deliberate: early money came in as deeply discounted convertible notes and non-voting preferred series that flip into Class N common at the IPO, while a contractual ownership cap, a waiver right, and rollover notes keep either partner from crossing a pre-agreed ceiling unless it chooses to. Control stays with a Founder LLC (50.1% of the ordinary vote via one Class F share) and the Long-Term Benefit Trust (majority of the board via Class T). The filing blanks the exact post-conversion percentages, which is why press stake estimates still disagree with each other and with the caps written into the notes.

How the S-1 describes Amazon’s instruments

Amazon’s position is a stack of old notes that never fully converted, plus later preferred purchases that reduced a new facility rather than drawing it. As of June 30, 2026, every note still outstanding was a non-interest-bearing Amazon rollover note with $6.203 billion of principal—about $2.186 billion under the 2023 note purchase agreement and $4.017 billion under the 2024 agreement—carried on Anthropic’s books at $108.6 billion of fair value. That fair-value mark, not cash interest, is why the notes dominate the income statement.

• The September 2023 Amazon NPA allowed up to $4.0 billion of convertible notes in two tranches. The November 2024 Amazon NPA allowed up to another $4.0 billion in three tranches, of which Anthropic issued about $1.3 billion (November 2024), $1.3 billion (May 2025), and $1.4 billion (November 2025). Second and third 2024 tranches are non-interest-bearing and mature in May 2035 and November 2035 unless earlier converted.
• Pre-split, those notes convert into non-voting redeemable convertible preferred at the lesser of a fixed price—$20.9280 per share for 2023 notes, $50.4779 for 2024 notes—and the price in the most recent preferred or qualified equity financing, subject to Amazon’s ownership cap. After the 10-for-1 split of Class A and Class N effected September 17, 2026, the fixed prices used at the IPO are $2.0928 and $5.0478. Before the cap, $6.2 billion of principal implies about 1.84 billion Class N shares (Finrep calculates 1,840,444,962).
• At the offering, notes convert into non-voting Class N at the lower of the fixed price and the IPO price, but only up to a 12% ownership cap that Amazon may waive. Any balance that would breach the cap stays outstanding as a rollover note, converts at each fiscal-quarter end at the lowest of the fixed price, the IPO price, and the 10-day trailing volume-weighted average price, still subject to the cap, and is repaid in cash at maturity if it never converts. Anthropic may not prepay without the holder’s consent. Maturity on the 2026-form notes is ten years from issuance.
• In May 2026 Amazon put $5.0 billion into Series H redeemable convertible preferred. That purchase cut the amount still available under the April 20, 2026 Amazon facility from $20 billion to $15 billion. As of June 30, 2026, nothing had been issued under that 2026 facility. Draws are tied to compute-capacity milestones and settle in Class N at the 30-day trailing VWAP of Class A, again subject to the 12% cap.
• Preferred that is non-voting (Series B, C-2, D-3, E-2 through E-5, F-2, G-2, G-3, G-4, H-2 and related series) converts into Class N immediately before the IPO; voting preferred series convert into Class A. Class N has the same economics as Class A and no vote except as required by law. It becomes voting Class A only if sold or transferred to an unaffiliated third party.

The economic implication is that Amazon’s conversion price is an order of magnitude below the last private round. Series H was issued at $589.0095 per pre-split share; post-split, that is on the order of $59, against note conversion prices of roughly $2–$5. The S-1 therefore describes a cheap equity option dressed as debt, not a conventional creditor claim. Because Anthropic elected the fair-value option, each rise in its own valuation increases the liability and runs through earnings. The $72.4 billion H1 2026 fair-value charge is mostly that mechanism; operating loss for the half was only $693 million.

How the S-1 describes Google’s instruments

Google’s older notes have already converted into named non-voting preferred series. What remains is preferred already purchased, a large undrawn facility, and an IPO option that can shrink that facility dollar for dollar.

• Under the September 2023 Google NPA, notes of up to $2.0 billion were issuable in three tranches: $500 million (September 2023), $750 million (September 2024), and $750 million (September 2025). In the 2025 Series E-1 financing, the first two tranches converted in full into 12.7 million Series E-3 shares at $44.8692 and 15.3 million Series E-4 shares at $50.4779, both contractual discounts to the Series E-1 original issuance price of $56.0865. The third tranche ($750 million still outstanding at December 31, 2025) converted in January 2026, with the Series G-1 closing, into 3.3 million Series G-4 shares at $233.2228.
• A Google 2024 note that bore 8.5% interest, compounded annually, and matured in August 2030 unless converted, converted in the same Series E-1 financing into 24.2 million Series E-5 shares at $53.2822, again a discount to $56.0865.
• In May 2026 Google invested $10.0 billion in the Series H financing. That purchase reduced the amount available under the March 31, 2026 Google NPA to $20.0 billion. As of June 30, 2026, no non-voting common stock or notes had been issued under the 2026 agreement. The original facility is described in the exhibits as $30 billion, calculated as $40 billion less what Google invested in Series G; the S-1’s $20 billion remaining figure is the post–Series H residual.
• The 2026 agreement runs through December 31, 2030. Notes under it can carry 9% interest. Google’s stake is capped at 14.99% of fully diluted capitalization unless Google elects otherwise. Amounts that would exceed the cap are funded with 0% rollover notes maturing ten years from issuance, which convert into Class N at each quarter-end once Anthropic is public. In a sale of the company other than an IPO, Google can take 125% of principal and accrued interest in cash.
• Google also holds an “IPO Acceleration Option” to buy up to $10 billion of Class N in a private placement concurrent with the offering, at the IPO price, on at least 20 days’ notice. Exercising it reduces the remaining facility by the same amount. After the listing, Google may elect either a convertible note on terms to be mutually agreed or Class N at the 30-day trailing VWAP of Class A. Revenue-linked draws can be deferred to February 6, 2029. The same agreement included a commitment to sell Google 38.6 million Series G-2 shares at $259.1364 (about $10 billion).
• Conversion shares are non-voting on change-of-control matters except where law or the charter requires a vote, and the conversion documents cannot impose non-compete, non-solicit, or non-hire covenants on Google beyond customary confidentiality.

Google’s path is therefore the inverse of Amazon’s residual notes. Most of Google’s 2023–24 principal has already become discounted non-voting preferred (Series E-3, E-4, E-5, G-4, and later G-2 and H). Amazon still holds the cheap unconverted principal that creates the $108.6 billion liability. Both end in the same place at the IPO: Class N, not a board seat.

Board, consent, and investor-rights provisions tied to these holdings

The rights that travel with these holdings are economic and contractual, not governance rights over the company. The S-1 states that after preferred and notes convert, Amazon and Google will each hold greater than 5% of outstanding capital stock, in non-voting Class N. Exact share counts and percentages are blank in this filing, so the document itself does not rank them against other holders.

• Class N has no vote except as required by law. It converts into voting Class A only on an open-market sale or transfer to an unaffiliated third party. That is the structural answer to “do they control Anthropic?”: not through these shares.
• Amazon has neither a board seat nor an observer seat. Reporting on the filing states that directly. Google’s earlier position, stated by Anthropic itself in 2025 antitrust filings, was no voting rights, no board seat, and no board observer, with a stake then capped near 15%. The S-1 does not hand either partner a designation right. The board majority is elected by the Long-Term Benefit Trust through Class T; the remaining directors and ordinary stockholder votes sit with the Founder LLC’s Class F share (50.1%). Spark Capital’s Yasmin Razavi is the venture director named in coverage of the filing, not an Amazon or Google designee.
• What they do have is holder-level consent on their own paper: Anthropic cannot prepay Amazon notes without Amazon’s written consent; ownership caps are waivable only by the holder; Google’s IPO option and the choice between notes and stock on post-IPO draws are Google elections. The 2026 agreements include forms of registration rights. In a non-IPO sale, Google’s 125% cash put is a creditor-style protection, not a veto over ordinary course operations.
• The S-1’s own risk factors treat the overlap as a conflict, not as control: Amazon and Google are simultaneously investors, compute suppliers, marketplace distributors, and competitors (Nova and Gemini). Anthropic has committed to spend at least $110 billion with Amazon through April 2036 and at least $114.5 billion with Google through July 2033, in both cases regardless of actual usage. Those purchase commitments are larger than the equity checks and are the practical leverage, not a preferred-stock consent right.

For anyone trying to read “investor rights” the way a venture round is usually read—board seat, protective provisions on new issuances, drag/tag, information rights that become public-company observer rights—the S-1’s answer is that those governance hooks were designed out. The substitutes are the ownership cap, the cheap conversion price, registration rights, and the multi-year take-or-pay compute contracts.

Cross-check against cumulative investment and estimated stakes

Cash invested and percentage owned are different numbers, and the S-1 is much clearer on instruments than on either investor’s fully diluted share. Public reporting fills the cash side from Amazon’s own filings and from secondary reads of the S-1, and then estimates stakes that the blank capitalization table does not confirm.

Amazon’s Q2 2026 Form 10-Q is the cleanest cash reconciliation: $8 billion of convertible notes bought from the third quarter of 2023 through the fourth quarter of 2025, plus $5 billion of Series G non-voting preferred and $5 billion of Series H non-voting preferred in the second quarter of 2026, totaling $18 billion as of June 30, 2026. The Series H check is what left $15 billion available under a facility capped at $20 billion. Secondary analyses of Amazon’s filings put the June 30, 2026 carrying value at about $190.4 billion, split roughly $97.9 billion of convertible notes and $92.5 billion of non-voting preferred. That holder-side carrying value is not the same as Anthropic’s $108.6 billion issuer-side fair value of the notes alone; the two books use different accounting, and Amazon’s preferred is marked for observable price changes rather than carried at Anthropic’s note fair value.

Google’s cash total is less cleanly stated by Alphabet and has to be rebuilt from the S-1. One reconstruction from the filing’s described instruments is about $13.3 billion before the Series G-2 commitment: $2.0 billion of 2023 notes, about $1.3 billion on the 2024 note, and $10.0 billion of Series H. Adding the contracted 38.6 million Series G-2 shares at $259.1364 (about $10 billion) moves the total toward the “about $23 billion invested” figure used in same-day coverage of the prospectus. Office Chai reads the S-1 as roughly $10 billion of Series G-2 in March 2026 plus $10 billion of Series H in May 2026, on top of the earlier note conversions. Those readings can be reconciled if “invested” sometimes includes converted notes at principal and sometimes includes the G-2 commitment as closed; they cannot all be true as a single cash number without the blank share table.

Stake estimates diverge more sharply, and the divergence is the point of the caps:

• The S-1 will only say each will hold more than 5% after conversion, in Class N. It does not publish the percentage.
• Notes alone, before the cap, are about 10.9% of fully diluted shares as of June on one read of the filing, and conversion is blocked if it would push Amazon’s total ownership above 12% unless Amazon waives.
• PitchBook’s calculation that full conversion would leave Amazon with roughly one-fifth of outstanding shares describes the uncapped math, not the contract. That one-fifth figure and Office Chai’s cited analyst range of 15–21% for Amazon post-IPO both sit above the 12% cap. They are estimates of what Amazon would own if it waived, or of pre-cap dilution, not of what the notes automatically deliver.
• Google is described as holding approximately 7% in PitchBook’s read of the filing, with an option to buy another $10 billion of non-voting stock at the IPO price, against a 14.99% cap it can waive. That 7% is hard to square with the 2025 court-filing figure of about 14% on roughly $3 billion invested, unless the 2026 mega-rounds diluted Google and the later $10 billion checks have not yet been fully reflected in a single percentage. Office Chai’s 10–15% analyst range for Google sits inside the contractual cap; the old 14% figure does not describe the post–Series H cap stack by itself.
• Headline cash figures in the press also slip. NDTV’s “Amazon has invested $13 billion” undercounts the second $5 billion preferred purchase that Amazon’s 10-Q includes in the $18 billion total. Yahoo’s “up to $15 billion” of additional Amazon financing is the remaining facility, not a new $15 billion on top of an undrawn $20 billion.

The practical divergence: the S-1 is a cap-and-instrument document. Press stake ranges that ignore the 12% and 14.99% ceilings, or that treat Amazon’s $190 billion carrying value as if it were a disclosed ownership percentage, are estimating a waiver scenario or a pre-dilution world the filing does not adopt. Amazon has never stated its percentage. Anthropic has not filled in the percentage columns.

Where the S-1 and the press actually part company

Three gaps matter more than the rest.

First, principal versus value. Press and Amazon’s disclosures correctly say Amazon has put in about $18 billion of cash and carries the position near $190 billion. The S-1’s headline liability is $108.6 billion of fair value on $6.2 billion of remaining note principal. Those are not competing estimates of the same thing. Using the carrying value as “what Amazon owns” overstates the legal claim; using principal understates the shares the notes can become. The conversion math—about 1.84 billion Class N shares before the cap, at roughly $2–$5 versus a post-split Series H price near $59—is the number that will dilute public buyers if the cap is waived or filled over subsequent quarters.

Second, the cap versus the “15–21%” narrative. Contractual automatic conversion stops at 12% for Amazon and 14.99% for Google unless the holder waives. Analyst ranges above those caps are scenarios, not descriptions of the notes as written. The S-1 also leaves open that uncapped balances remain outstanding and keep converting quarterly after the listing, so a waiver is not the only way the economic stake can keep growing—it can grow later, still inside the cap, as the share count and price path allow. Any notes left outstanding stay at fair value, so the non-cash charge does not necessarily end on listing day.

Third, governance folklore versus the filing. Older reporting that Google was capped near 15% with no board seat or observer is consistent with the direction of the S-1, and current coverage of the filing says Amazon likewise has no board or observer seat. What the S-1 adds, and some cumulative-investment stories still omit, is that the economic claim is now large enough to be a greater-than-5% holder twice over, while the vote and the board are structurally elsewhere. The investor-rights package tied to these holdings is the cap, the conversion price, registration rights, prepayment consent, and the 125% sale put—not a consent right over Anthropic’s strategy. The consent that actually binds Anthropic’s future cash is the compute commitment: at least $110 billion to Amazon through April 2036 and at least $114.5 billion to Google through July 2033, payable regardless of usage.

Until an S-1/A fills in the price range, the share count, and the pro forma ownership table, any percentage more precise than “greater than 5%, non-voting, and capped at 12% or 14.99% unless waived” is an estimate, not a filing fact. The filing’s own mechanism is already enough to see the trade: Amazon and Google financed a cheap, capped, non-voting claim and a giant take-or-pay cloud contract; the founders and the trust kept the vote and the board.

---

Recent Findings Supplement (October 2026)

Anthropic filed its preliminary S-1 on October 26, 2026, providing the first detailed public breakdown of Amazon’s and Alphabet/Google’s complex, multi-year holdings via convertible notes, preferred stock series, and related facilities.[1][2]

This filing (and contemporaneous Amazon 10-Q disclosures) updates prior public estimates with precise figures as of June 30, 2026, conversion mechanics, ownership caps, and undrawn commitments, while highlighting non-voting structures and cloud spending obligations that tie the relationships together.[3][4]

Amazon’s Holdings: Notes, Conversions, and Caps

Amazon’s position combines early convertible notes (starting 2023) with later preferred stock purchases. The S-1 shows ~$8 billion in notes issued through 2025, with $6.2 billion principal still outstanding as of June 30, 2026. These are slated to convert at the IPO into nonvoting shares (potentially ~1.84 billion shares or ~10.9% fully diluted as of June, subject to adjustment), subject to a 12% total ownership cap that Amazon can waive. Any excess would remain as notes convertible post-IPO.[1][3]

In Q2 2026, Amazon converted portions into two $5 billion investments in nonvoting preferred stock (Series G and H), bringing cumulative cash invested to ~$18 billion as of June 30, 2026—consistent with Amazon’s own 10-Q reporting.[2][5]

Amazon’s 10-Q further shows carrying values of ~$190.4 billion ($92.5 billion preferred + $97.9 billion notes) and significant unrealized gains (e.g., a $16.8 billion non-cash gain on the preferred conversion). An additional Amazon 2026 NPA provides up to $20 billion more (reduced to $15 billion available after the Series H purchase), drawable in $5 billion tranches tied to compute capacity milestones.[5][6]

Implication for competitors or entrants: These structures lock in massive non-controlling economic exposure plus compute commitments ($110 billion minimum AWS spend through 2036), creating deep interdependence that is hard to replicate without similar scale and strategic alignment.

Google/Alphabet’s Holdings: Preferred Purchases and Option Structure

Google’s involvement includes earlier note conversions into preferred stock and newer direct purchases. The S-1 details a March 2026 ~$10 billion Series G-2 preferred investment and a May 2026 $10 billion Series H preferred investment.[1][4]

A March 2026 Google 2026 NPA allows up to $30 billion in convertible notes or nonvoting common stock (reduced to $20 billion available after the Series H purchase; none drawn as of June 30). Google also holds an option to buy up to $10 billion in Class N common stock via private placement at the IPO price concurrent with the offering, which would further reduce NPA availability.[2][3]

Ownership is capped at 14.99% (waivable), with earlier reports noting a prior ~14% stake built quietly and capped near 15% without voting rights or board seats. Cumulative Google investment is reported around $23 billion in recent analyses cross-referenced to the filing.[7][8]

Implication: The IPO-timed option and rolling NPA give Google flexible additional exposure at public valuation while preserving the non-voting, capped structure—useful for strategic cloud partners seeking influence without control.

Conversion Terms, Share Classes, and Rights Provisions

Notes and facilities generally convert into nonvoting Class N common or preferred stock with economic rights but no voting power. Amazon’s notes have a 12% cap; Google’s latest agreement has a 14.99% cap (both waivable). Earlier Google notes converted into specific Series E-3/E-4 nonvoting preferred at defined prices (~$44.87 and $50.48 per share).[9][10]

Neither investor receives board seats, observer rights, or voting rights. The S-1 flags these intertwined investor-customer-competitor relationships as risk factors. Anthropic uses multiple share classes (including founder-controlled Class F with majority voting power and Class N for these strategic holders) to maintain founder control while accommodating large economic stakes.[4][11]

Investor rights appear limited to standard protective provisions in the note purchase agreements (e.g., maturity, conversion mechanics, sale-of-company options like 125% cash for Google in certain scenarios), with no broad consent/veto rights highlighted for day-to-day governance.[10]

Implication: This setup allows Big Tech to deploy capital and secure compute deals without triggering governance or antitrust concerns, a model that future AI startups may need to emulate or navigate when seeking similar partnerships.

Cross-Checks with Public Reporting and Estimated Stakes

Amazon’s 10-Q and S-1 figures align closely on the $18 billion cumulative (notes + preferred). Google’s ~$20–23 billion range matches S-1-described purchases plus prior activity. Analyst estimates around the filing project post-IPO stakes of ~15–21% for Amazon and ~10–15% for Google (at a hypothetical $2 trillion valuation, implying enormous paper values).[1][5]

The S-1 reconciles total capital raised (~$141 billion) to include the $11.3 billion in Amazon/Google convertible note proceeds (with ~$5 billion converted and $6.2 billion remaining principal carried at high fair value on the balance sheet). Minor differences appear in exact series labeling or timing across reports, but core economics are consistent.[12]

Cloud commitments add another layer: ≥$110 billion to Amazon (to 2036), ≥$114.5 billion to Google (to 2033), plus Microsoft exposure—totaling hundreds of billions when combined with marketplace revenue (48% of H1 2026 revenue via the three clouds).[5][8]

Implication: Press estimates of stakes and values are directionally accurate but the S-1 supplies granular mechanics (caps, conversion quantities, undrawn facilities) that were previously opaque, reducing uncertainty for IPO pricing while underscoring concentration risks.

Recent Context and What Has Changed

Prior to the October 26, 2026 S-1, details were pieced together from Amazon/Google quarterly filings, leaks, and partial announcements (e.g., 2025–early 2026 note issuances and preferred rounds). The filing is the primary new data point, confirming and expanding on Q2 2026 activity, adding the IPO-timed Google option and precise note conversion math, and quantifying cloud obligations.[2][13]

No major policy or regulatory shifts specific to these holdings are noted in the immediate post-filing coverage; the emphasis is on disclosure transparency ahead of the planned Nasdaq listing (ticker ANT).[11]

For anyone evaluating or competing in this space, the S-1 illustrates how AI labs can raise enormous sums from hyperscalers while using share class engineering and caps to limit influence—creating both moats (via committed spend and capital) and vulnerabilities (dependency and valuation sensitivity).

Anthropic’s October 26, 2026 S-1 separates who owns the economics from who controls the company. Seven co-founders hold nearly identical Class A stakes of roughly 232–244 million shares each, but the shareholder vote sits in a single non-economic Class F share, and the board majority sits with a trust that owns no equity.

Economics and control are different instruments

Public buyers will get Class A (ticker ANT on Nasdaq Global Select Market): one vote per share and pro rata economics. That vote is structurally capped. Immediately before the offering, Anthropic will issue one share of Class F to Anthropic Founders, LLC. That share is entitled to 50.1% of the voting power on matters submitted to stockholders generally, including election of the directors the trust does not pick, and it has no economic rights. All Class A holders together are limited to 49.9%.

The Long-Term Benefit Trust (LTBT) holds the sole share of Class T. Class T generally elects and removes the Class T directors, who will be a majority of the authorized board, and otherwise does not vote on ordinary stockholder matters and has no economic rights. Strategic partners, including Amazon and Google, hold non-voting Class N, which converts to Class A only on a transfer to an unaffiliated third party. Employees other than the co-founders receive Class E, which votes only on a narrow set of governance tie-breaks and has no economic rights.

Because the LTBT can elect a board majority, Anthropic says it will be a Nasdaq “controlled company,” and it expects to use that exemption only for the nominating and corporate governance committee.

Founder and executive holdings as of September 30, 2026

Beneficial ownership is as of September 30, 2026, after the September 17, 2026 10-for-1 split of Class A and Class N, and includes options exercisable and RSUs expected to settle within 60 days. Percentage columns in the S-1 are blank, so the company does not state ownership percentages. Share counts below are from the principal stockholders footnotes and table as extracted from the filing.

Holder	Class A beneficially owned	What is inside the number
Dario Amodei, CEO and director	245,259,100	237,813,080 shares held or controlled via trusts; 2,185,400 options; 639,500 RSUs; plus 4,621,120 shares he votes under irrevocable proxies for donor-advised funds but does not own
Tom Brown	244,173,040	Includes the same 2,185,400 options and 639,500 RSUs pattern used for the co-founders
Christopher Olah	242,882,420	Same near-even co-founder package
Jared Kaplan	242,351,170	Same
Jack Clark	242,271,840	Same
Daniela Amodei, President and Chair	241,408,940	238,319,540 shares held or controlled; her own 2,185,400 options and 639,500 RSUs; plus spouse options of 130,550 and spouse RSUs of 133,950
Sam McCandlish	231,809,780	Smallest of the seven co-founder stakes
Krishna Rao, CFO	9,497,900	Only 153,960 shares held of record; 7,843,940 options in his name; 1,500,000 options held by a trust he controls
All current directors and executive officers as a group (7 persons)	496,360,810	481,077,470 shares plus 13,845,290 options and 1,438,050 RSUs within 60 days

The seven co-founders in the Founder LLC are Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, and Christopher Olah. Benjamin Mann, often described as a co-founder, is not a member and has no say over the Class F share. Each continuing founder gets one vote inside the LLC on how Class F is cast. Their personal Class A shares are voted separately, at their own discretion. Each has pledged to give 80% of personal Anthropic equity to philanthropy; the S-1 says those pledges are voluntary and revocable, and that donations shrink economic exposure while Class F voting power stays put so long as the minimum retained-ownership condition for LLC membership is met.

Only three people are listed as executive officers: Dario Amodei, Daniela Amodei, and Krishna Rao. The other co-founders have senior titles (including chief compute officer and chief science officer) but are outside the named-executive compensation table. For 2025, Dario’s total compensation was $17,958,298 and Daniela’s was $16,408,998, each on a $654,167 salary, with most of the rest in stock and option awards. Rao’s 2025 total was $720,250, almost all salary; his wealth is in the option package above, not in 2025 equity grants.

Amounts each founder will sell in the IPO are blank.

Directors, the officer-and-director total, and employees

The current board has six directors. After the offering, seven seats are authorized:

• Class T (LTBT), four seats: Reed Hastings, Vas Narasimhan, Christopher Liddell (the “Specified Class T Director”), and one vacancy.
• Class A and Class F voting together, three seats: Dario Amodei, Daniela Amodei, and Yasmin Razavi (Spark Capital).

Today, the LTBT’s designees are Daniela Amodei, Hastings, and Narasimhan. Dario was elected by common stockholders, Liddell as a common-stock designee, and Razavi by senior preferred holders. Those election mechanics are replaced at the IPO; the six sitting directors continue until the next annual meeting.

The 496,360,810-share officer-and-director group is not the seven co-founders. It is the people who are directors or executive officers. That group is dominated by the two Amodeis plus Rao’s options. Non-employee directors show up mainly as small RSU positions (footnotes list 9,410 RSUs for at least two of them). Hastings, Liddell, Narasimhan, and Razavi are treated as independent under Nasdaq standards.

Employees other than the co-founders do not get a disclosed aggregate economic stake in the principal-stockholders table. What they do get, at the offering, is Class E: one vote per share, but only on specified matters (the Specified Class T Director if the nominating committee does not concur with the LTBT, certain LTBT amendments, and redemption of Class T), and no economic rights beyond par on liquidation. Ordinary employee wealth is in Class A options and RSUs, not in Class E. The company also contributed 846,840,960 Class N shares — 5% of fully diluted shares as of June 30, 2026, assuming full conversion of the Amazon notes — to its wholly owned Anthropic Public Benefit Fund. That is a company issuance, not employee or founder ownership.

How the Long-Term Benefit Trust works

The LTBT was created in 2023. It is a Delaware common-law noncharitable purpose trust, governed by an amended and restated trust agreement filed as an exhibit. It is meant to be an independent body of five financially disinterested trustees with expertise in AI safety, national security, public policy, and social enterprise. Anthropic consolidates it as a variable-interest entity because the company funds it, directs the activities that most affect its economics, and would receive its assets other than the Class T share if the trust ends. LTBT assets and liabilities on the balance sheet were not material.

What Class T can do. As sole holder, the LTBT generally elects and removes the Class T directors, a majority of the authorized board. It also votes on redemption of Class E and on certain amendments to Class F rights. It does not vote on election of the other directors, and it does not vote on matters submitted to stockholders generally except as required by law. The S-1 is explicit that neither the trust nor the trustees run the company day to day or set strategy. Their job is to pick directors who can balance three duties: stockholders’ pecuniary interests, people materially affected by the company’s conduct, and the company purpose — “the responsible development and maintenance of advanced AI for the long-term benefit of humanity.” A merger or sale of substantially all assets still needs board approval under Delaware law, and a majority of that board is LTBT-elected.

The check on one seat. Before nominating or removing a director, the nominating party must give the nominating and corporate governance committee (NGC) at least 30 days’ notice, waivable by the NGC. After the IPO the NGC is one Class T director and two non-Class T directors, acting by majority. For one seat only — the Specified Class T Director, initially Liddell — nomination, removal without cause, or redesignation needs NGC concurrence. If the NGC does not concur within 15 days, the question goes to a vote of Class E unless the LTBT withdraws. Approval then requires a majority of Class E votes cast, with a quorum of a majority of outstanding Class E. Any Class T director, including that specified seat, can be removed for cause without NGC or Class E approval. The LTBT and the Founder LLC have agreed to give each other and the NGC 30 days’ notice before a for-cause removal.

How trustees are appointed. There are five authorized trustee seats; one is vacant. Sitting trustees appoint successors by majority vote of trustees then in office. Removal is by majority of all authorized trustees, or by unanimous approval of the other trustees then in office plus NGC approval. Appointment or removal requires at least 30 days’ prior written notice to the NGC, waivable by the NGC. The Delaware Court of Chancery can also remove a trustee for disqualifying conduct. Terms are annual, with no term limit. The S-1 says that in practice the trustees and company executives have collaborated closely on sourcing, interviewing, and selecting trustees.

Current trustees, per the S-1:

• Dr. Neil “Buddy” Shah, Chair since June 2023 (global health, impact measurement, social enterprise; CEO of the Clinton Health Access Initiative).
• Richard Fontaine since April 2025 (foreign policy and national security).
• Dr. Ben S. Bernanke since June 2026 (macroeconomics, financial systems, institutional governance).
• Dr. Christopher Stubbs since October 2026 (experimental physics and AI applied to science and education).
• Fifth seat vacant.

Registration materials state that each trustee receives $400,000 a year in cash, and the chair an additional $25,000. Pay is cash only, may not be conditioned on company profits, and trustees may not directly buy or hold Anthropic securities.

Class T does not sunset on a clock. It can be amended or redeemed for nothing only if a “Trust Amendment Threshold” is met: if trustees favor the change, 80% of the authorized board and 80% of trustees in office; if they oppose it and Class F is still out, a combination of Class F (unanimous continuing founders in the first three years after the IPO, or 66⅔% thereafter), a Class E supermajority or majority depending on timing, and a majority of Class A votes cast; if Class F is gone and trustees oppose, 85% of outstanding Class A.

What the S-1 says about stake versus rights

The filing’s own summary is the point of the structure. Class T “does not otherwise have voting rights with respect to matters submitted to stockholders generally and has no economic rights.” The trust “generally has the authority to elect and remove the Class T directors, who together will constitute a majority of the authorized members of our board of directors.” It has “significant influence” over the board and none over day-to-day management or strategy. The arrangements are “designed to minimize the LTBT’s and the Trustees’ exposure to our financial performance.”

Put next to Class F, the split is complete. Founders keep 50.1% of the general stockholder vote through a share with no economics, even as they sell or donate Class A. The trust keeps a board majority through a share with no economics, even though it is paid in fixed cash and barred from owning the stock. Together, the S-1 says, the LTBT and the Founder LLC can elect every director, and the Founder LLC can determine the outcome of matters submitted to stockholders generally “for the foreseeable future.” Class F voting power is a fixed fraction, so new Class A issuance does not dilute it. Public Class A buys the residual economics and at most 49.9% of the general vote, and none of the seats the trust fills.

The risk factors say the obvious implication: the LTBT may have interests that differ from Class A holders, PBC balancing can favor purpose or affected stakeholders over the stock price, and founder donations can widen the gap between who has money at risk and who has the vote.

---

Recent Findings Supplement (October 2026)

Anthropic (PBC) filed its S-1 with the SEC on October 26, 2026, providing the first detailed public disclosure of its multi-class share structure, beneficial ownership table (as of September 30, 2026, post a 10-for-1 forward split on Class A and Class N shares), and Long-Term Benefit Trust (LTBT) mechanics.[1][2]

This filing (and contemporaneous analyses) represents the most recent and comprehensive data, superseding prior private-round or partial disclosures. No material updates to these figures or structures have appeared in the ~24 hours since filing.[3]

Founder, Executive, Director, and Employee Equity Holdings

The seven co-founders (Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, and Christopher Olah) each hold roughly 232–244 million Class A shares as of September 30, 2026. These stakes include direct holdings, shares in trusts, and (for Dario Amodei) voting proxies over ~4.62 million shares held by donor-advised funds. Each also holds exactly 2,185,400 currently exercisable stock options and 639,500 RSUs scheduled to vest post-IPO (in quarterly installments, subject to continued service).[3][2]

• Tom Brown holds the largest stake at 244,173,040 Class A shares; others range down to Sam McCandlish at 231,809,780. At a hypothetical $1.5–2 trillion IPO valuation, these would be worth ~$21–30 billion each (pre-donation pledges).[4]
• CFO Krishna Rao holds 9,497,900 Class A shares.[2]
• All executive officers and directors as a group (7 persons) beneficially own 496,360,810 Class A shares.[2]
• Non-founder employees receive Class E shares (limited voting rights, no economics mentioned for this class in summaries).[5]
• Co-founders have pledged to donate ~80% of their equity (voluntary and revocable per the filing), but this does not affect their Class F voting rights.[6]

Implication for competitors/entrants: This concentrated founder ownership (plus options/RSUs) creates strong alignment on long-term control but highlights the separation of economics from governance. Public investors in Class A will own economic upside without proportional voting power.

Long-Term Benefit Trust (LTBT) Structure and Class T Stock

The LTBT, implemented in 2023 as a Delaware noncharitable purpose trust governed by a Trust Agreement, holds the sole share of Class T common stock. Class T carries no economic rights (no dividends, no liquidation proceeds, no conversion) and generally does not vote on matters submitted to stockholders except as required by law or specific protective provisions.[1][1]

• Its primary power is the authority to elect and remove “Class T directors,” who together constitute a majority of the authorized board seats (phasing in over time based on milestones; a majority within four years of the Series C close).[1]
• Class T also votes on redemption of Class E stock and certain amendments to Class F rights.[1]
• The Trust has no day-to-day management or strategy authority.[1]

Implication: This creates a governance overlay focused on Anthropic’s public-benefit purpose (“responsible development and maintenance of advanced AI for the long-term benefit of humanity”) alongside stockholder and stakeholder interests, without economic skin in the game.[1]

Trustee Appointment, Composition, and Compensation

The LTBT has five authorized trustee seats; one is currently vacant. Trustees are financially disinterested (cannot directly buy or hold Anthropic securities) and serve one-year terms, with new trustees selected by a vote of existing trustees (in consultation with the company per older descriptions).[1][7]

Current trustees (per the S-1):
- Dr. Neil “Buddy” Shah, Chair (since June 2023) — global health, impact measurement, social enterprise.
- Richard Fontaine (since April 2025) — foreign policy and national security.
- Dr. Ben S. Bernanke (since June 2026) — macroeconomics, financial systems, institutional governance.
- Dr. Christopher Stubbs (since October 2026) — experimental physics and AI applications in science/education.
- Fifth seat: vacant.[1]

Each trustee receives a fixed $400,000 annual cash retainer; the Chair receives an additional $25,000. Compensation is not tied to company performance or actions.[1]

Recent development: Stubbs’s appointment (announced ~October 21, 2026) fills one of two prior vacancies and brings scientific/AI-application expertise.[8]

Implication: The fixed-pay, no-equity model insulates trustees from financial incentives, aiming for independence, but the self-appointment mechanism and opacity of the full Trust Agreement have drawn scrutiny in pre-IPO coverage.[9]

Economic Stake vs. Governance Rights (S-1 Summary)

The S-1 explicitly contrasts the Trust’s zero economic stake (Class T has “no economic rights”) with its substantial governance rights over board composition.[1]

• The Trust elects a board majority but has no role in daily operations or most stockholder votes.
• Complementary structure: A Founder LLC (owned by the seven co-founders) holds one Class F share carrying 50.1% of total voting power on general stockholder matters (including election of the non-Class T directors). Class F also has no economic rights.[5][6]
• Post-IPO board (7 authorized directors): 4 Class T seats (Reed Hastings, Vas Narasimhan, Christopher Liddell, and one vacancy) elected by the LTBT; 3 seats (Dario Amodei, Daniela Amodei, Yasmin Razavi) elected by Class A and Class F voting together.[1]
• Class A (public shares) carries one vote per share but is capped at no more than 49.9% of total voting power collectively.[10]
• Sunset: Class F voting power ends if fewer than three of the seven founders remain in the LLC for 90 consecutive days.[6]

The filing notes this structure has “no direct public company precedent” and may limit unsolicited takeovers or proxy contests.[11]

Implication for market participants: Investors gain economic exposure via Class A but face structural limits on influence, with control split between the mission-oriented LTBT (board majority) and founder voting bloc (other matters). This setup prioritizes long-term AI-safety governance over traditional shareholder primacy.[12]

No other post-April 2026 developments (e.g., updated ownership tables, trustee changes beyond Stubbs, or regulatory actions on the structure) appear in available sources. All quantitative details above derive directly from the October 26, 2026 S-1 and contemporaneous reporting on it.

Anthropic’s S-1, filed October 26, 2026, names almost no pure venture firms as large holders. The mechanism is deliberate dilution plus non-voting strategic stock: no traditional VC is reported above about 2%, so SEC’s 5% beneficial-ownership table never lists Menlo, Lightspeed, ICONIQ, Fidelity, or GIC individually. Spark appears dozens of times only because partner Yasmin Razavi sits on the board after leading Series C—not because Spark crossed 5%. Amazon and Google are the outside names the filing treats as greater-than-5% economic owners, and both hold non-voting Class N.

How the S-1 treats investors

Item 403 requires a principal-stockholders table for directors, officers, and anyone known to beneficially own more than 5% of a class of voting securities. Anthropic’s table is built around the seven co-founders (each with on the order of 230–245 million Class A shares), the Long-Term Benefit Trust’s Class T share, and the Founder LLC’s single Class F share that carries a fixed 50.1% of the vote. Spark is footnoted because Razavi is a director: the filing says she disclaims beneficial ownership of Spark entities, and it excludes Spark affiliates from her personal line. Business Insider reports the filing states Spark invested $777 million across six financings since 2023, including $75 million in the first round at $11.22 per share, and that Spark is the only traditional VC firm named.

After preferred stock and notes convert, Amazon and Google will each hold more than 5% of outstanding capital stock, in non-voting Class N. Amazon’s stake is capped at 12% (waivable); Google’s latest financing caps its stake at 14.99% (also waivable). Neither has a board seat or observer right. That is why the cloud partners show up in related-party and capitalization disclosure, while the growth funds that led the lettered rounds do not.

Newcomer, citing investor documents and people familiar with the rounds, reported on October 22, 2026 that no venture firm owns more than 2%. Bloomberg-sourced figures at the May 2026 $965 billion mark put ICONIQ near $17 billion after more than $6 billion invested, Menlo near $14 billion after more than $1 billion, GIC and Sequoia each around 1%, and Lightspeed under 2%. Those stakes are large in dollars and invisible as named 5% holders.

Round table (Series A through Series H)

Company announcements are the primary source for amount, date, and lead. Post-money valuation is company-disclosed from Series E onward. Earlier marks were not in the press releases; secondary databases and contemporaneous reporting disagree, and that disagreement is shown rather than smoothed.

Round	Date	Amount	Lead / co-leads	Publicly reported valuation	Named participants (selected)
Series A	May 28, 2021	$124 million	Jaan Tallinn	Not disclosed by the company. Secondary marks cluster around $550–623 million post-money (StockAnalysis $550M; Forge/Yahoo $623 million). One later analysis cites ~$845 million.	James McClave, Dustin Moskovitz, Center for Emerging Risk Research, Eric Schmidt
Series B	April 29, 2022	$580 million	Sam Bankman-Fried (FTX)	Not disclosed by the company. Later databases put it near $4.0–4.1 billion. Court records cited by Business Insider put Bankman-Fried’s ~$500 million at 13.56% at the time.	Caroline Ellison, Nishad Singh, Jim McClave, Jaan Tallinn, Center for Emerging Risk Research
Series C	May 23, 2023	$450 million	Spark Capital	Company did not disclose. Business Insider, citing the context of the filing, uses $4.5 billion. Reuters sources at the time said nearly $5 billion. Newcomer: about $4.6 billion before the new money. Yahoo/Forge series data: $4.55 billion post at $11.23/share.	Google, Salesforce Ventures, Sound Ventures, Zoom Ventures, Menlo Ventures (small check, ~$10 million per Newcomer), Factorial Funds
Series D	In market Dec 2023; closed early 2024 (press through Jan–March 2024)	$750 million	Menlo Ventures (Matt Murphy); more than $500 million via an SPV	Widely associated with $18.4 billion. Reuters on Dec 21, 2023 described $18.4 billion as the valuation before the round. Later Menlo and press accounts treat $18.4 billion as the round valuation (a roughly 4x step-up from Series C).	Lightspeed (~$100 million, per Newcomer), Bessemer introductions; Amazon and Google convertible commitments were being formalized around the same period and are not the equity lead
Series E	March 3, 2025	$3.5 billion	Lightspeed Venture Partners (~$1 billion, partly with top LP co-invest)	$61.5 billion post-money (company)	Bessemer, Cisco Investments, D1 Capital Partners, Fidelity Management & Research, General Catalyst, Jane Street, Menlo Ventures, Salesforce Ventures
Series F	September 2, 2025	$13 billion	ICONIQ; co-led by Fidelity Management & Research and Lightspeed	$183 billion post-money (company)	Altimeter, Baillie Gifford, BlackRock affiliates, Blackstone, Coatue, D1, General Atlantic, General Catalyst, GIC, Goldman Sachs Alternatives, Insight Partners, Jane Street, Ontario Teachers’ Pension Plan, Qatar Investment Authority, TPG, T. Rowe Price, WCM, XN
Series G	February 12, 2026	$30 billion	GIC and Coatue; co-led by D.E. Shaw Ventures, Dragoneer, Founders Fund, ICONIQ, and MGX	$380 billion post-money (company)	Accel, Addition, Altimeter, Baillie Gifford, Bessemer, BlackRock, Blackstone, D1, Fidelity, General Catalyst, Greenoaks, Goldman Sachs Alternatives, Insight, Jane Street, JPMorganChase, Lightspeed, Menlo, Morgan Stanley Investment Management, QIA, Sands Capital, Sequoia, Temasek, TPG, XN; a portion of previously announced Microsoft and Nvidia commitments
Series H	May 28, 2026	$65 billion	Altimeter, Dragoneer, Greenoaks, and Sequoia; co-led by Capital Group, Coatue, D1, GIC, ICONIQ, and XN	$965 billion post-money (company)	Includes $15 billion of previously committed hyperscaler money, of which $5 billion from Amazon. Also AMP PBC, Baillie Gifford, Blackstone, Brookfield, D.E. Shaw, DST, Fidelity, General Catalyst, Insight, Jane Street, Lightspeed, MGX, Temasek, T. Rowe Price, and others

Strategic checks sit beside the lettered rounds and dominate the economic cap table: Amazon’s cumulative cash in is reported in the teens of billions with a 12% cap and $6.2 billion of notes still outstanding as of mid-2026; Google’s cumulative investment is reported around the low-to-mid twenties of billions with a 14.99% cap and an option, disclosed in the S-1, to buy up to $10 billion of Class N at the IPO price. Those are not “Series” leads, but they are why Amazon and Google clear 5% and the VCs do not.

Press names that are not 5% holders

Flag these as disclosed in round announcements or later stake reporting, and not individually listed as greater-than-5% voting holders in the S-1 principal-stockholders table:

• Named in the filing, but not as a 5% holder: Spark Capital. Presence is a related-person/director fact (Razavi). Cumulative investment of $777 million is well below 5% after dilution.
• Round leads or co-leads that stay under the line: Menlo Ventures (Series D lead; press stake under 2%, marked near $14 billion at $965 billion), Lightspeed (Series E lead and Series F co-lead; under 2%), ICONIQ (Series F lead; first check in 2025 at the $183 billion round; press mark near $17 billion after more than $6 billion in), Fidelity Management & Research (Series F co-lead; also in E, G, H), GIC (Series G lead; about 1% after more than $4.5 billion, per Bloomberg-sourced reporting).
• Later co-leads also under 5%: Sequoia (Series H co-lead; about 1% after roughly $3 billion, starting in 2026), Coatue (Series G co-lead and later follow-on; one person put total investment near $4.75 billion), Altimeter, Dragoneer, Greenoaks, D1 Capital, Capital Group, Founders Fund, MGX, XN.
• Early or repeat participants that never approached 5% as firms: Sound Ventures, Salesforce Ventures, Zoom Ventures, Bessemer, Jane Street, General Catalyst, Insight Partners, Blackstone, BlackRock, T. Rowe Price, Baillie Gifford, General Atlantic, Goldman Sachs Alternatives, Ontario Teachers’, QIA, Temasek.
• Angels and the FTX complex: Jaan Tallinn, Eric Schmidt, Dustin Moskovitz, and James McClave are Series A/B names. The S-1 does not list them as 5% voting holders in the excerpts and secondary coverage available. Bankman-Fried’s Series B stake was sold by the FTX estate in 2024 for about $1.3 billion. The S-1 says the company bought 4,453,900 Series B shares from the U.S. government in October 2025 that had been forfeited by former holders in proceedings Anthropic was not a party to—consistent with the Caroline Ellison and Nishad Singh shares the government sold, which Business Insider reports fetched $627 million.

What this means if you are matching a cap table to the prospectus

Match lettered-round leads to the table above, then do not expect those names in the 5% column. The filing’s economic greater-than-5% outside holders are Amazon and Google, on non-voting stock, with contractual caps. The only venture name that recurs in the S-1 is Spark, and the recurrence is governance (board seat, related-person footnote, six-financing investment total), not a 5% line. Dollar stakes in the tens of billions at a $965 billion mark can still be under 2%, which is why ICONIQ, Menlo, GIC, Sequoia, and Lightspeed dominate the pre-IPO press and disappear from the individual 5% list. Early-round valuations before Series E should be treated as estimates: the company did not publish them, and secondary sources do not agree.

---

Recent Findings Supplement (October 2026)

Anthropic PBC (ticker ANT on Nasdaq) publicly filed its S-1 on October 26, 2026, marking the key recent development.[1][2]

This follows a confidential filing around June 2026 and provides the first detailed public look at finances, governance, and ownership ahead of an expected roadshow (~November 9) and potential listing in mid-to-late November 2026. The filing leaves price/share counts blank but supports estimates of a $1.5T–$2T IPO valuation (up from the $965B private valuation in its May/June 2026 Series H round).[3][4]

In the S-1, Spark Capital is the only pure-play venture firm individually named and detailed as a principal stockholder. Its GP Yasmin Razavi (board director since the 2023 Series C) triggers extensive disclosure (>50 mentions) due to the board seat and multiple investments totaling ~$777M across six rounds since 2023 (including a $75M check in the pivotal early round at $11.22/share and leading the $450M Series C at ~$4.5B valuation in May 2023).[5][6]

No other traditional VCs (e.g., Menlo, Lightspeed, ICONIQ) appear individually in the principal/selling stockholders table or with comparable detail. Amazon and Google each hold >5% in non-voting Class N shares (strategic, convertible notes/investments).[7][8]

Governance features a novel multi-class structure (inspired by precedents like Zuckerberg/Musk setups) that limits public shareholder influence: A Founder LLC (seven co-founders: Dario/Daniela Amodei et al.) holds one Class F share with 50.1% voting power on most matters (dilution-protected, no economic rights). A Long-Term Benefit Trust (disinterested trustees) elects a board majority via Class T. Public Class A shares (Nasdaq) are capped at 49.9% collective vote. Class E (non-founder employees) has limited rights; 5% of shares go to a Public Benefit Fund.[9][10]

Founders each beneficially own ~240M+ Class A shares (pre-IPO stakes worth ~$13.7B–$14.4B at Series H pricing of $58.90/share).[11]

Press reports (Bloomberg, PitchBook, etc., October 2026) detail larger stakes and gains for other investors based on the $965B private valuation, but these are not individually listed in the S-1 (consistent with sub-5% holdings or aggregation):[3][4]

• ICONIQ: >$6B invested (first in 2025 at $183B valuation); ~$17B stake at $965B val (potentially ~$35B at $2T IPO).
• Menlo Ventures: >$1B total (incl. $500M to lead Series D, its largest check); ~$14B position (<2% stake).
• Lightspeed Venture Partners: Participated in Series D and follow-ons; <2% stake.
• GIC: >$4.5B since 2025; ~1% stake.
• Sequoia: $3B invested (starting ~early 2026); ~1% stake (~$9.7B at current val).
• Others (e.g., Sound Ventures, Factorial, Bessemer) noted in cap table analyses but not S-1 principals.

These “press-only” investors reflect late or large follow-on checks in a capital-intensive AI environment; early/smaller checks (like Spark’s) or board seats drive S-1 visibility. Amazon/Google’s >5% strategic stakes are disclosed (non-voting Class N).[12]

Recent funding context (post-April 2026, per S-1/press): Series H (May–June 2026) raised ~$65.2B at pricing leading to the $965B valuation (77M+ Series H-1 shares to 305 investors + non-voting H-2 to 11 at ~$589/share). Earlier rounds (e.g., 2023 Series C led by Spark) are referenced but not exhaustively tabled in new sources. Total capital raised: ~$141B, with ~$112B+ cash/short-term investments as of mid-2026.[7]

Implications for competitors/entrants: The S-1 underscores how board seats and early institutional involvement (even at “high” valuations like 2023’s $4.5B–$5B) create outsized disclosure and returns in AI, while massive later-stage checks from growth/strategic players (ICONIQ, GIC, Sequoia) drive valuation without proportional S-1 naming if below thresholds. Governance innovations (founder control + trust) may set precedents for mission-driven AI/public-benefit companies seeking public capital without ceding control. Revenue trajectory ($16.3B H1 2026, up >1,300% YoY) and infrastructure costs highlight execution risks/opportunities for those building complementary AI infrastructure or enterprise tools.[13]

Summary Table of Key Investors (S-1 vs. Press, Recent Focus):

• Spark Capital: S-1 named (board via Razavi; $777M total); press highlights ~$7B stake at $965B val (93x+ on early $75M).
• ICONIQ, Menlo, Lightspeed, GIC, Sequoia: Not individually in S-1 principals; press reports large stakes/gains (e.g., ICONIQ ~$17B, Menlo ~$14B) from post-2023/2025 rounds.
• Amazon/Google: S-1 >5% (non-voting Class N); strategic notes/convertibles.
• Founders (via LLC): S-1 detailed control/ownership (~10% economic, 50.1% vote).

No comprehensive new full historical round-by-round table (Series A onward) emerged in post-April 2026 sources beyond Spark’s role and the Series H details; prior rounds are referenced anecdotally in cap table stories. Data aligns with the query’s focus on matching S-1 entries and flagging press-only investors.

Anthropic’s October 26, 2026 S-1 does not use a conventional dual-class structure. It splits control into two non-economic instruments—one share that locks 50.1% of ordinary stockholder votes with seven co-founders, and one share that lets an independent trust elect a board majority—while public buyers and the largest strategic investors hold the economics. The result is a company that can raise public capital, and take Amazon and Google money, without either group being able to outvote the founders or replace the trust-selected directors.

What the filing actually creates

Anthropic, PBC filed Form S-1 on October 26, 2026 (SEC file 333-299633, accession 0001628280-26-067168) to list Class A common stock on the Nasdaq Global Select Market under the symbol ANT. Price and share count are still blank. The company is a Delaware public benefit corporation whose certificate requires directors to balance stockholders’ pecuniary interests, people materially affected by the company’s conduct, and a stated purpose: “the responsible development and maintenance of advanced AI for the long-term benefit of humanity.”

Immediately before the offering closes, outstanding preferred stock converts under the certificate (the “Preferred Stock Conversion”). Voting preferred becomes Class A; non-voting preferred, the series used for strategic partners, becomes Class N. A single Class F share is then issued to Founder LLC. Class E is authorized for later issuance to employees other than the co-founders and is not outstanding at the IPO. On September 17, 2026 the company effected a 10-for-1 forward split of Class A and Class N; prospectus figures are adjusted for that split.

The five common classes that will exist after the IPO, taken from the S-1’s own “Our Capital Structure” table:

Class	Who holds it	Votes	What those votes cover	Economics	Conversion / exit
A	IPO buyers, co-founders, other existing holders; the listed stock	1 per share	All general stockholder matters and election of the non–Class T directors. Aggregate Class A voting power is capped at 49.9% because Class F is fixed at 50.1%	Pro rata dividends and liquidation	Not convertible
N	Primarily strategic partners (Amazon and Google will each hold more than 5% of capital stock in this class); also the company’s Public Benefit Fund	None, except one vote per share where Delaware law requires a class vote	Not general matters	Same pro rata economics as Class A	Converts 1-for-1 into Class A on a sale or transfer to an unaffiliated third party
F	One share, held by Founder LLC (initial members: Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, Christopher Olah)	A number of votes equal to 50.1% of total voting power on the record date—not a fixed multiple	General stockholder matters, including election of directors other than Class T directors; certain LTBT/Class T amendments; redemption of Class T and Class E	None (par value only on liquidation)	Not convertible. Redeemed for nothing on sunset or dissolution of Founder LLC. Transferable only in a board-approved reorganization of the LLC that does not change its members
T	One share, held by the Long-Term Benefit Trust	1, but only on specified matters	Election and removal of the Class T directors, who are a majority of the authorized board; redemption of Class E; certain amendments to Class F rights	None (par value only)	Not convertible. Redeemable for nothing only if the Trust Amendment Threshold is met. Transfer requires that same threshold
E	Employees, excluding co-founders (authorized at IPO; not issued yet)	1 per share, only on narrow matters	The “Specified Class T Director” if the nominating committee does not concur with the trust; certain LTBT or Class T amendments; Class T redemption	None (par value only)	Not convertible or transferable. Redeemed for nothing on specified events, including the Class F sunset path

Preferred stock does not survive the IPO as a separate voting class. Its protective provisions from the private rounds are replaced by the post-IPO charter mechanics above.

How the two control levers work

Founder vote, detached from ownership. Class F is not “10 votes per share.” It is one share whose vote count is defined as 50.1% of whatever the total voting power is on that record date. Issuing more Class A—or selling more stock to fund compute—cannot dilute it. Founder LLC votes that share as directed by a majority of its members, one vote each. If the continuing founders cannot agree, Class F is voted proportionally with the Class A votes actually cast. Benjamin Mann, often described as a co-founder, is not a member of the LLC and has no say in the Class F vote.

The structure is built to survive philanthropy and death. Each co-founder has pledged 80% of personal Anthropic equity to philanthropy; the S-1 states those pledges are voluntary and revocable, not binding obligations. Because Class F has no economic rights, donating or selling Class A does not reduce the 50.1%. The Founder LLC agreement lets each member designate a ranked, confidential list of up to 10 successors (the first eight must meet employment or public-benefit-organization criteria), subject to approval by five of the other six. A Class F Sunset Event occurs the first time fewer than three qualifying members have been in the LLC for 90 consecutive days. A 450-day transition period then runs, after which Class F (and Class E) is redeemed for no consideration.

Board majority, held by a trust with no equity stake. The Long-Term Benefit Trust, created in 2023 as a Delaware noncharitable purpose trust, holds Class T. Class T generally elects and removes the Class T directors, who together are a majority of the authorized board. Secondary readings of the S-1 put the post-IPO board at seven seats: four Class T, three elected by Class A and Class F voting together. Because Class F already has 50.1% of that combined vote, the founders effectively choose those three, including the CEO and the chair, without needing public support. The trust does not vote on ordinary stockholder matters and, per the filing, has no day-to-day management authority.

There is a narrow employee check. Before the trust nominates or removes a director it must give the nominating and corporate governance committee at least 30 days’ notice. One seat—the Specified Class T Director—needs that committee’s concurrence. If the committee does not concur, and the trust does not withdraw, Class E holders vote. Changing the trust’s terms or the rights of Class T (the “Trust Amendment Threshold”) is deliberately hard: if the trustees favor the change, it needs 80% of the authorized board and 80% of trustees in office; if they oppose it while Class F is outstanding, it needs a high founder supermajority plus Class E and Class A majorities. Class T does not sunset on a calendar.

Voting power versus economic ownership

The divergence is the point of the design, and it is larger than at Meta or Alphabet because the controlling instruments own nothing.

• Founder LLC / seven co-founders. 50.1% of general voting power and the practical ability to elect the non–Class T directors, with zero economic rights on Class F. They also hold large personal Class A stakes—secondary accounts of the principal-stockholders table put each co-founder in a range of roughly 232 million to 245 million Class A shares as of September 30, 2026, after the split—but those Class A votes sit inside the 49.9% residual and cannot override Class F. As they donate Class A under the 80% pledge, economic ownership falls and voting power on general matters does not.
• Long-Term Benefit Trust. Power to elect and remove a board majority, plus veto-level rights over amendments to its own charter position. Economic ownership: par value of one share. Trustees are paid cash that does not vary with company profits and may not hold Anthropic securities. Current trustees named in coverage of the filing include Neil “Buddy” Shah (chair), Richard Fontaine, Ben Bernanke, and Christopher Stubbs, with one of five seats vacant. The S-1 notes that, in practice, trustees and company executives have collaborated closely on selecting trustees.
• Class A public holders. Pro rata claim on dividends and residual value, and one vote per share, but collectively never more than 49.9% of general votes. They cannot elect the Class T majority and cannot outvote Class F on mergers, charter amendments, or the other directors. The prospectus states that Class A holders generally will not be able to affect the outcome of matters submitted to stockholders.
• Amazon, Google, and other Class N holders. Full economic participation, no general vote. The S-1 states Amazon and Google will each hold more than 5% of outstanding capital stock after conversion, in non-voting Class N. That stock becomes voting Class A only if sold to an unaffiliated third party—so a partner cannot quietly accumulate control by holding; it has to exit. Anthropic also contributed 846,840,960 Class N shares, 5% of fully diluted shares as of June 30, 2026, to its wholly owned Anthropic Public Benefit Fund, with a board framework that can add up to another 2% at successive 50% price milestones. Those shares are economic, non-voting, and consolidated.

Controlled-company status follows from the trust, not from the founders. Nasdaq treats a company as controlled when more than 50% of the voting power for the election of directors is held by an individual, group, or another company. Because Class T elects a board majority, Anthropic says it is a controlled company. It expects to use the exemption only for the nominating and corporate governance committee, which will not be entirely independent. It says the board already has a majority of independent directors and that the audit and compensation committees are fully independent.

How this compares with the precedents

Standard dual-class tech IPOs attach extra votes to shares the founders still own. Alphabet’s Class B is 10 votes and converts on transfer; Meta’s Class B is 10 votes, and as of October 1, 2026 Mark Zuckerberg held about 60.6% of voting power with about 13.4% economic ownership. Snap sold the public only non-voting stock. Those structures can be diluted if the founder sells the high-vote shares, and they usually convert on non-permitted transfer. Sunsets are inconsistent: Airbnb has a 20-year sunset; many founder-controlled companies, including Meta’s current Class B, have no hard time sunset. Jay Ritter’s data show dual-class structures in a large share of recent tech IPOs—15 of 31 tech IPOs in 2025, and about 41% of all 2025 IPOs.

Anthropic borrows the non-voting partner class (Class N works like Alphabet’s Class C or Snap’s public stock, but only for strategic holders) and the idea of keeping control while giving away economics. It does not copy the mechanism. Zuckerberg’s 2015–17 attempt to create non-voting Class C so he could donate voting shares and keep control was dropped after a shareholder suit. Anthropic reaches a similar economic result up front: the control share never had economics, so giving away Class A does not touch the vote, and new issuance cannot dilute the 50.1%.

Public-benefit-corporation precedents are thinner and usually weaker on the mission side. Allbirds went public as a Delaware PBC with a conventional 10-vote founder class. Critics argued that combination lets founders sacrifice stockholder returns without being answerable to the diversified holders who bear externalities. Anthropic keeps the PBC balancing duty—directors must weigh the public-benefit purpose, not just stockholder value—but adds an independent body that elects the board majority and that the founders cannot remove by stockholder vote. That is closer to a mission lock than Allbirds, and different from OpenAI’s nonprofit-parent model: the LTBT does not own the operating company and cannot set day-to-day strategy. The filing itself says the governance structure has no direct public-company precedent.

The practical bargain for a Class A buyer is therefore unusually explicit. You are underwriting economics and a PBC whose board majority is chosen by financially disinterested trustees, while ordinary stockholder decisions—including the other directors, mergers, and charter changes—are decided by seven people (or their confidential successors) through a share that cannot be diluted and that disappears only after the founder group itself falls below three members for 90 days, plus a 450-day tail. Index providers that exclude multi-class or controlled companies may limit passive demand. Unsolicited bids and proxy fights for control are structurally difficult, which the prospectus acknowledges may deprive stockholders of a premium offer. The offset the company is selling is that this is how it intends to keep safety and the stated public benefit from being outvoted by either public holders or the cloud partners who fund the compute.

---

Recent Findings Supplement (October 2026)

Anthropic, PBC filed its S-1 on October 26, 2026 (the most recent and directly relevant filing matching the query criteria), introducing a novel multi-class structure that decouples voting control and board election rights from economic ownership far more aggressively than standard dual-class tech IPOs while retaining its Public Benefit Corporation (PBC) status.[1][2]

This structure uses a single “golden” Class F share for founder voting supremacy on general matters and a separate Class T share for an independent Long-Term Benefit Trust (LTBT) to control a board majority, creating layered protections for its AI safety/mission focus. It qualifies as a “controlled company” under Nasdaq rules.[3][4]

Share Class Structure and Mechanics

Anthropic authorizes five classes of common stock (Class A, E, F, N, T) plus blank-check preferred. The setup takes effect around the IPO (Class A shares listed on Nasdaq under ANT). Class F and Class T are single shares with fixed, non-dilutable rights; most other classes have no or limited economic claims.[4][5]

• Class A: Public offering shares. 1 vote per share on general matters. Full economic rights (dividends, liquidation, etc.).
• Class F: 1 share held by Founder LLC (initially the 7 co-founders: Dario Amodei, Daniela Amodei + Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, Christopher Olah). Fixed 50.1% of total voting power on matters submitted to stockholders generally (including election of non-Class T directors). No economic rights. Protected against dilution. Terminates if fewer than 3 continuing founders remain in the LLC for 90 consecutive days (triggering a 450-day transition period, after which it is redeemed for nominal value).
• Class T: 1 share (post 1-for-1,000 reverse split) held by the LTBT. Authority to elect and remove Class T directors (who form a board majority, e.g., 4 of 7). Votes only on those director elections; no general voting rights. No economic rights. No expiration.
• Class E: Issued to non-founder employees. Limited voting rights (primarily on narrow matters such as certain charter amendments or PBC status changes). No economic rights.
• Class N: Held by strategic partners (e.g., Amazon, Google holdings noted in coverage). Generally no voting rights. Converts to Class A upon transfer to unaffiliated third parties. Full economic rights.

Conversion and other mechanics: No automatic conversion triggers for Class F like typical dual-class transfers (it is a single fixed-vote share). Preferred stock (if any) would convert per standard terms pre-IPO. The structure explicitly preserves PBC status post-IPO.[6]

Voting Power vs. Economic Ownership Divergence and Controlled Company Status

Voting power is heavily concentrated and decoupled from economics:

• Co-founders (via Founder LLC/Class F): Control 50.1% of general stockholder votes with effectively zero economic ownership in that share (no dividends or meaningful liquidation rights). Each co-founder also holds substantial Class A shares (roughly 200–240 million each per reports), but the Class F mechanism ensures their collective vote cannot be diluted. They elect the minority of directors not chosen by the LTBT.[7]
• LTBT (via Class T): Controls board majority election/removal with zero economic stake or general voting power. Trustees are financially disinterested experts in AI safety, policy, etc.
• Public Class A holders: Hold nearly all economic ownership in the offered shares but collectively no more than 49.9% of general votes and no independent ability to elect directors or override the above.
• Class N/strategic holders: Economic ownership without voting power (until conversion).
• Class E: Minimal influence, no economics.

This creates a “controlled company” under Nasdaq rules (due to the trust’s board control and/or combined insider dominance), exempting it from certain independence requirements (company plans to use the exemption only for the nominating/governance committee).[3][4]

Protective provisions include standard anti-takeover elements plus PBC-specific balancing requirements and Class E voting rights on mission-related changes. The LTBT provides an independent “kill switch” layer (historically noted as requiring high supermajority to alter in some reports).

Comparison to Dual-Class Tech IPO Precedents and PBC Governance

Standard dual-class structures (e.g., Meta Class B 10:1 for Zuckerberg’s ~60% control; Alphabet/Google similar for Page/Brin; Snap non-voting public shares) tie supervoting rights to proportional holdings that can dilute over time via transfers, sales, or death/disability triggers, often with time-based sunsets in newer IPOs.[8]

Anthropic’s approach is more extreme and hybrid: a single non-economic golden share locks in fixed 50.1% voting (immune to dilution), fully separates board control to a mission-focused trust, and layers this on PBC duties. It lacks typical sunset or transfer-conversion triggers for the core control mechanisms.[2]

PBC status (retained post-IPO) requires directors to balance three interests per Delaware law: stockholder financial returns, stakeholders materially affected by the company’s conduct, and the specific public benefit (“responsibly develop and maintain advanced AI for the long-term benefit of humanity”). A July 2026 Delaware Court of Chancery ruling (Drakes Landing) confirmed no traditional Revlon duty to maximize shareholder value in change-of-control scenarios; informed, disinterested decisions balancing the three interests receive safe-harbor protection.[9][10]

This goes beyond OpenAI’s PBC setup (Foundation-controlled board) by adding the independent LTBT. Broader 2026 context: Council of Institutional Investors data shows 36% of 1H 2026 IPOs used dual-class structures (up from prior periods), with many lacking time-based sunsets.[11]

Implications for Investors, Competitors, and Market Entrants

Public Class A buyers receive economic exposure to Anthropic’s growth (strong revenue traction noted in the filing) but minimal governance influence, insulating the company from short-term activist pressure while potentially complicating index inclusion or attracting scrutiny on shareholder rights. The structure tests the limits of founder/mission protection in public markets.[12]

Competitors or new entrants seeking similar control would need aligned founder groups willing to forgo economics in control shares and credible independent mission overseers; replicating the PBC + trust + golden-share combination may face investor pushback or regulatory attention. It sets a precedent for highly mission-driven AI/tech firms prioritizing long-term safety over one-share-one-vote norms.[13]

Summary Table of Key Share Classes (Anthropic S-1, filed Oct 26, 2026)

Class	Holder	Votes	Economic Rights	Key Notes
Class A	Public investors	1 per share (general matters)	Full	Offered in IPO; ≤49.9% aggregate voting power
Class F	Founder LLC (7 co-founders)	Fixed 50.1% (general matters, incl. non-T directors)	None	Single share; dilution-proof; terminates on founder attrition
Class T	Long-Term Benefit Trust	Director elections only (Class T seats)	None	Single share; elects board majority
Class E	Non-founder employees	Limited (specific matters)	None	Safeguard on mission changes
Class N	Strategic partners	None (generally)	Full	Converts to Class A on unaffiliated transfer

This represents the primary new development in the queried area as of late October 2026. Earlier 2026 dual-class or PBC examples exist but lack this integrated, multi-layered control mechanism.

Anthropic’s October 26, 2026 Form S-1 (accession 0001628280-26-067168, file 333-299633) is a classic preliminary prospectus: the commercial terms of the deal are blank, and several ownership and commitment totals that depend on those terms are blank with them. Anyone quoting a valuation, a post-IPO stake, or a single “compute commitment” number is using press arithmetic, not the filed document.

The blanks that are actually in the filing

The cover and the “The Offering” summary are the places the deal size is withheld. Anthropic is selling an unspecified number of Class A shares; selling stockholders are selling an additional unspecified number; the estimated price is “between $ and $”; the per-share and total columns for price, underwriting discount, and proceeds to the company and to selling stockholders are empty dollar signs; the underwriters’ option (greenshoe) size is blank; and both the prospectus date and the expected delivery date are blank. The same holes reappear in the offering table: shares offered by the company, shares offered by selling stockholders, shares outstanding after the offering (Class A, Class N, and the total), and estimated net proceeds (including the greenshoe case), all keyed to an assumed midpoint price that is itself blank.

[1]

[1]

[2]

Three other families of blanks matter more than the cover:

• Headline compute total. The prospectus summary says purchase commitments and leases, including short-term commitments and leases not yet commenced, have an aggregate committed spend of “more than $” (blank), of which approximately $ (blank) is expected before December 31, 2030, and approximately $ (blank) is due in the next 12 months. Named counterparties are priced in the notes and MD&A; the rolled-up figure is not. [1] [3]
• Recent operating metrics that the marketing story depends on. Revenue “for the three months ended September 30, 2026” is printed as a blank dollar amount in the overview, even though another passage states that quarter as 86 times revenue for the three months ended December 31, 2024. Run-rate revenue “as of October 31, 2026” is also blank—and that measurement date is five days after the filing. [3] [4] [5]
• Ownership percentages and selling amounts. The principal and selling stockholders table gives share counts (for example, Dario Amodei beneficially owned 245,259,100 Class A shares as of September 30, 2026, including trusts and near-term options/RSUs; the other co-founders sit between about 232 million and 244 million). The percentage columns are blank because the denominators are blank: the filing says percentages are based on “shares” of Class A and “shares” of Class N outstanding before the offering, and on a similarly blank post-offering count, after preferred conversion, convertible-note conversion, and RSU settlement. How many shares each selling stockholder will actually sell is likewise unfilled. Amazon and Google do not appear as voting 5% holders because their economic stakes sit largely in non-voting Class N. [6] [7] [8]

Dilution and capitalization tables that reference “the assumed initial public offering price of $ per share, which is the midpoint of the estimated offering price range set forth on the cover page” are the same mechanism: every sensitivity ($1 move in price, 1 million-share move in size, full greenshoe) is a formula waiting for inputs. The fee table uses a placeholder maximum aggregate offering price (reported as a $1 billion placeholder on which a registration fee of $87,000 was paid), which is not the deal size.
[9]
[5]

Why these items are omitted, and when they usually appear

This is the ordinary sequence, not a concealment unique to Anthropic. A company can file a public S-1—and must, at least 15 days before a roadshow, after confidential review—before underwriters and the board have agreed a bona fide price range or a final primary/secondary split. SEC practice treats a preliminary prospectus with a price range (the “red herring”) as the document used to market the deal. Until that range exists, proceeds, dilution per share to new investors, post-offering ownership percentages, and the greenshoe share count cannot be computed without inventing the variable they depend on. The cover’s “Subject to Completion” legend and the delaying amendment (so the registration does not become effective 20 days after filing by operation of law) are the legal wrappers around that gap.

[10]

[11]

What typically gets filled, and when:

• Price range and share count land in an S-1/A filed at or just before roadshow launch. That amendment is what first implies a valuation (shares × high end of the range, adjusted for the pre-deal cap table). Reports have pointed to marketing that could start in the week of November 9–10 and a listing targeted for the week of November 16, which would put the range amendment in a narrow window after this filing—but those dates are “people familiar,” not a company commitment, and the filing itself says the sale will begin as soon as practicable after effectiveness. [12] [13]
• Selling-stockholder amounts and post-offering percentages are usually completed in that same range amendment, because both the new-issue denominator and the secondary shares sold have to be known. They can still move in a later amendment if the book is resized.
• Final price, final share count, and underwriting discount in dollars are set at pricing, often after the close, and appear in the pricing amendment and the Rule 424(b)(4) prospectus. The price can print inside, above, or below the range. A move outside a customary band can require recirculation or a new amendment before the deal prices.
• Stale or incomplete operating numbers (here, the September quarter in dollars, and a run-rate dated October 31) are filled when management is willing to put them in a filed document—often the range amendment or a subsequent S-1/A before effectiveness. The 86× ratio already in the business section is a bridge, not a substitute for the dollar line.
• The aggregate commitment figure, if the company chooses to fill the summary blank rather than leave readers to add the notes, would appear in an amendment. Component amounts (Broadcom equipment leases of about $161.2 billion, SpaceXAI up to $120.5 billion, Google at least about $114.5 billion, Amazon at least $110 billion, and others) are already disclosed and do not depend on the IPO price.

What can still change after the range is printed: the range itself (up or down on roadshow feedback), the mix of primary shares (cash to Anthropic) versus secondary shares (cash to selling stockholders), the greenshoe (typically 15% of the base deal, exercisable for 30 days), which holders sell and how much, and therefore every post-offering percentage and every dilution figure. Lock-up waivers by Goldman Sachs and Morgan Stanley can also change how much stock is perceived as coming to market, even if the cover share count does not move.

How headline ownership figures mislead

The S-1 separates economics from control so cleanly that a single “founders own X%” sentence is almost always wrong in one of those dimensions.

Voting power is not a function of share count. Immediately before closing, a Founder LLC whose initial members are the seven co-founders (Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, Christopher Olah—not Benjamin Mann) receives one Class F share entitled to 50.1% of the vote on matters submitted to stockholders generally, including election of directors the Long-Term Benefit Trust does not elect. That 50.1% does not dilute when new shares are issued. Class A holders, in aggregate, are capped at 49.9% of that general vote. The Trust’s single Class T share elects a board majority and has essentially no economic rights and no general vote. Class N (strategic partners, including converted Amazon and Google positions) has no vote except as law requires, and converts to Class A mainly on transfer to an unaffiliated party. Class E, for non-founder employees, has no economic rights and votes only on a narrow set of trust-related matters. A reader who treats “shares beneficially owned” as “control” will invert the structure.
[14]
[15]
[16]

Beneficial ownership is not outstanding shares, and neither is fully diluted. SEC beneficial-ownership counts include options exercisable and RSUs expected to settle within 60 days of September 30, 2026, and exclude everything else. The percentage math the filing will eventually publish is on shares outstanding after preferred conversion, convertible-note conversion, and net RSU settlement (shares withheld for tax never hit the outstanding count). Fully diluted ownership—every option, every unvested RSU, every note that can still convert, the Public Benefit Fund’s Class N block—is a third number. One filing-derived disclosure already shows the gap in kind: Anthropic contributed 846,840,960 Class N shares, described as 5% of fully diluted shares as of June 30, 2026 assuming full conversion of the Amazon notes, to a Public Benefit Fund, with board discretion to issue up to another 2% each time the price rises 50% from the prior milestone, starting at the IPO price. That 5% is a fully diluted construct; it will not match any “outstanding after this offering” percentage.
[17]
[8]

Convertibles make the denominator path-dependent. The $80.7 billion first-half net loss is mostly a $72.4 billion non-cash charge for changes in fair value of convertible notes, derivatives, and SAFEs as the private valuation rose—not cash spent. Those instruments convert in connection with the offering, so the post-IPO share count is not the June 30 or September 30 historical count. Amazon’s notes are described in secondary coverage of the prospectus as converting at prices as low as $2.09, subject to an ownership cap (reported as 12%); Google’s stake is described as capped at 14.99% of fully diluted capitalization, with an IPO acceleration option to buy up to $10 billion of Class N at the IPO price and excess pushed into zero-coupon roll-over notes. Because the conversion outcome can depend on the IPO price and on whether a cap binds, a pre-deal “Amazon owns the mid-teens” estimate and a post-deal Class N percentage are not the same fact. PitchBook’s calculation that full conversion would leave Amazon with roughly one-fifth of outstanding shares is an outside estimate, not a figure in the percentage column—which is blank.
[18]
[7]
[19]
[5]

Trusts, proxies, and nominees are already inside the founder counts. Dario Amodei’s 245.3 million includes shares held of record by him or by trusts he controls, plus 2.2 million option shares and 0.6 million near-term RSUs, and he holds irrevocable voting proxies over additional Class A shares in donor-advised funds (about 4.6 million in one account of the table). Daniela Amodei’s count includes a spouse’s options and RSUs. Those are correctly included in beneficial ownership under SEC rules; they are not the same as shares the individual can sell without trust or proxy constraints, and they are not the Class F vote, which is cast by the LLC, one founder one vote, under an agreement the filing says will be entered shortly before effectiveness.
[20]
[8]

Founder economic percentages circulating in the press cannot all be right, because the filing refuses to state one. Using the filed counts (roughly 232–245 million Class A each, about 1.69 billion together) against an unstated denominator has already produced “roughly 10%” combined, “about 2% each / 14% combined,” and “1.4% each.” Those are different guesses about fully diluted versus as-converted outstanding shares after the September 17, 2026 10-for-1 split of Class A and Class N. The S-1’s own instruction is the disconfirming evidence: percentage columns blank, do not compute them until the denominator is filed.
[6]
[7]

A further misread: the founders have described pledges to give away most of their equity (press accounts say 80%). The filing treats those as voluntary. Class F voting power does not shrink as economic stakes are donated, so the gap between “how much they own” and “how much they control” can widen after the IPO. The Class F sunset is not dilution; it is a tenure test (fewer than three continuing founders in the LLC for 90 consecutive days, then a transition, then redemption for nothing).

Press estimates that already conflict with the document

Claim in circulation	What the October 26 S-1 actually does
Raise of about $100 billion at a $1.5–2 trillion valuation, listing the week of November 16	No price, no share count, no valuation. Timing is attributed to people familiar with the matter.
Q3 revenue of $16.8 billion (investor-day accounts in Reuters, FT, Bloomberg)	Dollar line blank. A separate sentence says the quarter was 86× the three months ended December 31, 2024; on the filed ~$197 million base that is about $16.9 billion, close to but not the same as $16.8 billion.
“$518 billion” of compute commitments (September leak coverage)	That figure is not the public S-1’s aggregate; the aggregate is blank. Named deals summed by outside readers land near $637 billion, mixing “at least” floors (Google, Amazon, Microsoft, Broadcom leases) with “up to” or cancellable amounts (SpaceXAI can be cancelled by either party on 90 days’ notice after initial periods). A further $125–150 billion of 2028–2031 TPU capacity is an estimate, not a fixed commitment.
Amazon ~one-fifth, or mid-teens, or Google ~7% vs ~14%	Caps and conversion mechanics are described; voting-table percentages for these holders are not, because Class N does not vote.

[12]

[21]

[22]

[23]

[4]

[24]

The leaked-draft compute total and the public filing’s blank are the cleanest example of why a headline from a prior document should not be treated as the current one. Each S-1/A replaces the last. The September figures were never the October 26 prospectus.

Key uncertainties

• Deal terms. Price range, primary shares, secondary shares, greenshoe, and therefore implied valuation, net proceeds to Anthropic (the cover says the company receives nothing from selling-stockholder shares), and dilution to new investors. A $100 billion primary raise and a largely secondary deal are different companies from the buyer’s perspective; the filing currently allows both.
• Denominator. Post-conversion Class A and Class N outstanding, after net RSU settlement and any cap-driven roll-over of Amazon or Google notes into non-share instruments. Until that number is printed, every founder, employee, and strategic percentage is an estimate.
• Whether caps bind. A higher IPO price can push more of a capped holder’s economics into notes rather than shares, which changes both the Class N count and the “outstanding” denominator used for everyone else.
• The commitment total, if filled. It may be a floor-only number, a sum of floors and caps, or something that still excludes post-June 30 deals and the $125–150 billion estimate. Energy, water, and delivery risk sit under all of those contracts regardless of which total is chosen.
• The September quarter and the October 31 run-rate, when they appear in dollars. The 86× bridge and the $16.8 billion investor-day figure can both be “right” under slightly different definitions (gross vs a rounded print, consumption vs total revenue) and still move when audited or updated.
• Who sells. All seven co-founders are identified in coverage as prospective selling stockholders, with amounts blank, under a standard lock-up the leads can waive. Secondary size changes the float narrative without changing Class F control.
• Process risk. SEC comments can force more amendments; the roadshow date can slip; the range can be cut or raised; the deal can be delayed past the Thanksgiving window the press is using. Effectiveness is not scheduled in this document.

For anyone underwriting a position before the range amendment: the only ownership fact the S-1 has locked is control—Trust elects a board majority, Founder LLC holds 50.1% of the general vote, Class A is economically exposed and politically junior. Every percentage of economics, and every dollar of valuation, is still an input the next S-1/A has not supplied.

---

Recent Findings Supplement (October 2026)

Anthropic’s October 26, 2026 public S-1 filing provides the most recent high-profile illustration of standard bracketed/blank fields in a preliminary prospectus.[1][2]

The filing (accession details referenced in EDGAR listings) leaves the following key items blank or bracketed on the cover and in related tables:

• Estimated initial public offering price per share (e.g., “between $ and $ ”) and related per-share and aggregate proceeds figures.
• Number of Class A common shares offered by the company and by selling stockholders.
• Total raise size/proceeds (company and selling stockholders portions).
• Post-offering ownership percentages in beneficial ownership tables.
• Certain midpoint or assumed figures in dilution, capitalization, and proceeds tables (often tied to an assumed price that is left blank).[3][4]

Some numeric commitments (e.g., aggregate multi-year compute obligations) appear with ranges or are noted as undetermined in summaries. The document explicitly states it is “subject to completion” and that information may change.[4]

These omissions are standard SEC practice for preliminary S-1/prospectuses. Pricing, share counts, and exact proceeds depend on market conditions, investor demand during the roadshow, and final negotiations with underwriters. They are typically filled in via S-1/A amendments (or directly in the final prospectus filed under Rule 424B) shortly before or at pricing, which for Anthropic is anticipated around mid-November 2026 (roadshow timing and potential pre-Thanksgiving pricing noted in contemporaneous reporting).[5][6]

The blanks allow the company to gauge demand flexibly without locking in terms prematurely. What could change includes the final price range (driven by valuation sentiment, currently discussed in the $1.5–2 trillion range in press but not in the filing), share allocation between primary/secondary sales, overall size, and any related pro forma calculations. Over-allotment options and selling stockholder participation can also adjust.[7]

Headline ownership and control figures in the filing (and press coverage) carry several disconfirming or cautionary nuances that could mislead without full context.[2][8]

• Voting vs. economic rights and multi-class structure: Seven co-founders control 50.1% of voting power via a single Class F share held by Founder LLC (no economic rights, non-transferable in key respects). A Long-Term Benefit Trust holds Class T for board majority election. Public Class A shares have one vote each but limited influence. Class N shares (e.g., held by Amazon/Google) generally have no votes until converted/sold to unaffiliated parties.
• Fully diluted vs. outstanding: Tables often use outstanding or pro forma bases; full dilution includes unexercised options/RSUs, convertible notes (Amazon notes with low conversion prices and caps noted), warrants, and potential future issuances. The filing references Amazon/Google stakes and conversion mechanics that could shift percentages significantly.
• Selling stockholders and secondary sales: Some founders are listed as potential sellers; the company receives no proceeds from secondary shares. This affects post-offering floats and percentages.
• Trust/nominee holdings and pledges: Shares pledged to charity (80% by each co-founder in some cases) or held via trusts/LLCs; press estimates may not align precisely with beneficial ownership definitions in the filing (Section 13(d) rules).
• Press vs. filing discrepancies: Pre-filing leaks and valuation chatter ($1.5–2T range) differ from the filing’s lack of a stated target; actual post-IPO ownership will depend on final share count and any concurrent conversions.

Key uncertainties remain around final pricing/terms, exact dilution impacts from convertibles/options, and how control structures interact with public float.[9]

No major new SEC policy changes specifically altering S-1 blank practices appear in post-April 2026 sources; related updates (e.g., equity compensation/S-8 guidance or filer status proposals) address other disclosure/registration mechanics but do not change the core preliminary prospectus flexibility for offering terms.[10]

Other recent S-1 filings (e.g., smaller biotechs or resale registrations in October 2026) follow similar patterns but lack the same visibility.[11]

For competitors or entrants analyzing IPO readiness, Anthropic’s filing underscores that blanks are intentional and temporary—focus on the underlying business/risk factors and monitor S-1/A updates for the finalized economics. Ownership analysis requires careful review of all share classes, conversion features, and footnotes rather than headline percentages.