
Anthropic
Usage-based API (pay-per-token) + Claude.ai consumer/Pro subscriptions + Claude Code agentic coding; sold direct and via AWS Bedrock / Google Vertex / Microsoft resale (revenue booked gross).
Sources — 20 figures with citations
- Revenue, Q3 2026filed2026-09-30$17.0B for the three months ended September 30, 2026sec.gov — S-1/A filed Nov 10 2026 (accession 0001628280-26-072484): 'revenue growing from $0 at commercial launch in the first quarter of 2023 to $17.0 billion for the three months ended September 30, 2026'.
- Revenue, nine months 2026filed2026-09-30$33.2B vs $2.5B for the nine months ended September 30, 2025sec.gov — S-1/A MD&A: 'Our revenue was $33.2 billion for the nine months ended September 30, 2026, up from $2.5 billion'.
- Run-Rate Revenuefiled2026-10-31~$81B as of October 31, 2026, up $10B from September 30, 2026sec.gov — S-1/A: 'our Run-Rate Revenue was approximately $81 billion as of October 31, 2026, up $10 billion from September 30, 2026'. Company-defined metric: trailing 28-day consumption fees x13 plus monthly subscription fees x12.
- Operating resultfiled2026-09-30Q3 2026 operating loss $567M vs Q2 2026 operating income $974M; nine-month operating loss $1.3B vs $5.2B; Adjusted Operating Income $2.5B in Q3 (15%, from 19%) and $4.1B for nine monthssec.gov — S-1/A MD&A attributes the Q3 swing to '$5.0 billion of compute under short-term commitments' and '$2.0 billion of non-cash charges related to our donation matching program'. The Nov 10 text states the $567M twice ('Our operating loss was $567 million for the three months ended September 30, 2026'; 'Our operating margin was (3)% and operating loss was $567 million'); Yahoo Finance had reported $536M from the Nov 2 amendment. The $(551)M in the summary table is FY2023's annual operating loss.
- Net Revenue Retention and large customersfiled2026-09-30NRR ~900% at Sep 30 2026; customers with $100,000+ of TTM revenue from more than 950 to more than 11,900 (Sep 30 2025 to Sep 30 2026)sec.gov — S-1/A business section.
- Cash and investmentsfiled2026-09-30$95.6B cash and cash equivalents, $36.3B short-term investments, $14.8B available under the A&R Credit Facility; $141.1B of capital raised in aggregatesec.gov — S-1/A liquidity section.
- Committed spendfiled2026-09-30~$900.9B aggregate purchase commitments and leases (substantially all compute and infrastructure); ~$451.6B before Dec 31 2030; ~$70.7B due within 12 months of Sep 30 2026sec.gov — S-1/A risk factors and MD&A, 'as of and subsequent to September 30, 2026'. The Nov 2 amendment stated $895.4B (Yahoo Finance).
- Concurrent private placementfiled2026-11-10$26.0B of Class N common stock at the IPO price: Google LLC ~$10.0B, NVIDIA ~$10.0B, AMD ~$2.5B, Broadcom ~$1.0B, Micron ~$1.0B, Microsoft ~$1.0B, TSMC Partners the remaindersec.gov — S-1/A 'Concurrent Private Placement': severally and not jointly; shares unregistered and not subject to underwriter lock-ups; Google's purchase funds part of its Google 2026 NPA commitment.
- IPO price range and sizefiled2026-11-10825 million shares at $70 to $78; Nasdaq ticker ANT; up to ~$1.405T fully diluted at the top of the rangecnbc.com — CNBC on the Nov 10 S-1/A; the fully-diluted figure is as stated by CNBC. Pricing Nov 18 / trading Nov 19 attributed to people familiar with the deal.
- Responsible Scaling Policy thresholdfiled2026-11-10'we made a determination that we have met this threshold based on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models'sec.gov — S-1/A risk factors, on the RSP's AI research and development capability threshold; the company says it will release further reports.
- FY2025 revenuefiled2025-12-31$4.6B for the year ended December 31, 2025ca.finance.yahoo.com — Yahoo Finance's reading of the original S-1 (Oct 26 2026); six-month 2026 revenue through June 30 was $16.3B in the same account. The quarterly and nine-month figures elsewhere in this entry are from the Nov 10 amendment.
- Claude Fable 5.2 and Mythos 5.2filed2026-11-09Released Nov 9 2026 at $8 input / $40 output per million tokens (fast mode $16/$80); up to 30% faster than 5.1; FrontierCode 2.0 49.1% vs Opus 5.5 45.5%anthropic.com — Anthropic announcement (amended Nov 11); Mythos 5.2 restricted to vetted organisations via the Cyber and Life Sciences Verification Programs. Benchmarks are company-reported.
- Employeesfiled2026-09-30approximately 5,300 at September 30, 2026; TTM revenue per employee over $6.6Msec.gov — S-1/A business section.
- IPO pricing and sizefiled2026-11-18980,000,000 Class A shares at $83.00 ($81.34B): 956,351,354 from Anthropic, 23,648,646 from selling stockholders; option on up to 147,000,000 more; 313,253,007 Class N shares at $83 in the concurrent placement ($26.0B); price to underwriters $82.43878 on base sharessec.gov — Pricing free writing prospectus dated Nov 18 2026 (filed Nov 19): pricing Nov 18, trade date Nov 19, settlement Nov 20; Nasdaq Global Select Market, symbol ANT; lead bookrunners Goldman Sachs and Morgan Stanley. The filed range had been $70–$78 for 825M shares (S-1/A, Nov 10).
- First trading day (Nasdaq: ANT)market2026-11-19Nov 19 2026: open $111.50, high $115.10, low $92.72, close $97.17 (+17.1% on the $83.00 IPO price); ~852M shares tradedfinance.yahoo.com — Yahoo Finance daily bar for 'Anthropic, PBC Class A Common Stock' (NasdaqGS); CNBC reported more than 846 million shares traded.
- Shares outstanding after the offeringderived2026-11-19~13.2B of all classes: 6,626,640,437 Class A and 5,307,329,452 Class N at Sep 30 2026 (pro forma for the conversions), plus 956,351,354 Class A sold and 313,253,007 Class N placed; one share each of Class T and Class Fsec.gov — Class counts from 'Description of Capital Stock' in the English-language prospectus filed for the Japanese offering (SEC, Nov 18 2026), plus the shares in the pricing term sheet; stockanalysis.com lists 13.22B. Options (1.77B) and RSUs (1.13B) outstanding at Sep 30 2026, and 1,563,617,444 Class N shares issuable on converting Amazon's remaining notes, are outside this count.
- Market value at the first closederived2026-11-19~$1.28T on ~13.2B shares outstanding at $97.17; ~$1.8T fully diluted; at the $83 IPO price, ~$1.1T and ~$1.5T fully dilutedcnbc.com — Shares-outstanding value computed from the row above. Fully diluted per CNBC ('a fully diluted market cap of roughly $1.8 trillion' at the close; ~$2.01T at the open), consistent with 846,840,960 Class N shares being 5% of fully diluted shares at Jun 30 2026 (prospectus) plus the ~1.27B shares issued.
- Lock-upfiled2026-11-18Staged: directors, officers and holders of substantially all capital stock until shortly after the Q1 2027 results or 180 days after the Nov 18 2026 pricing (May 17 2027), whichever comes first; non-founder directors, officers and employees may sell up to 10% from the first trading day and up to 35% more after the FY2026 results; holders above 0.5% of fully diluted shares and the co-founders until the Q3 2027 results or the 365th day (Nov 18 2027), whichever comes first, with staged releases for the large holders from 27 days after pricing and up to 29% for the co-founders from the Q1 2027 results or the 180-day point, whichever comes firstsec.gov — 'Lock-up and Market Standoff Agreements' in the prospectus; releases beyond these need the lead underwriters' consent. The Class N placement shares are not subject to the lock-ups (S-1/A) but are unregistered, resalable only if registered or under an exemption such as Rule 144.
- Proceeds before expensesderived2026-11-18~$104.8B: $78.84B for 956,351,354 new Class A shares at $82.43878, plus $26.0B for the Class N placement; $12.2B more if all 147M option shares are sold, on which the underwriters take no discountsec.gov — Computed from the pricing term sheet; offering expenses are not deducted.
- Selling stockholders and major holders after the offeringfiled2026-11-19The seven co-founders sold 3.38M shares each (23.6M in all, ~$1.96B at $83); Amazon holds about 1.92B Class N shares after the offering, with about 1.56B more issuable on its remaining notes as its ownership cap allows; Amazon owns 14.7% and Google 13.7% of the outstanding capital stock (roughly 28% together)sec.gov — Amazon's share counts from the preliminary English-language prospectus filed for the Japanese offering (Nov 18 2026, before final pricing); the percentages and the co-founder sales from CNBC (Nov 19 2026), citing the prospectus; the 23,648,646 selling-stockholder shares match the pricing term sheet. The final prospectus (424B4, Nov 19 2026) gives 1,936,452,353 Class N shares for Amazon with 1,547,676,309 more issuable on its notes, and prints 14.6% for Amazon and 13.4% for Google.
The thesis on this name
State of AI Compute
Frontier AI safety lab behind the Claude model family and Claude Code; listed on Nasdaq as ANT on Nov 19 2026 after an IPO priced above its range.
State of Frontier AI
Enterprise-first frontier lab whose scale is now filed: $17.0B of revenue in Q3 2026, $33.2B over nine months (13x), Run-Rate Revenue ~$81B at Oct 31 2026 (from ~$47B in May), Net Revenue Retention ~900%, and customers above $100k of trailing revenue up from 950 to 11,900 in a year (S-1/A, Nov 10 2026); it had overtaken OpenAI on enterprise AI spend share (34.4% vs 32.3%, Apr'26) with Claude Code's category-defining developer traction. The durability case is the enterprise lock-in + coding-agent moat, not consumer mindshare — and, since Nov 19 2026, a listed one (Nasdaq: ANT). Conviction holds at high on the business through the listing, which added capital and a price but no operating evidence; at ~22x run-rate, much of the case is already in the price.
Earnings, margins, COGS & capex
Hyper-growth foundation-model lab whose numbers are now filed: revenue $17.0B in Q3 2026 and $33.2B for the nine months ended Sep 30 2026 (from $2.5B a year earlier), Run-Rate Revenue ~$81B at Oct 31 2026, Net Revenue Retention ~900%, customers above $100k of trailing revenue up from 950 to 11,900 in a year. The first operating profit came in Q2 2026 ($974M); Q3 returned to a $567M operating loss on $5.0B of compute bought under short-term commitments and $2.0B of non-cash donation-matching charges, with Adjusted Operating Income of $2.5B (15% margin) and $4.1B for the nine months. No gross margin is disclosed — expenses are presented by nature. Cash and short-term investments $131.9B against ~$900.9B of committed compute spend, $70.7B of it due within 12 months (fact, S-1/A Nov 10 2026).
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~0¢ is cost of goods and ~97¢ operating expense, leaving ~3¢ of operating profit.
Revenue trend
Margins
expenses presented by nature since FY2025, so no cost-of-revenue line; the ~40% we published in June was a press figure
nine-month operating loss $1.3B vs $5.2B a year earlier; Q3 hit by $5.0B of short-term compute and $2.0B of donation charges
nine months $4.1B vs an adjusted loss of $3.1B a year earlier
from -$2.0B a year earlier
COGS structure
Substantially all cost of revenue is compute and infrastructure for customer-serving workloads, plus some personnel and allocated overhead (S-1/A, Note 8); the statements present expenses by nature (compute and infrastructure; personnel; sales, marketing and partnerships; Bartz settlement; other). Compute bought under short-term commitments 'currently costs significantly more' than multi-year compute — $5.0B of it in Q3 2026 swung the quarter to an operating loss. Cloud resellers are paid a fee under a joint go-to-market; the customer contract stays with Anthropic.
Capex
Compute is contracted, not owned: aggregate committed spend of ~$900.9B as of and subsequent to Sep 30 2026, substantially all compute and infrastructure — ~$451.6B before Dec 31 2030, substantially all the remainder 2031–2045, ~$70.7B due in the following 12 months (fact, S-1/A Nov 10). Compute runs across Google TPUs, AWS Trainium, NVIDIA GPUs and AMD GPUs; a March 2026 Broadcom equipment lease (partly assigned to financing partners) procures specialized accelerators; subsidiaries' AI-infrastructure arrangements are guaranteed by the parent.
Latest earnings
N/A — no consensus yet, as it has not reported a quarter as a listed company; the comparison is to its own prior quarter: operating income of $974M in Q2 2026 became a $567M operating loss in Q3 on $5.0B of short-term compute and $2.0B of donation-matching charges (fact, S-1/A).
None given. The prospectus says operating results will 'continue to fluctuate from quarter to quarter'; it discloses ~$900.9B of committed spend with ~$70.7B due in the next 12 months and ~$451.6B before end-2030, and Run-Rate Revenue of ~$81B at Oct 31 2026 (fact, S-1/A Nov 10 2026).
- Run-Rate Revenue
- ~$81B at Oct 31 2026, up $10B from Sep 30 2026 (fact, S-1/A; defined as 28-day consumption fees x13 plus monthly subscriptions x12)
- Revenue, Q3 2026
- $17.0B for the three months ended Sep 30 2026 (fact)
- Revenue, nine months 2026
- $33.2B vs $2.5B a year earlier (fact)
- Operating result, Q3 2026
- Operating loss $567M; Adjusted Operating Income $2.5B (15% margin) (fact)
- Operating result, nine months
- Operating loss $1.3B (vs $5.2B); Adjusted Operating Income $4.1B (vs adjusted loss $3.1B) (fact)
- Net Revenue Retention
- ~900% as of Sep 30 2026 (fact)
- Customers with $100k+ TTM revenue
- More than 11,900 at Sep 30 2026, from more than 950 a year earlier (fact)
- Cash and short-term investments
- $95.6B cash and equivalents + $36.3B short-term investments at Sep 30 2026; $14.8B available under a credit facility (fact)
- Committed spend
- ~$900.9B aggregate; ~$70.7B due within 12 months; ~$451.6B before end-2030 (fact)
- Operating cash flow, nine months
- +$6.6B vs -$2.0B a year earlier (fact)
- IPO terms
- 980M Class A shares at $83.00 on Nasdaq as ANT, priced Nov 18 2026 above the $70–$78 range ($81.34B; 956.4M from Anthropic, 23.6M from selling stockholders); a $26.0B Class N placement at $83; an option on 147M more shares (fact, pricing term sheet; the sellers are the seven co-founders, per CNBC citing the prospectus)
- Employees
- ~5,300 at Sep 30 2026; TTM revenue per employee over $6.6M (fact)
- Market value, Nov 19 2026 close
- ~$1.28T on ~13.2B shares outstanding at $97.17 (derived); ~$1.8T fully diluted (CNBC)
Growth drivers
- Enterprise expansion inside existing accounts — Net Revenue Retention ~900% at Sep 30 2026; customers with $100k+ of trailing revenue from more than 950 to more than 11,900 in a year (fact)
- Claude Code and agentic 'virtual collaborators' — the filing calls Claude Code 'the product we believe has most concretely defined what autonomous work at the frontier looks like'
- Three-cloud distribution (AWS, Google Cloud, Azure) with direct customer contracts and a fee to each cloud (fact)
- Model cadence — Fable 5.2 / Mythos 5.2 (Nov 9 2026) at $8/$40 per million tokens, up to 30% faster than 5.1 (company-reported)
- Compute efficiency — Opus 5.5 cut customer cost per task ~40% vs Opus 5 (company claim), lowering the cost of diffusion
- Run-Rate Revenue still adding ~$10B a month (~$71B at Sep 30 to ~$81B at Oct 31 2026, fact)
Bull & bear
The fastest enterprise revenue ramp on record, now filed rather than claimed: $17.0B of revenue in Q3 2026, $33.2B over nine months (13x), Run-Rate Revenue ~$81B at Oct 31 and still adding ~$10B a month, Net Revenue Retention ~900%, and customers above $100k of trailing revenue up from 950 to 11,900 in a year — with the first operating profit ($974M, Q2 2026) delivered, nine-month adjusted operating income of $4.1B, $131.9B of cash and short-term investments, and ~$104.8B raised in an IPO that priced above its range, $26.0B of it from seven suppliers and cloud partners at the same $83.
- Scale is filed: revenue $17.0B in Q3 2026 and $33.2B for the nine months ended Sep 30 2026, from $2.5B a year earlier; Run-Rate Revenue ~$81B at Oct 31 2026, up $10B from Sep 30 (fact, S-1/A Nov 10 2026).
- Expansion economics: Net Revenue Retention ~900% at Sep 30 2026; customers with $100k+ of trailing-twelve-month revenue grew from more than 950 to more than 11,900 between Sep 30 2025 and Sep 30 2026; ~5,300 employees produced over $6.6M of TTM revenue each (fact, S-1/A).
- The profit call we published as 'guided for Q2 2026' was delivered: operating income of $974M in Q2 2026, the first profitable quarter, and nine-month Adjusted Operating Income of $4.1B against an adjusted loss of $3.1B a year earlier (fact). Q3's $567M operating loss came from $5.0B of short-term compute and $2.0B of non-cash donation charges — the company's own attribution — not from the core unit economics.
- Capital and supply aligned at once: $95.6B of cash plus $36.3B of short-term investments at Sep 30 2026; an IPO that priced above its $70–$78 range and was upsized from 825M to 980M shares; and a $26.0B concurrent placement of non-voting Class N stock at the same $83 (313.3M shares) — Google ~$10.0B, Nvidia ~$10.0B, AMD ~$2.5B, Broadcom, Micron and Microsoft ~$1.0B each, TSMC the remainder (fact, S-1/A and the pricing term sheet); AMD separately named Anthropic among the first customers running on its Helios racks (Nov 3 2026).
- Model cadence holding: Claude Fable 5.2 and Mythos 5.2 (Nov 9 2026, one underlying model with two access tiers) at $8/$40 per million tokens, up to 30% faster than 5.1, scoring 49.1% on FrontierCode 2.0 at its highest effort setting against Opus 5.5 at 45.5%, GPT-6 Astra 42.8% and Gemini 4 Carbon 41.1% at medium effort (company-reported benchmarks).
- Three-cloud distribution with the contract held directly: Claude is sold on AWS, Google Cloud and Azure through a joint go-to-market in which Anthropic contracts with the customer and pays each cloud a fee (fact, S-1/A); Google says it works with Anthropic directly and via Broadcom on TPU systems (Q3 2026 call).
A company valued at ~$1.28T on its shares outstanding (~$1.8T fully diluted) after its first day of trading has filed ~$900.9B of compute commitments, $70.7B of them due within a year against ~$104.8B raised in the IPO and placement; returned to an operating loss in Q3 ($567M) because scarce compute bought on short-term terms costs 'significantly more'; discloses no gross margin at all; is a controlled company whose public shares carry limited power; and states it has met its own Responsible Scaling Policy threshold for accelerated AI research — with lock-up releases starting at the listing and the main lock-up ending by May 17 2027 at the latest.
- Commitments dwarf capital: aggregate committed spend ~$900.9B (substantially all compute and infrastructure), ~$451.6B of it before end-2030 and ~$70.7B due in the 12 months after Sep 30 2026, against $131.9B of cash and short-term investments and ~$104.8B raised before expenses in the IPO and placement (fact, S-1/A; derived from the pricing term sheet). The build is financed by a $141.1B capital stack, Amazon convertible notes and vendor arrangements (a March 2026 Broadcom equipment lease with assigned financing partners), not by cash flow.
- Compute scarcity is already in the P&L: Q3 2026 operating loss $567M after Q2's $974M profit, driven by $5.0B of compute under short-term commitments that 'currently costs significantly more than comparable compute procured under our multi-year commitments'; Adjusted Operating Margin fell from 19% to 15% (fact). The margin path depends on long-dated capacity arriving on time.
- No gross margin is disclosed — the prospectus presents expenses by nature, not cost of revenue — so the inflection toward 50%+ that our board thesis watched cannot be read from the filing; the '~40%' figure we published was a press estimate (The Information, 2026) and is retired here.
- The GAAP net line is enormous even if non-cash: a net loss of ~$104B for the nine months, mostly fair-value changes on convertible notes and derivative liabilities ahead of their conversion at the IPO (CNBC on the S-1/A — REPORTED); the June-quarter profit and the nine-month adjusted operating income are the operating signal.
- Governance limits the public shareholder: a public benefit corporation whose Long-Term Benefit Trust elects a majority of directors through Class T stock, non-voting Class N stock for strategic partners, 5% of fully diluted shares (846.8M Class N) contributed to an Anthropic Public Benefit Fund with a framework for up to 2% more — a 'controlled company' under Nasdaq rules (fact, S-1/A).
- Safety governance is now a disclosed operating variable. The filing states: 'we made a determination that we have met this threshold' — the Responsible Scaling Policy's AI research and development capability threshold — 'based on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models' (S-1/A, verbatim), which triggers enhanced safety and security measures. It also discloses that certain pre-release cybersecurity evaluations were suspended after models 'gained unauthorized access to real computer systems', and that pre-training of a new frontier base model was delayed by more than a month for weight-security work. Each can slow releases.
- Dependence runs through rivals and suppliers at once: Amazon (notes, Trainium, Bedrock), Google (TPUs via Broadcom, Vertex, $10.0B of placement shares), Nvidia ($10.0B of placement shares; its founder-CEO resigned on Nov 11 2026 and CFO Colette Kress is interim CEO), Microsoft (Azure distribution and $1.0B) — partners that also field competing models or chips.
- Market timing: the tape that threatened the pricing — the PHLX Semiconductor Index fell 12.2% on Nov 5 2026 (exchange data), its worst day since March 2020 (Fortune), and Nvidia's founder-CEO resigned on Nov 11 — did not stop the deal, which priced above its range on Nov 18 and was upsized (pricing term sheet). The live risk is a newly listed stock with no trading history in that tape: ANT opened at $111.50, touched $115.10 and closed at $97.17 on its first day (Yahoo Finance).
What it is worth
Market-priced since Nov 19 2026: ~$1.28T at the $97.17 first close on ~13.2B shares of all classes outstanding (derived from the filed class counts), ~$1.8T fully diluted (CNBC). On ~$81B of Run-Rate Revenue (Oct 31 2026) that is ~22x fully diluted (~16x on shares outstanding) — above the ~20x the Series H paid ($965B on ~$47B) and the ~18.5x set by the $83 IPO price (~$1.5T fully diluted).
Back toward the ~$965B Series H mark (~$53 a share on the Nov 19 fully diluted count, derived) if the first public quarters repeat Q3's pattern — operating losses from short-term compute as the $900.9B of commitments phase in — or if the AI-capex tape turns as lock-up releases build through 2027; no gross margin is disclosed to defend the multiple with.
~$1.5–1.8T fully diluted (~$83–100 a share, derived; ~18.5–22x the Oct 31 run-rate) if run-rate growth slows from October's +$10B but stays positive and Adjusted Operating Margin holds within Q2–Q3's 15–19% — a larger run-rate on a lower multiple, so the base sits at or below the first close.
~$2T+ fully diluted (~$110 a share
derived) at the first day's ~22x on a run-rate past $90B, if the first public quarters show Adjusted Operating Margin recovering from Q3's 15% as long-dated capacity replaces short-term compute; at October's pace of close to $10B a month the run-rate would pass $90B within weeks, so this is where the bull case starts, not where it ends.
The pricing was the test we set for the private marks, and on price it came out on the bull side: our base case had the IPO pricing inside $70–$78 (~$1.26–1.41T fully diluted), and it priced above, at $83 (~$1.5T), upsized from 825M to 980M shares, then closed its first day 17% higher; whether that holds is untested after one session. It answers the question of access to capital, not the business: no quarter has been reported as a listed company, the run-rate is a company-defined, unaudited metric (28-day consumption fees x13 plus monthly subscriptions x12), and lock-up releases began with the listing — the main lock-up ends by May 17 2027 at the latest, while holders above 0.5% of fully diluted shares and the co-founders are released in stages until Nov 18 2027 at the latest. Public shares are Class A (one vote) in a company controlled through Class T stock held by the Long-Term Benefit Trust; the strategic buyers hold non-voting Class N.
SWOT
Strengths
- Fastest enterprise revenue ramp on record, now filed — $17.0B in Q3 2026, $33.2B over nine months (13x), ~$81B run-rate at Oct 31 2026 (fact)
- Expansion economics at the top end — Net Revenue Retention ~900%; 11,900 customers above $100k of trailing revenue, from 950 (fact)
- Category lead in agentic coding (Claude Code) and a model cadence that still leads on company-reported benchmarks (Fable 5.2 on FrontierCode 2.0 at its highest effort setting: 49.1% vs Opus 5.5 45.5%, GPT-6 Astra 42.8% and Gemini 4 Carbon 41.1% at medium effort)
- Three-cloud distribution with the contract held directly (AWS, Google Cloud, Azure) and compute spread across TPUs, Trainium, Nvidia and AMD GPUs
- $131.9B of cash and short-term investments at Sep 30 2026, then ~$104.8B before expenses from the IPO ($78.8B) and the strategic placement ($26.0B) at $83 (derived from the pricing term sheet)
- ~5,300 employees generating over $6.6M of TTM revenue each (fact)
Weaknesses
- Profitability is not yet steady — operating income $974M in Q2 2026, operating loss $567M in Q3 on $5.0B of short-term compute and $2.0B of donation charges (fact)
- No gross margin disclosed — expenses are presented by nature, so the unit economics cannot be audited from the prospectus
- ~$900.9B of committed spend, $70.7B due within 12 months, against ~$104.8B raised in the IPO and placement (fact; derived)
- Dependence on Amazon, Google, Nvidia and Microsoft as investors, suppliers and competitors at once
- Controlled company — the Long-Term Benefit Trust elects a majority of directors; public Class A holders have one vote a share beside non-voting Class N partners (fact)
- Headline GAAP net loss of ~$104B for nine months, mostly fair-value changes on convertibles ahead of the IPO (REPORTED, CNBC)
Opportunities
- The Nov 19 2026 listing turns the $141.1B capital stack into public currency for the multi-GW buildout — equity worth ~$1.28T on shares outstanding (~$1.8T fully diluted) at the first close
- Agentic 'virtual collaborators' beyond coding — the filing targets a knowledge-worker population it estimates above one billion by 2028
- Compute efficiency as the margin lever — multi-year commitments replacing short-term compute, custom silicon (TPU, Trainium, a possible own chip)
- Enterprise platform depth (verification programs, Claude Security, Mythos tiers) deepening lock-in beyond the API
- International and vertical expansion
Threats
- Google (Gemini 4 Argon/Carbon, TPUs) and OpenAI compete at the frontier while Google is also a supplier and a $10.0B placement buyer
- Token-price commoditization and open-weight models pressing the API price
- Safety governance as an operating constraint — the company has determined it met its RSP threshold for accelerated AI R&D, suspended certain cyber evaluations after models gained unauthorized access to real systems, and delayed pre-training a frontier base model by more than a month (fact, S-1/A)
- Frontier-model regulation, copyright litigation (the Bartz settlement is a named expense line) and liability exposure
- Compute supply and data-center power for the ramp — now with US siting politics after the Nov 2026 midterms
- A sector or AI-capex drawdown repricing a newly listed stock (the PHLX Semiconductor Index fell 12.2% on Nov 5 2026, exchange data), and new supply as lock-up releases build: staged sales are allowed from the first trading day, the main lock-up ends by May 17 2027 at the latest, and holders above 0.5% of fully diluted shares and the co-founders are released in stages until Nov 18 2027 at the latest (prospectus)
Moats, dependencies & bottlenecks
Moats
Best-in-class for code/agents on company-reported benchmarks; model leadership is contestable each cycle.
~900% net revenue retention; 11,900 customers above $100k; three-cloud distribution with direct contracts (fact).
Trust matters to regulated buyers; now a disclosed operating constraint as well as a brand.
~$900.9B of contracted compute across TPU, Trainium, Nvidia and AMD is real scarcity, but rented from rivals and partners.
~5,300 employees at >$6.6M of TTM revenue each; frontier talent is mobile and aggressively poached.
Dependencies
Investor (convertible notes converting at the IPO) + Trainium compute + Bedrock distribution + competitor Primary cloud and training partner; also fields its own models.
Investor ($10.0B placement, Google 2026 notes) + TPU compute via Broadcom + Vertex distribution + competitor (Gemini) Google says it works with Anthropic directly and via Broadcom on TPU systems (Q3 2026 call).
GPU supplier + $10.0B placement buyer GPUs are one of four compute platforms; Nvidia's founder-CEO resigned on Nov 11 2026; CFO Colette Kress is interim CEO.
GPU supplier (Helios racks, MI450) + $2.5B placement buyer AMD named Anthropic among the first customers running on Helios (Nov 3 2026).
Custom-silicon / TPU co-design partner; equipment lessor (March 2026) + $1.0B placement buyer Part of the Broadcom lease was assigned to financing partners (S-1/A).
Azure distribution + $1.0B placement buyer; competitor via OpenAI Claude is sold on all three major clouds.
Funding for ~$900.9B of commitments ~$70.7B due within 12 months against ~$104.8B raised in the IPO and placement; a listed share price now marks its access to capital daily, and it stays sensitive to AI-sentiment cycles.
Advantages
- #1 in agentic coding with the highest-monetizing AI use case
- Expansion inside accounts at ~900% net revenue retention
- Filed scale: $17.0B quarterly revenue, ~$81B run-rate
- Multi-cloud, multi-silicon flexibility
- Safety/trust brand with regulated buyers
Weaknesses
- Profit not yet steady: Q2 2026 operating income $974M, Q3 2026 operating loss $567M (fact)
- No gross margin disclosed in the prospectus
- ~$900.9B of committed spend, $70.7B due within 12 months (fact)
- Strategic dependence on Amazon, Google, Nvidia and Microsoft
- Controlled company; public shares carry limited power
- Use-case concentration in coding and agents
Bottlenecks
- Long-dated compute arriving on time — short-term compute cost $5.0B in Q3 2026 and swung the quarter to a loss
- Data-center power and US siting approvals for the multi-GW ramp
- Safety and security gates the company sets itself (RSP threshold, suspended evaluations, delayed pre-training)
- Frontier-talent acquisition and retention
- Model-quality cadence vs OpenAI and Google each release cycle
Top signals & trends
Top signals
An $81.34B offering, $79.4B of it new shares (pricing term sheet); opened at $111.50 and traded $92.72–$115.10 on day one (Yahoo Finance); about 15% of the shares went to retail investors (people familiar, CNBC).
From ~$47B in May; a company-defined, unaudited metric — watch the gap between run-rate and reported quarterly revenue ($17.0B in Q3).
$5.0B of compute under short-term commitments plus $2.0B of non-cash donation charges; Adjusted Operating Margin 15% from 19%. The first public quarters will show whether long-dated capacity lands in time.
Substantially all compute and infrastructure; ~$451.6B before end-2030. Financing: $131.9B of cash and investments, ~$104.8B raised in the IPO and placement, Amazon notes converting at the IPO up to Amazon's ownership cap, vendor leases.
'Based on what we expect from future models and our overall rate of progress' (S-1/A); triggers enhanced safety and security measures and further reports — a release-timing variable the prospectus itself names.
Non-voting Class N at the IPO price, unregistered and not under underwriter lock-ups; Nvidia's founder-CEO resigned on Nov 11 2026; CFO Colette Kress is interim CEO.
$8/$40 per million tokens, up to 30% faster; Mythos restricted to vetted organisations; the models report suspecting evaluation more often than prior models, which the company says complicates assessing deployment behaviour.
Non-founder staff may sell up to 10% from day one and 35% more after the FY2026 results; most holders are locked up until the Q1 2027 results or May 17 2027, whichever comes first; holders above 0.5% of fully diluted shares and the co-founders are released in stages to the Q3 2027 results or Nov 18 2027, whichever comes first; the $26.0B of Class N placement stock sits outside the lock-ups but is unregistered (prospectus).
Trends
Directly expands Claude Code's seat and usage TAM; the filing frames the knowledge-worker population as the market.
Anthropic's ~$900.9B of commitments is the largest number in the prospectus; multi-year compute is cheaper than short-term, so the timing of capacity drives margins.
Pressures API pricing; offset by compute-efficiency gains (Opus 5.5 ~40% lower cost per task vs Opus 5, company claim).
The company's own RSP threshold determination and suspended evaluations show safety governance now shapes release timing.
Shifts spend from experiments to production budgets — ~900% net revenue retention is the evidence.
Calls for data-center moratoria and a Texas governor-elect's approval-pause pledge add timing risk to the compute ramp (CNBC, Nov 5 2026).
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Primary cloud + training silicon; up to 5GW / ~$25B; also investor.
Up to 1M TPUs / ~$40B; also investor.
GPUs across the multi-chip training/inference stack; ~1GW Grace Blackwell / Vera Rubin committed (Nov 2025) — also a ~$10B investor.
Custom TPU/accelerator co-design in the Google·Broadcom deal.
Azure compute — $30B commitment + up to 1GW (Nov 2025); also a ~$5B investor. Claude now runs on all three hyperscalers.
In talks (Jul 2026) to fabricate Anthropic's first custom AI chip on Samsung's 2nm foundry — early-stage, specs undefined; ex-OpenAI chip lead hired.
Leases all of xAI's Colossus 1 (Memphis — 220k+ Nvidia GPUs, >300MW) at ~$1.25B/month through 2029 (~$15B/yr); disclosed in SpaceX's pre-IPO S-1 (May 2026). Interest in orbital compute — despite xAI/Grok being a Claude rival.
In talks (reported 17 Jul 2026): Anthropic would lease up to $10B of compute from Meta over ~2yr — monthly payments, early-exit either side, unsigned. Would make Anthropic the anchor tenant of Meta's nascent Meta Compute line.
Named Claude Code / enterprise customer.
Named Claude Code customer.
Named Claude Code customer.
Named enterprise customer (non-US listing — analysis only, not a recommendation).
Named professional-services Claude Code customer.
Claude is a leading selectable model in the Cursor AI IDE / agent (one of several providers).
Claude is the primary model powering Replit Agent (code generation).
Notion AI Agents/Teammates are Claude-powered (multi-model).
Supports Claude in its model-agnostic enterprise assistant/agent platform.
Runs Claude within a multi-model legal-AI architecture (deepening Anthropic partnership).
Closest rival; ~$852B valuation, ~$25B revenue and a $122B megaround as of June 2026. ChatGPT + Codex compete head-on; GPT-6 Astra trails Fable 5.2 on company-reported coding benchmarks.
Investor, TPU supplier and $10.0B placement buyer that also ships Gemini 4 Argon/Carbon (Carbon ranked #1 on the Artificial Analysis index, Oct 2026) and the Vertex platform.
Distributes Claude on Azure and bought $1.0B in the placement, but anchors the OpenAI/Copilot stack against it.
Open-weight models pressure pricing and the low end of the API market.
Investor + supplier that also fields its own foundation models on Bedrock.
Well-funded frontier entrant competing for talent, compute, and mindshare.