Fixed Against Floating
Anthropic filed its S-1 on 26 October and an amendment on 2 November that adds the third quarter of 2026. The price range, share count and offering size are still blank, so there is no valuation to test. What the filing does give is nine months of results under the company's own name, enough to score claims made before those numbers were public. In August, Benzinga asked Tessera's founder, Chan Ahn, what Anthropic would need to justify a $2 trillion valuation, and reported his answer as roughly $725 billion of revenue by 2036. Against the amended filing, revenue has run ahead of the figures he used: $33.2 billion in the first nine months of 2026, against $2.5 billion a year earlier, with $17.0 billion in the third quarter alone. His full-year estimate of $50 billion to $55 billion would need a fourth quarter of roughly $16.8 billion to $21.8 billion, so it looks more likely to prove low than high. The piece then turns to margins, stock-based compensation and about $895.4 billion of commitments, and sets out where the filing counts against him and where it does not.

4 November 2026. Figures are from Anthropic's amended registration statement of 2 November 2026 unless the text names another source.
Interest disclosure: T-OpenAI and T-Kalshi are tokenized loan participation rights providing economic exposure linked to those companies' pre-IPO valuations, each issued through a dedicated issuing subsidiary. T-SpaceX, issued through its own dedicated issuing subsidiary, provides economic exposure linked to SpaceX; following the SpaceX listing, the investment underlying T-SpaceX is being divested. Redemption opens only once a Redemption Start Date is announced, and as of 4 November 2026 none has been. This piece reviews Anthropic's registration statement, and Tessera's interest in how that listing is received runs in two directions. OpenAI competes with Anthropic. SpaceX sits on the other side as a supplier: Anthropic's filing records compute agreements of up to $120.5 billion with SpaceXAI, a SpaceX subsidiary, and those agreements depend on Anthropic paying. SpaceX's own prospectus also names Anthropic among its competitors, and on 14 August 2026 SpaceX acquired Anysphere, the company behind the coding tool Cursor. Tessera has no arrangement with Anthropic, issues no product referencing it, and plans none. It has no arrangement with Amazon, Google, Microsoft, Broadcom, Nscale or Lambda. Tessera's founder, Chan Ahn, worked earlier in his career at Goldman Sachs, JPMorgan and Credit Suisse, now part of UBS. Goldman Sachs is a representative of the underwriters of this offering, and JPMorgan and UBS are among the underwriters; Tessera has no arrangement with any of the three. Tessera, and entities under its control, hold related exposures. Read what follows with that in mind.
Anthropic filed its S-1 on 26 October and, on 2 November, an amendment that adds the third quarter of 2026. The price range, the number of shares and the size of the offering are all still blank, so there is no valuation in the document to test, and this piece does not attempt one. What the filing does contain is the first set of results Anthropic has published under its own name, now covering nine months. That is enough to score claims made before any of those numbers were public.
In August, Benzinga asked Chan Ahn, Tessera's founder, what it would take for Anthropic to justify a $2 trillion valuation. Its article of 24 August reported his answer as roughly $725 billion of revenue by 2036, assuming a 10% cost of equity, a 25% free cash flow margin and a 25x terminal multiple, or about $950 billion at a 13% cost of equity. The longest passage it quoted was: "You can underwrite the growth or you can underwrite the margin. Underwriting both at once is the leap being asked of public investors." Below, those claims, and the fuller written workings he sent Benzinga in mid-August, are set against the amended filing. Several of them come out against him.
Revenue Ran Ahead of the August Figures
Anthropic reports $33.2 billion of revenue for the first nine months of 2026, against $2.5 billion a year earlier. By quarter it was $4.7 billion, $11.6 billion and $17.0 billion (S-1/A, quarterly results of operations). Writing on 28 August from Wall Street Journal reporting, Chan used $4.8 billion for the first quarter and a projected $10.9 billion for the second, the same figures that went into his workings for Benzinga. The first quarter came in slightly below that and the second above it.
His workings put the full year at $50 billion to $55 billion. Reaching that range needs a fourth quarter of roughly $16.8 billion to $21.8 billion (our arithmetic), and the third quarter alone was $17.0 billion, so the range looks more likely to prove low than high. It cannot be scored until the year is reported, and the filing warns of lower usage around holiday periods in summer and winter: total customer spend fell about 20% in the two weeks to 2 January 2026, against the two weeks before.
Benzinga also reported his view that "Model leadership is temporary and must be repeatedly regained ..." Two disclosures bear on it. The filing states that "model versions released in the six months ended September 30, 2026 accounted for over 80% of our revenue for the three months ended September 30, 2026." That figure records customers moving to Anthropic's own newest models, so it is no evidence of leadership being lost. What it does measure is how much of one quarter's revenue rests on the latest releases, which is the exposure Chan was pointing to. The second disclosure touches the product Benzinga quoted him on as a moat, Claude Code, "where the customer is buying completed work rather than tokens". Claude Code is not broken out. The filing does report that over half of its customers paying $100,000 or more a year used both its developer platform and at least one of its own applications in the third quarter, up from about 30% in the fourth quarter of 2025. That fits the workflow position he described. It says nothing about how large the moat is.
The Second Quarter Beat the August Margin, and the Third Did Not
The margin figure Benzinga printed was "Anthropic's projected Q2 operating margin is just 5.1% ..." In Chan's workings that was $559 million of operating income on $10.9 billion of revenue, before stock-based compensation. The filing's Adjusted Operating Income excludes stock-based compensation too, and shows a margin of 19% ($2.2 billion) in the second quarter and 15% ($2.5 billion) in the third. It also excludes non-cash charitable contributions and a copyright settlement, so it is not quite his measure. Adding back stock-based compensation alone, as his figure did, gives about 19% for the second quarter, which beats the 5.1% he used and counts against him. The same calculation gives about 3% for the third quarter, below his figure; the third quarter reaches 15% only once its $1.96 billion donation-matching charge is excluded as well (our arithmetic, from the filing's quarterly reconciliation).
GAAP results, after stock-based compensation, are thinner. The second quarter shows operating income of $974 million on $11.6 billion of revenue, a margin of 8%. The third shows an operating loss of $567 million on $17.0 billion, a margin of minus 3%. Of the seven quarters the amended filing presents, the second quarter of 2026 is the only one with an operating profit.
Anthropic gives two reasons for the third-quarter loss: "$5.0 billion of compute under short-term commitments, which currently costs significantly more than comparable compute procured under our multi-year commitments, and $2.0 billion of non-cash charges related to our donation matching program." Compute and infrastructure, a line that covers research and model development as well as serving customers, took 58% of revenue in the third quarter against 48% in the second (S-1/A). The first reason sits awkwardly with the simple form of Chan's argument, because the expensive compute in the third quarter came from too little capacity under long-term contract rather than too much. The filing says its margins for the nine months "have expanded significantly year over year", and also that as new compute comes online, "we expect operating margin and Adjusted Operating Margin may compress in those periods."
Chan had conceded the point before the filing. On 28 August he set Anthropic's 2024 cash burn beside the projected 2026 figures, a first operating profit of $559 million among them, and called the distance between the two in eighteen months "the strongest evidence against the case I have made this week." The claim Benzinga printed looked out to 2036 on a 25% free cash flow margin, and his workings paired that margin with close to 30% compound annual growth for a decade. Two quarters settle that in neither direction.
The profit figures carry one more caveat, and the filing supplies it. Anthropic has granted employees restricted stock units whose performance condition is met when the registration statement becomes effective. "As of September 30, 2026, no stock-based compensation expense had been recognized for such RSUs ..." Had the offering completed on 30 September, the amended filing says, Anthropic would have recorded "$8.2 billion of cumulative stock-based compensation expense ..." at once, and a further "$24.1 billion of unrecognized ..." expense over about 2.0 years. Its other awards carry about $2.1 billion not yet recognized, and grants of about 47 million further RSUs approved since 30 September add an amount the filing leaves blank. The nine-month results already include $3.3 billion of stock-based compensation on those other awards. Chan's workings for Benzinga said the gap between the profit figure and GAAP would be "large and durable", and the filing now puts a size on it. The expense itself is non-cash. It lands on reported profit, and on holders through dilution, rather than on the free cash flow his August arithmetic used, although Anthropic intends to use part of the offering proceeds to pay the tax withholding due when the RSUs settle, at an amount the filing leaves blank.
The Split the August Workings Asked For
One part of Chan's workings never appeared in the Benzinga article: the one number he wanted from the S-1, which he said was not revenue. He asked for "the split between committed-spend enterprise agreements and pure on-demand usage. If a meaningful share of that run-rate sits under annual commitments, the mismatch largely closes and most of what I have said here softens considerably." The closest the article came was his objection to comparing Anthropic's multiple with Palantir's and Nebius's, because "annualized consumption revenue lacks the certainty of contracted revenue."
The filing gives no such split as a number, and its account of enterprise contracts shows why the two categories overlap: "A growing number of enterprise customers agree to non-cancelable annual or multi-year contracts with minimum spend commitments in exchange for a committed price list. These contracts result primarily in consumption-based revenue." Its own figure for consumption-based revenue, 79% of the nine months' total, therefore does not answer the question either. The nearest figure is remaining performance obligations, the part of customers' non-cancelable commitments not yet recognized as revenue: $13.8 billion at 30 September, about three-quarters of it due to be recognized within a year. That is a floor. It leaves out what committed customers have already used and anything they spend above their minimums. Even so, it is smaller than the third quarter's revenue, and the risk factors describe the revenue in terms close to Chan's: it "is primarily consumption-based, and much of it is not subject to long-term contractual commitments, while our compute and infrastructure commitments generally extend over multiple years." On the evidence the filing gives, the mismatch does not close.
The Commitments Now Have a Total
The amendment totals Anthropic's commitments, and the total is larger than the reporting Chan worked from in August. Counting the contractual obligations table, leases signed but not yet commenced, short-term contracts and agreements signed after 30 September, the filing gives "approximately $895.4 billion", substantially all of it for compute and infrastructure. About $447.5 billion is expected to fall due before the end of 2030, and most of the rest between 2031 and 2045. Arrangements with no fixed dollar amount sit outside that total; the filing estimates them at a further $150 billion to $175 billion. Three agreements carry minimums payable "regardless of actual usage": at least about $121.3 billion to Google, at least $110.0 billion to Amazon and at least $44.9 billion to Microsoft. Part of the total is conditional. About $213.8 billion relates to arrangements subject to termination rights, or to conditions that must be met before they start, such as the counterparty securing financing.
The SpaceX arrangement is the exception, and it is where Chan's August workings need a correction. They treated the SpaceX compute deal as a fixed commitment of $1.25 billion a month through May 2029, the figure in SpaceX's prospectus dated 11 June. That prospectus also said that after an initial three-month period "the agreements may be terminated by either party upon 90 days' notice", and on 17 August Chan had himself written that a run rate resting on cancelable agreements "is a contract, not an annuity ..." The amended filing puts the arrangement at up to $120.5 billion, about $3.1 billion a month once all capacity is delivered, which Anthropic expects in December 2026. It stays cancelable: after initial periods ending in January and March 2027, either party may end each agreement on 90 days' notice.
The original S-1 left blank the figure that puts both sides over the same period. The amendment fills it in: about $69.8 billion of obligations falls due in the 12 months after 30 September, of which about $4.3 billion is conditional or terminable. Over the same 12 months, about $10.4 billion of remaining performance obligations is expected to be recognized (our arithmetic, on the filing's 75%). Third-quarter revenue at an annual rate comes to about $67.8 billion (our arithmetic), roughly the size of that year of commitments, while the contracted backlog covers a fraction of them. The filing does not say how much of the rest is firm, or how the cancelable SpaceXAI agreements are counted; its total also includes about $17.4 billion of minimum payments under contracts cancelable without penalty within a year or with a year or less to run. At 30 September Anthropic held $95.6 billion of cash and $36.3 billion of short-term investments. In August, Chan's workings said the mismatch would show up as margin compression and a derating rather than as insolvency.
The counterparties could finance part of the fixed side themselves. At 30 September Anthropic had $77 billion of undrawn, conditional capacity to issue notes to Google, Amazon and Broadcom, or, for Google and Amazon, non-voting stock. Amazon's capacity opens as compute-capacity milestones are met, and Google's on compute conditions and revenue targets. Broadcom's $42 billion of notes may be issued, if needed, "solely to finance quarterly rent obligations under the 2026 Lease Arrangement", Anthropic's lease of Broadcom equipment, part of which Broadcom assigned in October to a newly formed special purpose financing entity. On 12 October Chan wrote that he could not find Anthropic's filing on EDGAR, and that Broadcom's report did not name the customer whose notes it described; the S-1 now shows a $42 billion note agreement between Anthropic and Broadcom. Those notes would let part of the rent be paid in paper rather than cash. The conditions are set below current results: the filing says Google's revenue targets are ones "which we have historically surpassed", and Broadcom's notes depend on revenue growth conditions "significantly below our actual revenue growth rates", on a minimum liquidity requirement and on continued performance under the leases. Google's facility can also be reduced by up to $10.0 billion if Google exercises its option to buy shares in a private placement alongside the offering. If revenue fell short for long enough, though, the conditional facilities are the ones that may not open.
The Lock-Up Follows the Earnings Calendar, as SpaceX's Did
The SpaceX half of the Benzinga article reported Chan's view that the fall in SpaceX shares "was driven more by earnings scrutiny and disclosure than by the August insider-share unlock", and that the main risk for later IPOs is the "first earnings report" rather than the lock-up expiry. SpaceX's lock-up released its first large tranche on the second trading day after its first earnings report, as Chan set out on 2 September. Anthropic's is drafted on the same calendar. Apart from releases from the first trading day and on four further dates, shares under the 180-day lock-up, and under an extended 365-day lock-up, are released on "the second full trading day on Nasdaq immediately following the public release of our annual financial results for the fiscal year ending December 31, 2026 ...", and on five dates counted from it. The share counts are blank, but the percentages are filed. For one group of stockholders under the 180-day lock-up, the earnings trigger releases up to 12.5% of their locked-up shares and each of the five later dates up to 9.5%, so up to 60% in that window; for non-founder directors, officers and certain employees, up to 10% is released on the first trading day and up to a further 35% after the earnings release. On Chan's reading of SpaceX, the decline came with the earnings and not on the day the shares became free to sell. Anthropic's schedule sets up the same test, with the two events two full trading days apart.
What the Filing Settles and What It Does Not
This piece does not value Anthropic, and it takes no view on whether the listing will go well or badly. The filing contains no price, and the valuation figures and 2028 forecast now circulating are reported, not filed. Nothing here is a forecast of failure.
What the filing settles is narrower. Revenue is running ahead of the figures Chan used in August, and the second quarter's margin before stock-based compensation beat his, which counts against the way he put it; the third quarter's fell short of it on the same basis. Under GAAP, the second quarter of 2026 is still the only one of seven with an operating profit, and it precedes a stock-based compensation expense on the performance-condition RSUs that the filing puts at about $32.3 billion on 30 September figures, $8.2 billion of it at once. He called the SpaceX contract fixed, and it was not. On the test he set in August, the filing gives a floor in place of the split, and the floor does not close the mismatch: about $69.8 billion of commitments falls due in the next 12 months, against about $10.4 billion of contracted backlog. Much of the revenue floats, and most of the commitments are fixed. If the 2026 annual results are the first Anthropic reports as a public company, as the lock-up's drafting implies, they will be the first time that gap is priced by investors who owe the company nothing.
Sources: Anthropic, PBC, Form S-1, SEC EDGAR accession 0001628280-26-067168, filed 26 October 2026, and Amendment No. 1 (Form S-1/A), accession 0001628280-26-068676, filed 2 November 2026: quarterly results of operations; MD&A, including Adjusted Operating Income, contractual obligations, infrastructure service arrangements and note purchase agreements; risk factors on stock-based compensation, revenue and compute commitments; Notes 13 and 17 and the lease note; lock-up table; read 28 October to 3 November 2026. Space Exploration Technologies Corp., prospectus (Form 424B4) dated 11 June 2026 and filed 12 June 2026, and Form 8-K filed 14 August 2026. Benzinga, Namrata Sen, "Anthropic Wants Investors to Buy Into a $2 Trillion Dream—But a Wall Street Veteran Says SpaceX's IPO Offers a Warning for AI Investors (UPDATED)", 24 August 2026. Chan Ahn, written answers to Benzinga, prepared 15 August 2026 and sent 16 August 2026 (unpublished). Chan Ahn, "Who Counts the Capex", 17 August 2026, "The Money Goes Round", 28 August 2026, "The Announcement Outlasts the Holder", 2 September 2026, and "In Exchange for the Guarantees", 12 October 2026, LinkedIn, as published.
This is market commentary, not investment advice. It is not a recommendation to acquire, hold or redeem any Tessera product, or to take or avoid exposure to any company mentioned.
T-Tokens are tokenized loan participation rights, not equity. High risk. DYOR. Not financial advice. Not available in the US or other restricted territories. tessera.pe/terms
