Source: five podcast readings of the public S-1, all fetched 2026-11-01:
raw/Everything_Important_in_Anthropic_s_IPO_Prospectus.md— The AI Daily Brief (ADB). The most complete numeric walkthrough.raw/Anthropic_Just_Filed_Its_S-1._Here_s_What_The_Headlines_Are_Missing..md— Nate B Jones (NBJ). The customer side.raw/What_Anthropic_s_S-1_Tells_Us_About_the_A.I._Business_Hard_Fork.md— Hard Fork (NYT), with NYT reporter Erin Griffith.raw/Anthropic_S-1_Shocker_Greatest_Business_Ever_Google_s_500B_Wipeout_Midterm_Predictions.md— All-In ep. 295 (Anthropic segment only).raw/Anthropic_s_2T_Cash_Machine_the_End_of_Neoclouds_and_Chinese_Models_Catch_Up_306.md— Moonshots (Peter H. Diamandis) ep. 306 (Anthropic segment only).
Added 2026-11-09:
raw/Ep._249_-_Anthropic_Files_for_IPO_AI_and_the_Meaning_of_Work_Microsoft_s_60K-Employee_Buyout_Offer.md— The Artificial Intelligence Show (AIS) ep. 249, Paul Roetzer and Mike Kaput, recorded 2026-11-03 (https://www.youtube.com/watch?v=_OODHaVlMH8). The contract-by-contract commitment figures.raw/Companies_Are_Spending_5x_More_on_AI_Now.md— The AI Daily Brief headlines (https://www.youtube.com/watch?v=INC89yHLexM). IPO odds and investor pushback after the November sell-off.raw/Big_Tech_s_3T_AI_Bet_Claude_Writes_90%_of_Anthropic_s_Code_AIs_Editing_AI_Brains_EP_308.mdandraw/Anthropic_s_IPO_in_Question_Midterms_Hit_AI_Nasdaq_Drops_5%_and_RSI_Accelerates_309.md— Moonshots eps. 308 and 309 (https://www.youtube.com/watch?v=x09f_BuJsoY, https://www.youtube.com/watch?v=WF-ADiC7Khs). Internal-usage figures and the hosts’ views on IPO timing.
Added 2026-11-13:
ai-research/anthropic-s-1-a-amendment-2-excerpts-2026-11-13.md— the filing itself: Amendment No. 2 to Form S-1, filed 2026-11-10, read by keyword search (not end to end). The first figures in this article taken from the document rather than from a host.ai-research/yahoo-benzinga-anthropic-ipo-could-top-1-5-trillion-2026-11-13.md— Benzinga’s relay of the Financial Times (FT original paywalled, not read).raw/Wall_Street_Has_No_Idea_What_Anthropic_is_Worth.md— The AI Daily Brief (https://www.youtube.com/watch?v=gCOMQPMhIL0), the roadshow and valuation debate.raw/Jensen_OUT_as_Nvidia_CEO_-_Spy_Scandal_Export_Restrictions_Market_Impact.md(All-In) andraw/Jensen_s_Departure_NVIDIA_s_succession_plan_AI_doomer_response_and_Demis_talks_RSI_311_Live.md(Moonshots ep. 311) — IPO terms and demand as relayed on air.
Creator: five shows, listed above · URL: https://www.youtube.com/watch?v=IwCkMupiZ4s (ADB) · https://www.youtube.com/watch?v=EAEixlPC6L4 (NBJ) · https://www.youtube.com/watch?v=EGC7jht_mis (Hard Fork) · https://www.youtube.com/watch?v=p1Dy4e4ff0s (All-In) · https://www.youtube.com/watch?v=WXELsIk4o0g (Moonshots) · Duration: not recorded · Platform: YouTube
Anthropic made its S-1 public in the last week of October 2026. Five shows read it within days. This article puts their figures side by side, names the show behind each one, and marks where they disagree. Everything before the public filing (leaks, the confidential draft, the Q3 investor day) is in Anthropic’s Position, August 2026 and is not repeated here. Most figures below are hosts reading the filing on air; the 2026-11-13 “Amendment No. 2” section is from the filing itself. A November update (2026-11-09) adds a sixth reading, which lists the compute commitments contract by contract, and the state of the IPO after the November 5 market sell-off.
Key Takeaways
- Checked against the filing (2026-11-13). Amendment No. 2 gives nine-month revenue of 17.0B, about 34.9B) that either side may now terminate. The FT reports the IPO may price above the $70-78 range. See “Amendment No. 2, read from the filing”.
- Revenue: 4.6B for all of 2025 (ADB, NBJ, All-In, Moonshots). All-In and Moonshots put that at about 14× the first half of 2025, which Moonshots gives as 11.6B (All-In). Hard Fork’s host said “16.3B, against 16.8B Q3 figure comes from the investor day, not the filing (ADB).
- First GAAP operating profit: 600–700M, against more than 16.3B is about 693M operating loss** for H1.
- Gross margin rose from about 40% in 2025 to about 85% in H1 2026. ADB and NBJ both say this is calculated from the expense disclosure, and NBJ notes it excludes R&D (new-model training). All-In’s Friedberg calls 85% the incremental gross margin on each new dollar of revenue.
- **The 72B is an accounting charge on convertible debt, caused by the rising share price (ADB). All-In (Sacks) and Hard Fork give the same explanation: notes issued to infrastructure partners such as Amazon were revalued as Anthropic’s valuation rose. AIS’s exact figures: an 72.4B was the non-cash charge.
- Compute efficiency. Revenue per compute dollar went from 2.08 in 18 months (Moonshots), with revenue up 29× while compute cost rose 6×. Compute spend is 8.9B of it in H1 2026, and about 70% goes to internal work (research and training) rather than serving customers (ADB).
- **Customers paying 1M and about 250 pay more than 100k–1M, about 1,650 at 10M, and about 40 million individuals and small businesses below $100k.
- Net revenue retention was 900% as of September (ADB, NBJ). Wiki readers saw this figure before: in October it was single-sourced to All-In’s investor-day account. It is now in the filing.
- Concentration is falling. The top five customers were about 36% of revenue at the end of Q1 and are now 21% (ADB). Moonshots says two customers are “about 20%” of H1 revenue. 48% of H1 revenue came through cloud partners and marketplaces (ADB, NBJ). Moonshots names Amazon, Google and Microsoft and calls them “also its competitor”.
- Commitments: 600B signed in total, and estimates that could reach 20–40B a year. Moonshots says “more than 500 billion”, including **114.5B, Amazon 44.9B, Broadcom about 120.5B. The first four add up to $430.6B, which matches ADB’s non-cancellable figure, and the SpaceX contract can be cancelled after early 2027. ^[inferred] Details in the section “Compute commitments, contract by contract” below.
- Cash: 141B raised in total (ADB). Interest on it was 112B, Sacks “about 100 billion”, Moonshots’ Mostaque 130B raised. ^[ambiguous] All of them agree on the point: Anthropic has burned only about 10–20B.
- The IPO itself: ticker ANT on the Nasdaq. The listing is targeted for the week of November 16 (WSJ via Moonshots; All-In says November 16). Goldman Sachs and Morgan Stanley lead (All-In). Anthropic is aiming to raise about 1.5–2T (All-In, Moonshots). Polymarket gave 78% odds of raising at least 1T (ADB). Details in the section “IPO pricing and timing after the November sell-off” below.
Customers: how far into the budget Anthropic has reached
NBJ argues that the customer data matters more than the headline loss. These are the figures he and ADB read from the filing:
- More than half of the Fortune 500 are substantial paying customers. Among the publicly traded ones, the average spends less than 1,000 of total expenses, measured on the September quarter (NBJ). ADB gives this as “less than 0.1%”.
- Anthropic’s 10 largest Fortune 500 customers spend more than 8× the average share of expenses, and that top group is still below 1% (NBJ).
- The filing calls out the 10 largest banks and 9 of the 10 largest pharma companies as customers (NBJ).
- The filing says most enterprises have deployed Claude “across a very small number of teams and use cases” (NBJ, paraphrasing the filing).
- Multi-surface buying is rising. In the September quarter, more than half of $100k+ customers used both the Claude platform (API) and a first-party app such as Claude Code. The share was about 30% in Q4 2025 (NBJ).
- Recent models do the earning. More than 80% of September-quarter revenue came from model versions released in the previous six months (NBJ).
- Opus 5.5 cut cost per task by about 40% against Opus 5, through lower prices and better task efficiency (NBJ, citing the filing). This matches the launch claim recorded in Claude Opus 5.5.
- NBJ reads the filing as saying Cowork, the non-technical product, is growing faster than Claude Code did. The filing also names Salesforce and Legora among companies building products on Claude.
- Inside Anthropic: Claude writes 90% of the company’s code. Internal Claude use, valued at API prices, passed payroll as an expense in April 2026 (ADB).
- Training customers’ engineers: Anthropic has committed $100M to a Claude Frontier Academy, which aims to train 10,000 engineers by the end of 2027 to lead Claude deployments in their own organisations (NBJ). Compare the “86 trained” figure recorded in August in Anthropic’s Position.
Financing beyond the cash pile (ADB)
- 77B more from Amazon, Broadcom and Google as promissory notes. Sacks on All-In gives “another 80 billion”.
- **A 15 revolving”; the unit is assumed to be billions.)
- TPU purchases go through third-party vehicles: a 63B vehicle, again with Broadcom. Both were structured with Apollo and Blackstone. The June vehicle is probably the 50B Buildout]].
- Credit rating. The S-1 says little about a path to investment grade. Tony Ziringer (Impax Asset Management) says the balance sheet supports investment grade “at today’s leverage”, but that the rating will depend on “the commitment schedule, short operating history, and circular counterparty relationships.”
- The founders’ letter says the compute spending “may seem irresponsibly large at first glance” and calls it “bold but sane”.
Risk factors
- Length. ADB describes a 400-page document with more than 90 pages of risks. Hard Fork says “300 plus” pages and a 93-page risk section. The October reporting in Anthropic’s Position cited 80 risk pages in a 261-page prospectus. That was probably the confidential draft.
- What the risk section dwells on. Chamath’s team counted words in it: compute comes first (data centres, power, chips, grid), then regulation (policy, sanctions, export controls), then zero data retention, privacy and IP, then AI safety. Customer traction barely appears (All-In).
- Insurance. The filing says Anthropic may be uninsurable: “No single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure” (ADB).
- Growth will slow. Revenue grew about 1,000% year on year, and Anthropic says it does not expect that to continue and cannot predict future growth (ADB). Hard Fork reports Q3 growth of 44% and attributes the slowdown to OpenAI catching up on coding, companies pulling back on “token maxing”, and cheaper open models. ADB gives about 45%.
- Compute partners. SpaceX, Google and AWS may not always be aligned with Anthropic’s long-term interests and may not honour contracts as competition heats up. The filing gives this as the reason Anthropic has started building its own data centres (ADB).
- Existential risk, stated in the passive voice. “Concerns have been raised both within our organization and by other researchers” that advanced AI could pose “catastrophic or existential risks to humanity”. Hard Fork credits Matt Levine with pointing out the passive voice.
- Model welfare can delay deprecation. “Considerations relating to model safety, interpretability, alignment, and welfare may lead us to limit or delay the deprecation of older models.” Hard Fork credits Joe Weisenthal with spotting it. See Claude Model-Welfare Policy.
- Distillation. Mostaque says the S-1 reports Anthropic banned 7% of user accounts, which he takes to be distillation attacks. On All-In, a host relays “scuttlebutt” that some of the Q3 slowdown came from Anthropic voluntarily cutting off “99% of distillation”. Neither figure is confirmed.
Governance, as the five shows describe it
- Founders: the seven co-founders hold 50.1% of voting power through a new Class F share (ADB, All-In, Moonshots). Hard Fork counts five classes of stock in total.
- The Long-Term Benefit Trust (LTBT) holds Class T shares, which let it appoint and remove a majority of the board (Moonshots, Salim Ismail). Hard Fork says the founders also keep a confidential list of successors that they all have to approve.
- Other pledges in the filing: the co-founders pledge 80% of their equity to charity (Hard Fork, Jasmine Sun). The founders’ letter promises to keep products free of advertising: “Once advertising becomes part of a company’s revenue targets, the pressure to optimize for engagement tends to grow” (ADB).
- Competitors named: only OpenAI and Google (ADB).
- Market size: Anthropic gives no clear total addressable market (TAM). It values global knowledge work at about 22.5T of cumulative economic impact by 2031 (ADB). All-In’s Jason calls the $40T figure the TAM.
- Early investors (Moonshots): the 2021 Series A (43B. The Series B (500M stake would now be worth about 1.3B in 2024.
- Opinions, recorded as opinions. Ismail calls the LTBT “embedding the MTP [massive transformative purpose] into governance”. Alex Wissner-Gross compares it to Chinese Communist Party ideological control of companies. Mostaque says Anthropic “should not be a public company”. Sacks calls Anthropic a coming “AGI monopoly” pursuing regulatory capture.
The argument the filing does not settle: cheaper tokens, more revenue?
- The Jevons case. Tomasz Tunguz (via ADB): after Opus 4.5 launched in November 2025 with lower cost per task than Opus 4.1, quarter-on-quarter revenue growth went from 48% to 132%. Gavin Baker (via ADB): revenue per GPU can rise while token prices fall, provided per-GPU throughput rises faster. Anthropic says customers generally expand usage after releases that combine more capability with lower task cost (NBJ).
- Outside spending data. In Ramp’s token-management sample, total token volume rose about 52% between the weeks ending September 13 and October 11. Spend rose about 23%, so spending grew even as cost per token fell. The caption is garbled, so the pairing is approximate. ^[ambiguous] OpenRouter found that nearly 90% of September’s agent tokens were cached content (NBJ).
- Margin, not just price. Sacks (All-In) says the cost of a given task falls about 10× a year. He says Anthropic passes on about half of that as lower prices and keeps the rest as margin, through “compute multipliers” (algorithmic and data improvements). Sacks also cites an Anthropic August blog post reporting that 26% of next-model development tasks already used recursive self-improvement. That is his reading, and the post is not in this wiki.
- The bear case. The ROI case for 1M+ customers are Global 2000 companies, with the rest AI natives. He also warns that AI-sector borrowing is now competing with US Treasury issuance (All-In). Hard Fork: many contracts are non-cancellable, so if demand falls short Anthropic is “sitting on all these contracts”.
NBJ’s five June questions, answered from the filing
- Are heavy users getting cheaper to serve? Not shown. Revenue grew faster than compute spend, but the filing has no per-cohort serving cost.
- Does gross margin improve as usage grows? Yes: about 40% in 2025, about 85% in H1 2026, and a GAAP operating profit in Q2.
- Are customers buying software that scales, or custom deployment work? Both, and the filing does not separate how much growth depends on forward-deployed engineers.
- Are customers building real workflows? Yes. The filing names insurance underwriting, clinical reports and portfolio analysis, and more than half of $100k+ customers use both the platform and an app.
- Is forward-deployed engineering a bridge or a permanent requirement? Unknown. The Frontier Academy is a bet on making customers self-sufficient, and that link is not proven yet.
Compute commitments, contract by contract (added 2026-11-09)
Mike Kaput read the commitments line by line on AIS ep. 249. These are the most specific figures the wiki has for them.
| Counterparty | Commitment | Runs through | Terms as read on air |
|---|---|---|---|
| at least $114.5B | 2033 | Minimum spend that applies regardless of usage, “subject to limited exceptions” | |
| Amazon | $110B | 2036 | Same minimum-spend terms |
| Microsoft | $44.9B | 2033 | Same minimum-spend terms |
| Broadcom | about $161.2B in payments, excluding interest | 5-year equipment lease | AI compute hardware |
| SpaceX AI | $120.5B | 2029 | ”More flexible”: after initial lock-in periods ending in January and March 2027, either party can cancel on 90 days’ notice |
- How this reconciles the earlier readings. Google, Amazon, Microsoft and Broadcom add up to **417–431B non-cancellable”, which ADB said included exactly those four counterparties. Adding SpaceX gives **600B signed” would then include contracts not read on air.
- The SpaceX deal is cancellable, as All-In claimed earlier (see the SpaceX compute deal), but only after the 2027 lock-ins.
- Why sign this much (Roetzer, AIS): the S-1 still lists running short of compute as a risk. Roetzer says Mark Zuckerberg and Sam Altman have both said in recent interviews that they bought too little compute in 2024, and paraphrases Dario Amodei as saying Anthropic never buys enough compute, but that buying enough would bankrupt it. Kaput’s summary: “if you want to bet on something, bet on… the cloud providers.”
- Scale in power (Moonshots ep. 309, Alex Wissner-Gross’s own triangulation from public sources): about 4 GW of Anthropic compute by the end of 2026, aiming for 16–18 GW in 2027–2028.
IPO pricing and timing after the November sell-off (added 2026-11-09)
- Timing as of November 3: The Information (via AIS) pointed to the week of November 16. ADB, the day after the crash, says Anthropic was aiming to list before Thanksgiving, which would require the roadshow to start the following week.
- Odds after the crash (ADB): on Thursday, November 5, prediction-market odds of an IPO before Thanksgiving fell from about 85% to 60%; odds of listing by year-end fell from 95% to 66% over the week. Rumours circulated that the IPO might be pulled.
- All-In’s version: Calacanis cites Polymarket at a 71% chance of listing this year, down from 95% before the midterms, and says the Wall Street Journal reported the valuation “haircut could be baked in now”, around 1.5–2 trillion (
raw/Midterm_Blood_Bath_4T_AI_Sell_Off_Data_Center_Freeze_What_s_Next_for_Trump.md). ADB’s year-end figure above is 66%; the two were probably read at different times. - Price pushback (WSJ, via ADB): three major investors who plan to take part say the valuation needs to come down, one of them to “closer to 100B raise. Thinly traded secondary markets still priced Anthropic at about **2T target.
- Run rate (WSJ, via ADB): **65B reported at the end of July. Salim Ismail on Moonshots rounds it to “close to… 100 billion a year”. Reuters’ earlier expectation of $110B by year end is in Anthropic’s Position.
- The Moonshots hosts disagree on timing (ep. 309, opinion): Wissner-Gross says there is “lots of public reporting” that bankers are trying to force the valuation down, but Anthropic may want liquidity before the regulatory environment expected in January. Dave Blundin doubts a delay because the roadshow was triggered when prediction markets already showed the election outcome; he compares it with Meta’s IPO, which he says Wall Street priced down and which then traded lower before rising many times over. Ismail expects Anthropic to push through.
- **Where 5T, Nvidia 3.9T, Alphabet 2.6T, TSMC 1.8T, Saudi Aramco 1.6T, Meta 1T. Roetzer: “a very very strong probability that Anthropic will be worth more than Meta when they IPO.” Market moves are in AI Bubble Signals.
Price range and private placement (added 2026-11-11)
Sources: The Neuron Daily, November 11 (raw/newsletter-theneurondaily-com-14c86ec601.md), checked against CNBC’s report of the amended filing (ai-research/cnbc-anthropic-ipo-price-range-2026-11-11.md, https://www.cnbc.com/2026/11/10/anthropic-sets-price-range-of-70-to-78-in-upcoming-ipo.html, published 2026-11-10). The filing itself was not fetched.
- Range: 825 million Class A shares at 64.35B, valuing Anthropic at up to about **965B in its May funding round.
- Private placement: AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC agreed to buy $26B of Class N common stock at the IPO price; those shares are not registered in the offering and not subject to underwriter lock-ups. The Neuron calls the range “corrected”.
- Timing and demand (CNBC’s unnamed sources): pricing November 18, trading November 19; about 20% retail allocation targeted against a typical 5-10%; the book already covered at the range.
- Numbers in the amended filing (CNBC): revenue of 104B net loss for the period, mostly from changes in the fair value of convertible notes and derivative liabilities; run-rate revenue of about **10B in a month. A linked CNBC headline gives 895.4B of purchase commitments, which matches Intelligent Machines’ “17 billion” and “900 billion” (see Open Questions).
- Valuation pressure: CNBC reports investors had looked toward 1.5T was now in doubt; the range lands below that.
Roadshow week, as discussed on Moonshots ep. 310 (added 2026-11-11)
Source: Moonshots ep. 310 (raw/Jensen_s_Spy_Probe_Anthropic_s_Debut_Week_and_Elon_s_30M_Robot_Workforce_310.md, https://www.youtube.com/watch?v=ILQwMvCA2_o). The transcript has no speaker labels; attributions are from context.
- Roadshow and valuation (Diamandis): Anthropic “launched their IPO road show today”, pricing at about **1.5T and the 1.405T and the November 18 pricing date above.
- Why price it lower (Blundin, opinion): pricing down “to try and give it a pop”; he adds that Anthropic does not need to raise the full $100B on day one and could use a shelf registration and a secondary offering soon after. He expects Anthropic to pass OpenAI in capital raised and valuation at listing.
- Autonomy levels (Mostaque): he says Anthropic “put in their S1” that it has reached “autonomy level two” on its own scale, meaning models “just now need high-level human oversight”, and that level three means models can do all AI research “at a solid level”. No other show in this article reads this from the filing, and the scale’s name may differ in the S-1.
- Employee wealth (Mostaque, estimate): the IPO will make “almost 5,000 millionaires” and “dozens of billionaires” at Anthropic. Not from the filing.
- SpaceX as supplier (Wissner-Gross, opinion): he calls Anthropic SpaceX’s “largest customer” and SpaceX a “neo hyperscaler”. Compare the $120.5B SpaceX AI commitment in the contract table above.
Amendment No. 2, read from the filing (added 2026-11-13)
Source: the S-1/A filed November 10 (ai-research/anthropic-s-1-a-amendment-2-excerpts-2026-11-13.md). Only keyword-located passages were read; the financial statements and risk factors were not read end to end.
- Offering structure: 825,000,000 Class A shares (801,351,354 from Anthropic, 23,648,646 from selling stockholders) at 26.0B Class N private placement at the 10.0B (exercising a 10.0B, AMD about 1.0B each, TSMC Partners about 303M of Class A shares.
- Net proceeds (StockTitan’s summary of the use-of-proceeds section): about 93.7B if the option is exercised. The AI Daily Brief’s “around $100 billion” of stock floated is the gross figure at the top of the range with the option.
- Nine months to September 30, 2026 (unaudited): revenue 2,545M a year earlier; compute and infrastructure 1,260M (down from 104,039M, of which 41M (2023), 4,585M (2025).
- Q3 alone: revenue $17.0B, up 1,127% year over year, while compute and infrastructure expense grew 382%.
- Operating metrics: run-rate revenue about 100k+ of trailing revenue (from more than 950 a year earlier); net revenue retention about 900%; revenue per employee over 1.4M). Models released in the prior six months produced over 80% of Q3 revenue.
- The Claude Code spending gap: Claude Enterprise users who did not use Claude Code spent under $75 a month on average in Q3; the top 5% who did use it spent about 20 times that.
- Concentration: no customer above 10% of nine-month revenue; the top five about 25% of nine-month revenue; about 79% of Q3 revenue came from outside the top five (64% in Q1). So the top five were about 21% in Q3, which matches ADB’s “21%”.
- Cash: capital raised to September 30 was 95.6B plus 131.9B together, close to ADB’s $132B).
- Commitments: about **451.6B due before the end of 2030 and $70.7B in the next 12 months. Contract by contract:
| Counterparty | Filing (S-1/A, Nov 10) | As read on AIS ep. 249 (Nov 3) |
|---|---|---|
| at least about **125-150B (not in the obligations table) | at least $114.5B through 2033 | |
| Amazon | at least $110.0B, May 2026-April 2036 | $110B through 2036 |
| Microsoft | at least **49.2B, November 2026-June 2033 | $44.9B through 2033 |
| SpaceXAI | up to **3.1B a month once all capacity is delivered (expected December 2026); cancellable on 90 days’ notice after lock-ins ending January and March 2027 | $120.5B through 2029, same cancellation terms |
| AMD | no fixed amount; over $20.0B expected if the first tranche is taken (not in the table) | not read |
| Nscale | up to about $44.6B into 2033, payable only as capacity is delivered | not read |
| Lambda | about **34.9B) is now terminable by either side because Lambda did not secure its facility lease in time. Neither side has given notice; Anthropic does not currently intend to terminate | not read |
- What changed between the readings: the Google and Microsoft minimums rose by about $9B combined, which fits NLW’s “10 billion in new commitments with Microsoft and Google”. The filing says both agreements were amended through October and November.
- Other items: RSU vesting was moved so a November 20 service condition is met on November 19, the day after the expected pricing. The prospectus uses “AI” and “SI” (super intelligence) interchangeably, citing Executive Order 14434 of September 29, 2026. “Ode with Anthropic” is an AI services company launched with private-equity firms for mid-sized companies.
The roadshow and the price debate (added 2026-11-13)
Sources: Benzinga’s relay of the FT (ai-research/yahoo-benzinga-anthropic-ipo-could-top-1-5-trillion-2026-11-13.md); The AI Daily Brief (raw/Wall_Street_Has_No_Idea_What_Anthropic_is_Worth.md); All-In and Moonshots ep. 311 as noted.
- Above the range? (FT): underwriters are considering pricing above 1.5T and proceeds above $100B. Goldman Sachs and Morgan Stanley lead and began meeting investors on November 10. Five people told the FT demand was particularly strong; three backers said Anthropic cares more about raising money for compute than maximising valuation.
- Australian retail offer: open November 11-17 under a prospectus filed with Australia’s regulator. On November 10 banks briefly received details of a $69.3B raise that Anthropic withdrew as inaccurate.
- Forecasts: Dan Ives, near 1.2T; John Ma (Artemis, via NLW) says the consensus is “overvalued at 1.5 trillion” but sets out a contrarian case in which retail demand pushes the market cap to $3T. Opinions, not filing data.
- Multiples (Sonali Basak, EY Capital, via NLW): about 40× trailing 12-month revenue at 35.3B, which supports the 40×.
- NLW’s view (opinion): the 1.6T and 70-78 range is not a retreat from a target; “investors do not have a clear idea of what to expect.”
- Amodei’s pitch: “an age of miracles”, with risks treated as problems to solve on the way. To Bloomberg’s Emily Chang he said military decisions are “not ours to make” apart from “very high-level boundaries” (NLW’s paraphrase).
- All-In and Moonshots, same week: Jason Calacanis, citing Reuters, gives 825M shares at 1.4T, and a “total raise” of 64B of share sales plus the $26B placement), which matches the filing’s top-of-range shares plus placement. ^[inferred] Chamath Palihapitiya says the book is “six times over subscribed already” (unverified; no other source gives a multiple). Emad Mostaque (Moonshots ep. 311) says the IPO “is next week” and expects it to put “so much liquidity in the ecosystem” (
raw/Jensen_OUT_as_Nvidia_CEO_-_Spy_Scandal_Export_Restrictions_Market_Impact.md;raw/Jensen_s_Departure_NVIDIA_s_succession_plan_AI_doomer_response_and_Demis_talks_RSI_311_Live.md). - Lambda as a signal (NLW): the missed deadline raises the question of whether AI demand falls short first or neocloud financing and construction does. See The Compute Economics of the AI Buildout.
More from the filing, read on air in November (added 2026-11-09)
- Ownership caps (AIS): Amazon cannot hold more than 12% of Anthropic’s stock and Google no more than 14.99%. Current stakes are not disclosed.
- Risk factors, per Bloomberg’s summary and Roetzer’s own reading (AIS): catastrophic or misaligned AI behaviour; insurance that “may prove insufficient”; securing enough compute; government scrutiny; data-center opposition; hackers; TSMC and the chip supply chain, including a China–Taiwan conflict; and company culture: keeping mission-driven hiring as the company scales, and the risk that employees made wealthy by the IPO leave once they can sell, possibly to start competitors. AIS counts “over 90 pages” of risk factors, matching ADB.
- How Anthropic describes its product: the S-1 introduces “virtual collaborators” (capitalised in the filing): AI systems that “carry the access rights, system connections, and context that their organizations grant them” and do multi-step professional work. It describes Claude Cowork as extending Claude Code’s agentic architecture to knowledge workers, and aims for Claude to take on open-ended tasks “lasting hours, days, or weeks”, ending with “a country of geniuses in a data center”.
- Consumption, not seats: about 80% of revenue is consumption-based (Kaput’s reading). See Enterprise AI Spend, November 2026.
- Revenue per employee above 700,000 a year (AIS). What the revenue-per-employee figure means for other employers is in AI and Work, Early November 2026.
- Internal use (AIS; Moonshots ep. 308): the value of internal Claude usage at list prices grew more than 250× from January to September 2026, and first exceeded monthly personnel expense in April 2026. Moonshots adds that more than 90% of September 2026 code contributions were authored by Claude, and that daily code contributions per engineer were 45× what they were two years earlier. On ep. 309, Blundin recalled the figure as “235x in the last 6 months”.
- Customer counts, confirmed again (AIS): just under 12,000 customers at 1M+, and about 250 at $10M+.
- The $100M Claude Frontier Academy (10,000 engineers by end-2027) is read the same way by AIS. Roetzer’s critique of training only engineers is in AI and Work.
Try It
- Budget for model turnover, not a fixed model. More than 80% of Anthropic’s revenue runs on models less than six months old, and the filing says welfare considerations may delay deprecations. Plan re-evaluations around each release rather than pinning one model for a year.
- Price your own AI work per task, not per token. The filing’s evidence for growth is cost-per-task falling (Opus 5.5 about −40%) while total spend rises. Track cost per completed task in your own workflows to see which side of that curve you are on. See Token Economics Primer.
- If you sell deployment services, read the customer data as a market map. Big buyers spend under 0.1% of expenses on Anthropic, deployments are concentrated in a few teams, and Anthropic is funding a 10,000-engineer academy. That marks the size of the gap and the incumbent’s plan to fill it. See Forward-Deployed Engineer Playbook.
- Buy through your existing cloud contract where it helps. 48% of Anthropic revenue already comes through cloud partners, so procurement through an existing AWS or Google Cloud commitment is the common route.
- Cite the filing section, not the shows, where it exists. Since 2026-11-13 this article has the amended filing’s own figures for revenue, cash, concentration and commitments; use those over the on-air readings.
Open Questions
- H1 revenue: 6.3B? Four shows say 6.3B. It is probably a slip or a caption error, but this is unconfirmed. ^[ambiguous] 2026-11-09: AIS makes it five shows for 1.2B a year earlier; the $6.3B reading is very likely a slip.
- Cash on hand: 125B “entering June” (Chamath), 110B (Mostaque) or “about 100 billion” (Sacks)? The measurement dates may differ. Resolved for September 30 (2026-11-13): the S-1/A gives 36.3B short-term investments, $131.9B together, which matches ADB.
- Commitments: 600B signed (ADB), against “more than 500 billion” (Moonshots) and the 450B figures recorded in October. These need reconciling against the filing’s own table. Mostly reconciled 2026-11-09: AIS’s contract list sums to 551.1B with SpaceX (see “Compute commitments, contract by contract”). ^[inferred] Updated 2026-11-13: the S-1/A totals about 44.6B) and Lambda (about 895.4B figure does not appear in the fetched filing text; it may come from a different table or an earlier draft. ^[ambiguous] Still open: how the Broadcom lease (about 35B and $63B Broadcom/Apollo/Blackstone vehicles.
- Intelligent Machines’ figures (added 2026-11-09): the host relays Anthropic “quarterly revenue… 17 billion”, “12x year-over-year”, a “23 billion" loss, and spending commitments that "have ballooned to 900 billion, 70 billion of which is due in the next 12 months" (`raw/Unsustainably_High_-_Tech_Billionaires_vs._Voters.md`). None of these matches the filing figures above (16.3B for the half year; 17B of Q3 revenue and 23 billion” loss and “12x” are still unmatched. 2026-11-13: the filing gives Q3 revenue up 1,127% (about 12×) and “about 23 billion” loss is unexplained.
- Concentration: “two customers about 20%” (Moonshots) against “top five 21%” (ADB). Both cannot be true unless the other three top customers are tiny. One reading is probably misstated. ^[ambiguous] Partly resolved 2026-11-13: the S-1/A says no single customer exceeded 10% of nine-month revenue, the top five were about 25% of nine-month revenue, and about 79% of Q3 revenue came from outside the top five, so ADB’s 21% is a Q3 figure. Moonshots’ H1 claim is not contradicted by the filing but is not supported either.
- **The SpaceX 1.25B a month recorded in the SpaceX compute deal, and to All-In’s earlier claim that the deal can be cancelled within 60–90 days, is unresolved. Partly resolved 2026-11-09: AIS reads the filing as allowing either party to cancel on 90 days’ notice after lock-ins ending in January and March 2027, which supports All-In. The link to the $1.25B a month is still open.
- The 7% banned-accounts figure (Mostaque) is not repeated by any other show.
- Cowork growing faster than Claude Code is NBJ’s reading, and no growth figure is quoted.
- Run rate: about 80B in October (WSJ via ADB), “close to 100 billion” (Ismail, Moonshots), and Reuters’ earlier expectation of $110B by year end. The first is reported, the second rounded on air, the third a forecast.
- Internal Claude usage growth: more than 250× from January to September (AIS, Moonshots ep. 308) against “235x in the last 6 months” (Blundin, Moonshots ep. 309, from memory). The 250× figure has two readings behind it.
- “Amended S-1”: Wissner-Gross on Moonshots ep. 308 refers to Anthropic “filing its amended S1”. No other show mentions an amendment, and whether the internal-usage figures come from an amendment is unknown. 2026-11-13: Amendment No. 2 was filed November 10, so at least one earlier amendment exists; the internal-usage figures were not among the passages read.
- “Autonomy level two” (added 2026-11-11): Mostaque (Moonshots ep. 310) says the S-1 reports this milestone. It needs checking against the filing; it may correspond to a level in Anthropic’s Responsible Scaling Policy rather than a new scale. ^[inferred] 2026-11-13: a keyword search of Amendment No. 2 found no “autonomy level” or “level two” outside fair-value accounting; “varying levels of autonomy” appears only in the definition of AI. The claim is unsupported by the amended filing.
- Retail allocation: CNBC’s sources say about 20%; the S-1/A mentions a minority allocation to retail investors but no percentage was found in the passages read.
- The final price is not set as of 2026-11-13; the FT’s above-range report rests on unnamed sources.
- Speaker names are from auto-captions. “Tony Ziringer” (Impax) and “Boxer and Levy” (quoted by ADB on the spend headroom) are as captioned and could not be checked.
Related
- Anthropic’s Position, August 2026 — the pre-filing record: leaks, the confidential draft, the Q3 investor day, the Public Benefit Fund
- Anthropic’s $50B Buildout (Epoch AI) — the first Broadcom/Apollo TPU vehicle, before the $63B follow-on
- Anthropic + SpaceX Compute Deal — the SpaceX commitment the filing now sizes
- The Compute Economics of the AI Buildout — supply side, neocloud credit stress
- AI Bubble Signals — the market backdrop, including the November 5 sell-off that put the IPO price in question
- AI and Work, Early November 2026 — revenue per employee and the Frontier Academy critique
- Claude Opus 5.5 — the 40% cost-per-task cut the filing cites
- Claude Model-Welfare Policy — the context for the deprecation risk factor
- AI News, November 11-13, 2026 — the roadshow week in context