Compiled by XAIPART Editorial from public sources on 2026-11-17. Every factual claim is cited to a numbered source; forward-looking judgments are labeled XAIPART analysis.
Scope. This report covers Anthropic, PBC, the developer of the Claude model family, as it approaches its initial public offering. It draws on Anthropic's newsroom, its IPO prospectus and investor factsheet, SEC filings, Reuters, Bloomberg-sourced reporting, and trade press. Reported facts and XAIPART analysis are kept separate, and source conflicts are recorded rather than resolved silently.
1. Executive summary
Anthropic is expected to price its IPO on November 18, 2026, and to begin trading on Nasdaq under the ticker ANT on November 19. The offering covers 825,000,000 Class A shares at an expected price of $70 to $78 each, for gross proceeds of up to about $64.4 billion, including shares sold by existing stockholders. A concurrent private placement of $26 billion in Class N common stock will go to seven strategic investors. At the top of the range, the fully diluted valuation is about $1.4 trillion. Bloomberg reported that the order book was about six times oversubscribed. S1S22S64
Operating momentum is unusual for a company at this stage. Third-quarter 2026 revenue was $16.956 billion, up 1,127% year over year. Revenue for the first nine months was $33.242 billion, and run-rate revenue reached about $81 billion in October. S19S23S18
The balance sheet carries the other side of that growth. The amended S-1 discloses about $895.4 billion of purchase commitments and lease agreements, substantially all related to compute and infrastructure. About $447.5 billion is scheduled to be incurred before December 31, 2030, and about $69.8 billion falls due within 12 months. S19S23
On November 9, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, its most capable models for long-running technical work, at $8 per million input tokens and $40 per million output tokens. S12
XAIPART analysis The central tension in the IPO is not demand for Claude. It is the mismatch between fixed, multi-year compute commitments and revenue that is mostly usage-based. Amazon and Google are simultaneously investors, distribution channels, compute suppliers, and competitors. That combination supported Anthropic's growth; it also concentrates risk in a small group of counterparties. S20
2. Company timeline
The table lists milestones that can be traced to a public source. Where a source gives only a month, the entry stays at month level.
| Date | Event | Source |
|---|---|---|
| 2021 | Dario Amodei, Daniela Amodei, and other former OpenAI researchers found Anthropic as an AI safety and research company. | S17, S39 |
| 2021-05 | Series A raises $124 million to support research on reliable, steerable AI systems. | S40 |
| 2022-04 | Series B raises $580 million in a round led by Sam Bankman-Fried. | S41 |
| 2023-03-14 | Anthropic releases its first Claude model. | S60 |
| 2023-05 | Series C raises $450 million. | S59 |
| 2023-07-11 | Claude 2 is released. | S61 |
| 2023-09-25 | Amazon agrees to invest up to $4 billion; AWS becomes Anthropic's primary cloud and training partner. | S47 |
| 2023-10-27 | Google agrees to invest up to $2 billion. | S62 |
| 2024-03-04 | Anthropic releases the Claude 3 family: Opus, Sonnet, and Haiku. | S63 |
| 2024-06-20 | Claude 3.5 Sonnet is released. | S43 |
| 2024-11-25 | Anthropic publishes the Model Context Protocol as an open standard. | S46 |
| 2025-02-24 | Claude 3.7 Sonnet and a research preview of Claude Code are released. | S44 |
| 2025-03-03 | Series E raises $3.5 billion at a $61.5 billion post-money valuation. | S6 |
| 2025-05-22 | Claude Opus 4 and Claude Sonnet 4 are released. | S45 |
| 2025-09-02 | Series F raises $13 billion at a $183 billion post-money valuation; run-rate revenue exceeds $5 billion. | S5 |
| 2025-10-23 | Anthropic expands its use of Google Cloud TPUs, with up to 1 gigawatt of capacity. | S48 |
| 2025-11-12 | Anthropic commits $50 billion to US AI infrastructure, including work with Fluidstack. | S49 |
| 2025-11-18 | Microsoft and NVIDIA announce strategic partnerships with Anthropic: $30 billion of Azure compute, an NVIDIA investment of up to $10 billion, and a Microsoft investment of up to $5 billion. | S10 |
| 2026-02-05 | Claude Opus 4.6 is released. | S58 |
| 2026-02-12 | Series G raises $30 billion at a $380 billion post-money valuation; run-rate revenue is $14 billion. | S4 |
| 2026-02-24 | Responsible Scaling Policy version 3.0 takes effect. | S31 |
| 2026-04-06 | Anthropic signs a multi-gigawatt next-generation TPU agreement with Google and Broadcom, with capacity starting in 2027. | S8 |
| 2026-04-20 | Anthropic signs an agreement with Amazon for up to 5 gigawatts and more than $100 billion of AWS spending over ten years; Amazon invests $5 billion immediately and up to $20 billion more in the future. | S7 |
| 2026-05-06 | Anthropic signs a compute deal with SpaceX for all capacity at Colossus 1, more than 300 megawatts and more than 220,000 NVIDIA GPUs. | S9 |
| 2026-05-28 | Series H raises $65 billion at a $965 billion post-money valuation; run-rate revenue is $47 billion. | S3 |
| 2026-06-09 | Claude Fable 5 and Claude Mythos 5 are released. | S11 |
| 2026-06-12 | A US export-control directive suspends access to Fable 5 and Mythos 5 for all foreign nationals. | S34 |
| 2026-06-30 | The export controls are lifted; Fable 5 returns globally on July 1. | S33 |
| 2026-07-24 | Claude Opus 5 is released. | S51 |
| 2026-09-15 | Claude Fable 5.1 and Mythos 5.1 are released. | S52 |
| 2026-09-22 | Claude Opus 5.5 is released. | S14 |
| 2026-09-28 | Claude Sonnet 5.5 is released. | S15 |
| 2026-10-07 | Claude Haiku 5.5 is released. | S16 |
| 2026-10-18 | Anthropic establishes the Public Benefit Fund, committing 5% of its shares and an additional 2% for every 50% increase in company value. | S38 |
| 2026-10-26 | Anthropic files its Form S-1 with the SEC. | S2 |
| 2026-11-02 | Anthropic files Amendment No. 1 to the S-1, adding third-quarter results and compute commitments. | S19 |
| 2026-11-09 | Fable 5.2 and Mythos 5.2 are released; a filing copy that showed 843.3 million shares at $70 to $80 is briefly circulated and then retracted. | S12, S21 |
| 2026-11-10 | The roadshow begins with 825 million shares at $70 to $78, ticker ANT, and a $26 billion private placement. | S1 |
| 2026-11-18 (expected) | IPO pricing. | S64 |
| 2026-11-19 (expected) | Nasdaq trading begins. | S22 |
3. Products and technology
3.1 The Claude model family
Anthropic released at least 11 model versions between February and November 2026. The table below focuses on the Claude 5 generation that is currently on sale. Prices are standard API prices per million tokens. S13S50
| Release | Model | Positioning | Context | Price (input/output) |
|---|---|---|---|---|
| 2026-11-09 | Claude Fable 5.2 / Mythos 5.2 | Highest capability for long-running technical work; Mythos is limited to vetted organizations | 1M (Claude 5 documentation) | $8 / $40; fast mode $16 / $80 S12 |
| 2026-10-07 | Claude Haiku 5.5 | High-volume, low-cost small model for summaries, subagents, and browser use | Not separately disclosed; Haiku 4.5 is 200K | Unit price not disclosed; Anthropic says about 75% cheaper than Haiku 4.5 S16 |
| 2026-09-28 | Claude Sonnet 5.5 | Everyday tasks, bug fixes, documents, and spreadsheets | 1M | $2 / $10 S15 |
| 2026-09-22 | Claude Opus 5.5 | Primary model for agentic coding and knowledge work | 1M | $4 / $20 S14 |
| 2026-09-15 | Claude Fable 5.1 / Mythos 5.1 | Frontier coding, knowledge work, and research | 1M | Not separately verified S52 |
| 2026-07-24 | Claude Opus 5 | Long-running agents and complex enterprise work | 1M | $5 / $25 S50S51 |
| 2026-06-30 | Claude Sonnet 5 | Frontier performance at scale | 1M | $2 / $10 S53 |
| 2026-06-09 | Claude Fable 5 / Mythos 5 | Mythos-class model; Fable for general use, Mythos for controlled access | 1M | $10 / $50 S11S50 |
The context column follows the Claude 5 documentation. Anthropic has not separately disclosed context limits for versions 5.1, 5.2, and 5.5; if the documentation changes, the documentation should govern. The earlier lineage, in release order, runs from Claude 1 in March 2023 through Claude 2, Claude 3, Claude 3.5 Sonnet, Claude 3.7 Sonnet with Claude Code, Claude 4, Claude Sonnet 4.5, and Opus 4.5. Opus 4.6, 4.7, and 4.8 followed on February 5, April 16, and May 28, 2026, respectively, according to Anthropic's lifecycle table. S50S45S44S43S60S61S63
Fable and Mythos require a separate explanation. They share the same underlying model. Fable carries stronger cybersecurity and biology safeguards and is available to all users. Mythos lifts some of those safeguards and is available only to vetted organizations through the Life Sciences and Cyber Verification Programs. S11S12 Anthropic says version 5.2 produces output up to 30% faster than 5.1 and costs $8 / $40 per million tokens, down from $10 / $50. The full safety evaluation is in the system card. S12S42
3.2 Claude Code, Cowork, and the enterprise product line
Anthropic discloses more about Claude Code than about any other product. It reached $1 billion of run-rate revenue in November 2025 and more than $2.5 billion in February 2026; enterprise customers accounted for more than half of that revenue. S56S4 By April 2026, more than 1,000 customers were spending more than $1 million a year, double the number two months earlier. S8 Anthropic says Claude wrote more than 90% of the contributions to its own codebase in September 2026. S18
Cowork launched in January 2026 and extends Claude Code's engineering capabilities to sales, legal, finance, and other knowledge work, with 11 open-source plug-ins. S4 The enterprise line also includes Claude for Enterprise and unified access through AWS, Google Cloud, and Microsoft Foundry. S7S8S10 On October 22, 2026, Anthropic donated the Agent Skills standard to the Agentic AI Foundation; the company says more than 70 agent platforms have adopted it. S57
On benchmarks, Anthropic reports that Fable 5.2 scores 49.1% on FrontierCode 2.0, ahead of Opus 5.5 at 45.5%, GPT-6 Astra at 42.8%, and Gemini 4 Carbon at 41.1%. S12 These are vendor-run results. Anthropic does not include independent replications, so cross-vendor comparisons should keep the methodological differences in view.
4. Commercial and financial profile
4.1 Funding history
| Date | Round or event | Amount | Post-money valuation | Source |
|---|---|---|---|---|
| 2021-05 | Series A | $124 million | Not disclosed | S40 |
| 2022-04 | Series B | $580 million | Not disclosed | S41 |
| 2023-05 | Series C | $450 million | Not disclosed | S59 |
| 2023-09 | Amazon strategic investment | Up to $4 billion | Not disclosed | S47 |
| 2025-03-03 | Series E | $3.5 billion | $61.5 billion | S6 |
| 2025-09-02 | Series F | $13 billion | $183 billion | S5 |
| 2025-11-18 | Microsoft and NVIDIA | Up to $15 billion ($5 billion plus $10 billion) | Not disclosed | S10 |
| 2026-02-12 | Series G | $30 billion | $380 billion | S4 |
| 2026-04-20 | Amazon additional investment | $5 billion immediate; up to $20 billion future | Not disclosed | S7 |
| 2026-05-28 | Series H | $65 billion | $965 billion | S3 |
| 2026-11 | IPO plus private placement | Up to $64.4 billion including secondary shares, plus $26 billion | About $1.4 trillion fully diluted | S1S22 |
The $26 billion private placement is made up of Class N common stock bought by seven companies: NVIDIA and Google at $10 billion each, AMD at $2.5 billion, Broadcom, Micron, and Microsoft at $1 billion each, and TSMC at $500 million. S30S54 The shares carry no voting rights and are priced at the IPO offer price. S54
4.2 Revenue trajectory
| Period | Metric | Value | Source |
|---|---|---|---|
| Full year 2025 | Revenue | Close to $4.6 billion | S20 |
| 2025-08 | Run rate | More than $5 billion | S5 |
| End of 2025 | Run rate | About $9 billion | S7 |
| 2026-02-12 | Run rate | $14 billion | S4 |
| 2026-04 | Run rate | More than $30 billion | S7S8 |
| 2026-05 | Run rate | $47 billion | S3 |
| 2026 Q3 | Quarterly revenue | $16.956 billion | S19S23 |
| First nine months of 2026 | Revenue | $33.242 billion | S19S23 |
| 2026-10 | Run rate | About $81 billion | S18 |
Of 2025 revenue, about $3.8 billion came from usage-based billing and $789 million from subscriptions. The United States accounted for nearly two-thirds of sales. S20 XAIPART analysis Usage-based billing dominates, which means revenue moves with customer token consumption rather than with seat-based renewals.
4.3 Income statement and unit economics
In the third quarter of 2026, Anthropic reported revenue of $16.956 billion, up 1,127% year over year. Compute and infrastructure expense was $9.863 billion. Total operating expenses were $17.523 billion. The company reported an operating loss of $567 million and adjusted operating income of $2.463 billion, a 15% adjusted operating margin. The net loss was $23.291 billion, including a $21.936 billion non-cash charge from changes in the fair value of convertible notes, derivative liabilities, and SAFEs. S19S23
For the first nine months, revenue was $33.242 billion, the operating loss was $1.260 billion, and the net loss was $104.039 billion, including $94.324 billion of fair-value changes. Operating activities generated $6.569 billion of cash, and financing activities provided $110.735 billion. At September 30, liquidity was $146.6 billion, comprising $95.6 billion of cash, $36.3 billion of short-term investments, and $14.8 billion available under a revolving credit facility. S19S23 The IPO materials separately list about $141 billion of capital raised, about $96 billion of cash, and about $36 billion of short-term investments. S18
Full-year 2025 revenue was close to $4.6 billion, and the operating loss exceeded $8 billion. S20 The absolute loss is still rising.
Gross margin is a source of genuine conflict. BSIC calculates a gross margin of about 85% for the first nine months of 2026 from the S-1. S27 The same filing reports $9.863 billion of third-quarter compute and infrastructure expense, equal to 58% of quarterly revenue. S19S23 The two figures cannot be subtracted directly because the S-1 does not present inference and serving costs under a conventional cost-of-goods-sold line. The defensible conclusion is narrower: the gross margin depends on how much compute expense is classified as cost of revenue, and Anthropic has not disclosed a comparable GAAP gross margin.
| Metric | Value | Period | Source |
|---|---|---|---|
| Quarterly revenue | $16.956 billion | Q3 2026 | S19S23 |
| Nine-month revenue | $33.242 billion | January to September 2026 | S19S23 |
| Run-rate revenue | About $81 billion | October 2026 | S18 |
| Adjusted operating income | $2.463 billion, 15% margin | Q3 2026 | S23 |
| Net loss | $23.291 billion | Q3 2026 | S19S23 |
| Cash and short-term investments | $131.9 billion ($95.6 billion plus $36.3 billion) | 2026-09-30 | S23 |
| Purchase and lease commitments | About $895.4 billion | 2026-11-02 | S19S23 |
| TTM revenue per employee | More than $6.6 million | Q3 2026 | S18 |
4.4 Commitments and off-balance-sheet obligations
The amended S-1 discloses about $895.4 billion of purchase commitments and lease agreements, substantially all related to compute and infrastructure. S19S23 About $447.5 billion is scheduled to be incurred before December 31, 2030, with most of the remainder falling between 2031 and 2045. About $69.8 billion comes due within 12 months. About $213.8 billion is tied to arrangements with exercisable termination rights or pre-commencement conditions, including financing, facilities, and regulatory approvals. S19S23
Under the September 30, 2026 contractual-obligations table, fixed obligations total $319.993 billion: $4.224 billion of operating leases, $1.950 billion of finance leases, and $313.819 billion of purchase commitments. Outside that table sit about $451.2 billion of executed but not-yet-commenced lease obligations, $100.2 billion of commitments entered after September 30, and $150 billion to $175 billion of arrangements without fixed dollar commitments, including $125 billion to $150 billion for Google TPU capacity expected between 2028 and 2031. S19S23
According to Unite.AI's reading of the S-1/A, minimum commitments by counterparty are about $121.3 billion to Google from April 2026 to July 2033, $110.0 billion to Amazon from May 2026 to April 2036, and $44.9 billion to Microsoft from November 2026 to June 2033. Five-year equipment leases with Broadcom and financing entities total about $161.2 billion. S23 Reuters, as reported by International Finance, puts the Google commitment at no less than $111.1 billion and the Microsoft commitment at no less than $47.2 billion. S22 The two sets of figures use different dates and definitions. This report records the conflict and does not choose one as definitive.
Anthropic also describes a model called Anthropic Direct. Under it, part of the company's sourcing shifts from fixed-term cloud agreements to directly leased chips hosted in dedicated facilities. The structure secures data-center capacity earlier and defers accelerator purchases until demand is more certain. S19S23 XAIPART analysis The structure lowers near-term capital expenditure, but it leaves utilization risk with Anthropic.
5. Compute and the supply chain
5.1 From upstream to downstream
| Stage | Disclosed facts | Basis |
|---|---|---|
| Advanced process and packaging | The S-1 risk factors state that Anthropic has already experienced hardware delivery delays because TSMC could not meet market demand, and that alternative capacity would take years to build and qualify. S19S24S26 VLSI says leading-edge logic wafers go mainly to TSMC, while memory value accrues to the memory makers. S25 | Fact plus third-party analysis |
| HBM and memory | SK hynix, Samsung, and Micron are named strategic infrastructure partners in the Series H announcement. S3 Micron is also investing $1 billion in the IPO private placement. S22 The S-1 lists a potential HBM shortage as a risk. S19S24 VLSI estimates cumulative HBM demand of 8 billion to 13 billion Gb, with a single customer potentially accounting for a single-digit share of global HBM bit demand in peak years. S25 | Fact plus estimate |
| Accelerators | On AWS, Anthropic uses more than 1 million Trainium2 chips, and its agreement covers Trainium2 through Trainium4. S7 On Google, the company uses TPUs developed with Broadcom, with multiple gigawatts starting in 2027. S8 On NVIDIA, it has committed to up to 1 gigawatt of Grace Blackwell and Vera Rubin capacity. S10 AMD is supplying GPUs and has committed compute capacity. S22 SpaceX's Colossus 1 provides more than 220,000 NVIDIA GPUs. S9 | Fact |
| Cloud and data centers | AWS is the primary cloud and training partner, with more than $100 billion committed over ten years and up to 5 gigawatts. S7 Google and Broadcom will provide up to 5 gigawatts starting in 2027. S8 Azure has a $30 billion commitment and up to 1 gigawatt. S10 SpaceX provides all capacity at Colossus 1, more than 300 megawatts. S9 | Fact |
| Financing structure | Broadcom is providing up to $42 billion of financing for Anthropic's data-center leases, and Anthropic expects to become Broadcom's largest compute customer in 2027. S22 Five-year equipment leases total about $161.2 billion. S23 Seven suppliers are buying $26 billion of IPO stock. S30S54 | Fact |
| Downstream implications | Multi-gigawatt power, liquid cooling, networking, and AI server manufacturing demand are being locked in early. HBM and advanced packaging are tighter constraints than GPU die area. Suppliers gain order visibility, but they also become more exposed to a single customer's demand cycle. | XAIPART analysis |
5.2 Where the bottlenecks are
XAIPART analysis The tightest links in this chain are not GPU die area. They are HBM supply, advanced packaging capacity, and grid connections. Anthropic's multi-chip strategy spans Trainium, TPU, NVIDIA, and AMD. That reduces single-supplier risk, but it raises software porting and qualification costs. The $895.4 billion of take-or-pay commitments give suppliers demand visibility while leaving downside risk on Anthropic's balance sheet.
XAIPART analysis The $26 billion private placement ties suppliers' balance sheets to Anthropic's capital spending. TSMC's $500 million is small relative to the $895.4 billion commitment, so it is more a signal than a financial hedge. Micron's $1 billion, together with the strategic partner roles of SK hynix and Samsung, shows that memory suppliers have gained bargaining power in this expansion.
XAIPART analysis By counterparty, Google and Amazon account for about $231 billion of commitments, Microsoft for about $45 billion, and Broadcom leases for about $161 billion. The first three are cloud partners; Broadcom is a TPU partner and financier. Anthropic's compute base is therefore not a single-cloud lock-in. It is AWS first, with Google TPU as the second growth curve and Azure and NVIDIA as supplements. That structure can win more allocations when supply is tight. When demand is weak, it creates several payment obligations that mature at the same time.
XAIPART analysis HBM and advanced packaging capacity expands over quarters to years, while Anthropic's revenue is billed by token and moves month to month. Once upstream capacity is switched on, it is hard to slow quickly. If downstream demand cools, inventory and depreciation appear first in supplier results and then feed back to Anthropic through long-term contract prices. The $42 billion Broadcom lease financing adds another financial layer; interest rates and lease terms become new sensitive variables.
XAIPART analysis TSMC's $500 million subscription is too small to count as consideration for locked capacity. The actual capacity locks are the multi-year agreements and prepayments with Broadcom, Amazon, and Google. For second-tier suppliers, order visibility rises, but customer concentration rises with it. If Anthropic changes one supplier's allocation in 2027, the effect runs through an already scheduled chain of materials and packaging.
What is not disclosed. Anthropic has not published wafer starts by supplier, HBM bit allocations, individual data-center power contracts, or site-level commissioning schedules. Any attempt to convert the $895.4 billion commitment into chip volumes lacks the intermediate parameters. This report does not make that conversion.
6. Safety and governance
Responsible Scaling Policy version 3.0 took effect on February 24, 2026 and is a comprehensive rewrite of the policy first published in 2023. S31 The current framework ties capability thresholds to the ASL-3 security and deployment standards. For models with chemical or biological weapons capabilities, Anthropic requires safeguards at least as strong as its initial Constitutional Classifiers. S32 The Frontier Safety Roadmap published on February 22, 2026 lists the ASL-3 measures that apply to current models. S32
The Fable 5 launch shows how the policy works in practice. For cybersecurity and biology-related queries, the system routes requests to the next-most-capable model, Opus 4.8. Anthropic says conservative tuning causes false positives, but the safeguards trigger in less than 5% of sessions. S11 Mythos 5 is deployed through Project Glasswing with the US government and is available to defensive cybersecurity organizations. S11
The June 2026 export-control episode was a stress test for governance. On June 12, the US government invoked national-security authorities to suspend access to Fable 5 and Mythos 5 for all foreign nationals, including Anthropic's foreign employees. Anthropic therefore disabled both models for all customers. S34 The controls were lifted on June 30, and Fable 5 returned globally on July 1. Anthropic also proposed a jailbreak-severity scoring framework with Amazon, Microsoft, Google, and Glasswing partners. S33
For external evaluation, Anthropic has testing or information-sharing arrangements with METR, the UK AI Security Institute, the US Center for AI Standards and Innovation, and the Australian AI Safety Institute. S35S36 On August 4, 2026, the UK AISI published a report describing anomalous behavior by Mythos 5 during cybersecurity testing. Anthropic then published an alignment assessment in response. S37 XAIPART analysis The episode shows that third-party testing can now affect market access directly.
On governance, Anthropic is a public benefit corporation. The Long-Term Benefit Trust currently has four financially disinterested trustees and generally has the authority to elect a majority of the board. S18S39 The Public Benefit Fund announced on October 18, 2026 commits 5% of Anthropic's shares, with an additional 2% for every 50% increase in company value; the board has discretion over the additional 2%. S38
XAIPART analysis The export-control episode turned model access into a policy variable. For enterprise buyers, the practical implication is that access to a frontier model depends not only on a commercial contract but also on nationality, use case, and government direction. Procurement contracts should include fallback models and exit clauses.
7. Competitive position
| Company | Latest public valuation | Latest public revenue measure | Flagship model | Relationship with Anthropic |
|---|---|---|---|---|
| OpenAI | $852 billion post-money (2026-03-31) S28 | About $50 billion annualized (FT, October 2026) S29 | GPT-6 Astra | Direct competitor; Microsoft is also an Anthropic investor and cloud partner S10 |
| Alphabet is publicly listed | Gemini revenue is not separately disclosed | Gemini 4 Carbon | Investor, cloud supplier, distribution channel, and competitor S20 | |
| DeepSeek | Not disclosed | About $1 billion annualized, according to The Information; single-source and unverified S55 | V4.1 Flash S65 | Open-weight price competition |
| xAI / SpaceX | Not disclosed | Not disclosed | Not disclosed | Landlord of Colossus 1 and a potential competitor S9 |
| Anthropic | About $1.4 trillion fully diluted at the IPO S22 | About $81 billion run rate (October 2026) S18 | Fable 5.2 / Mythos 5.2 | — |
Revenue comparisons need a qualification. Anthropic recognizes cloud-marketplace transactions on a gross basis and records the platform's share as a sales expense. OpenAI has told investors that this approach inflates Anthropic's reported revenue by billions of dollars. Anthropic says it is the principal in those transactions and that gross recognition follows accounting rules. S20 The two companies' revenue measures are not directly comparable, and the $81 billion and $50 billion figures should not be compared by simple subtraction.
XAIPART analysis Anthropic's relative strength is in enterprise coding and multi-cloud access. Its weakness is that distribution runs through Amazon and Google. In 2025, 47% of revenue was settled through those two cloud marketplaces. The platforms can see pricing and commercial terms, and they influence compute allocation. S20 That relationship is an amplifier in a growth phase and a source of leverage in a conflict.
XAIPART analysis The gap between Anthropic and OpenAI is mainly distribution, not model quality. OpenAI has the ChatGPT consumer entry point and earlier brand recognition. Anthropic has enterprise coding, Claude Code, and access to three clouds. Both companies are using multi-year compute commitments to win cloud partners. Anthropic's revenue is more dependent on usage-based billing, which makes it more sensitive to token prices and inference costs. DeepSeek's low-cost open-weight models compress pricing from below and make a durable "frontier capability premium" harder to sustain.
8. Risks and uncertainties
- Duration mismatch. About $69.8 billion of the $895.4 billion commitment matures within 12 months, while revenue is mostly usage-based and many customers have no long-term contracts. S23S20 If token demand grows more slowly than committed capacity, Anthropic will need financing or price cuts to absorb idle capacity.
- Channel and customer concentration. In 2025, 47% of revenue was settled through Amazon and Google marketplaces, with about $351 million paid in distribution fees. At the end of 2025, those two companies were responsible for collecting 60% of receivables, and two unnamed customers each accounted for 12% of revenue. S20 Channel friction affects both revenue recognition and cash flow.
- Circular transactions and related parties. Suppliers are investing $26 billion before the IPO while holding multi-year purchase commitments. Amazon has invested about $18 billion and holds commitments of at least $110 billion; Google has invested about $33.3 billion and holds commitments of at least $111.1 billion; Broadcom is providing up to $42 billion of lease financing. S22 Critics say this ties supplier revenue and shareholder returns to the same demand assumption. Supporters note that the Class N shares have no voting rights and that the purchases are priced at market rates, an arrangement that is common in capital-intensive industries. Both sides point to the same fact: the value of these arrangements depends on Anthropic converting commitments into revenue. The issue remains unresolved.
- Supply concentration. The S-1 explicitly lists dependence on TSMC manufacturing and a potential HBM shortage as risks, and says alternative capacity would take years to build and qualify. S19S24S26 The expansion plans for up to 5 gigawatts with Amazon, and for up to 5 gigawatts with Google and Broadcom, require advanced packaging, HBM4, and grid connections to arrive on schedule together.
- Policy and export controls. The June 2026 suspension of Fable 5 and Mythos 5 shows that a single directive can cut off all customer access within hours. S34 The scope and triggers of future controls are not disclosed.
- Model safety and product migration. Fable 5 launched with universal-jailbreak protections and conservative blocking; controlled access to Mythos depends on verification programs. S11 Model retirement is also accelerating. Sonnet 4.5 is scheduled for retirement on November 30, 2026, so customers must keep migrating. S50
- Valuation and IPO execution. On November 9, a filing copy showing 843.3 million shares at $70 to $80 and proceeds of up to $69.3 billion was retracted; the official terms on November 10 were 825 million shares at $70 to $78. S21S1 The earlier $2 trillion expectation has come down, and some investors see fair value closer to $1 trillion. S21 The offering was about six times oversubscribed, and allocation priority for existing backers shows strong demand, but pricing remains a negotiation. S22
- Accounting comparability. Gross margin, cloud-marketplace revenue recognition, and adjusted operating income all involve definitional differences. Cross-company comparisons require line-by-line reconstruction. S20S27
XAIPART analysis These risks are not independent. Channel concentration, circular transactions, and duration mismatch all point to one variable: whether Anthropic can push annualized revenue toward the $150 billion range by the end of 2027 while keeping compute cost growth below revenue growth. If all three deteriorate at once, a valuation adjustment will appear before the fundamental data does. In the other direction, fourth-quarter revenue above $20 billion, a doubling of Claude Code's run rate, and on-time TPU deliveries in 2027 would relieve all three risks at once.
9. Outlook and falsifiable triggers
The following statements are XAIPART analysis. Each uses an observable threshold rather than a directional opinion.
| Claim | Window | Threshold and trigger | What falsification would mean |
|---|---|---|---|
| The IPO can hold its range | by 2026-12-18 (about one month after listing) | If ANT closes below $56, 20% below the $70 low end, for 20 consecutive trading days, the IPO has failed by this measure. | The window for large AI IPOs closes, and other AI companies delay listings. |
| Revenue growth can cover the commitments | by 2027-03-31, when fourth-quarter results are disclosed | Fourth-quarter 2026 revenue of at least $20 billion and an adjusted operating margin of at least 15%. | Revenue below $18 billion or a margin below 10% would leave the $1.4 trillion valuation without current support. |
| 2027 compute comes online as planned | by 2027-06-30 | Google and Broadcom confirm that the first new 2027 TPU capacity is online. | A delay of more than two quarters would lower the 2027 revenue ceiling. |
| Claude Code keeps driving growth | by 2027-02-28 | Claude Code run-rate revenue of at least $5 billion, double the $2.5 billion disclosed in February 2026. | A figure below $4 billion would remove the strongest public data point behind the coding thesis. |
| The commitment structure is not renegotiated | within 12 months | Any single commitment of at least $50 billion is terminated, reduced, or delayed by more than two quarters. | Supplier revenue visibility would fall, and the discount applied to circular transactions would widen. |
10. Method and limitations
This report takes November 17, 2026 as its cutoff. It draws on 19 independent registered domains, prioritizing Anthropic's newsroom, its IPO materials, SEC EDGAR, and OpenAI's official announcements. Media sources are used for cross-checking and background. The key SEC figures ($895.4 billion, $447.5 billion, and $213.8 billion) were confirmed in the S-1/A through EDGAR full-text search. S19
The report has three limitations. First, the IPO had not been priced at the time of writing, so the offer price, final proceeds, and first-day trading performance may change. Second, the detailed commitment schedule relies in part on public summaries and third-party reconstructions; not every footnote in the S-1/A was rechecked line by line. Third, management did not grant an interview, so judgments about strategy are labeled XAIPART analysis.
Source conflicts recorded in this report
- Google and Microsoft minimum commitments: $121.3 billion and $44.9 billion in the Unite.AI reconstruction of the S-1/A, versus at least $111.1 billion and $47.2 billion in Reuters reporting. The dates and definitions differ, and the conflict remains unresolved. S23S22
- Gross margin: about 85% in BSIC's calculation versus a 58% ratio of quarterly compute expense to revenue in the S-1/A. The filing does not present a conventional cost-of-goods-sold line, so the conflict remains unresolved. S27S19
- IPO terms: an earlier circulating copy showed 843.3 million shares at $70 to $80 and up to $69.3 billion; the official terms are 825 million shares at $70 to $78. The earlier copy was retracted. S21S1
- Revenue recognition: Anthropic reports cloud-marketplace sales on a gross basis; OpenAI says this inflates Anthropic's revenue by billions. Anthropic says it is the principal in the transaction. The conflict remains unresolved. S20
Các lựa chọn và chu kỳ sống
| Phần | Quan hệ | Chu kỳ sống | Cơ sở |
|---|---|---|---|
| — | Không có tùy chọn nào trong tệp CẦN DỮ LIỆU | — | Yêu cầu match_type trước khi bất kỳ nhãn tiền thân nào có thể xuất hiện |
“Tiền thân” và các nhãn thay thế chỉ đến từ một loại khớp đã được ghi lại, trạng thái chu kỳ sống hoặc văn bản nguồn. Khi không có khớp nào được ghi lại, không có nhãn nào được hiển thị. Các danh sách của chúng tôi không phải là một tham chiếu thị trường.
Bức tranh cạnh tranh
- 6 danh sách trực tiếp trong loại này trên XAIPART
- 3 các nhà sản xuất đại diện
- 55 các nhà sản xuất được nhắc đến trong kho lưu trữ tin tức của chúng tôi trong 30 ngày qua
Số lượng phản ánh chỉ catalog của XAIPART — chúng không phải là thị phần.
Các phần bị ảnh hưởng
Không có khớp catalog. Câu chuyện này không khớp với bất kỳ linh kiện nào được liệt kê trên XAIPART hiện tại. Chúng tôi không suy luận khớp và chúng tôi không hiển thị danh sách chờ.
Áp lực chi phí theo chiều
| Dimension | FRED series ID | Unit | Lag | As of |
|---|---|---|---|---|
| Input cost | PCOPPUSDM, PALUMUSDM, PNICKUSDM | USD/mt | 52 d | |
| Supply/demand | PCU334413334413, IPG3344S | index(1982=100) / index(2017=100) | 46 d | |
| FX & energy | DEXCHUS, DCOILBRENTEU, DHHNGSP | rate / USD/bbl / USD/MMBtu | 5 d |
No free authoritative interface for: gold · silver · rare earth · polysilicon · lithium carbonate · market share — listed rather than approximated.
Bình luận
Trở thành người đầu tiên bình luận.
🔐 Vui lòng đăng nhập để đăng một bình luận.
Bạn không có tài khoản? Đăng ký