Pre-IPO primer · Information technology / AI foundation models · As of 21 October 2026
Anthropic
Neutral pre-listing primer on the Claude developer. Sources are company releases, media reports on the draft prospectus, expert-call transcripts, partner press releases and market data; sell-side broker research is excluded by request.
1Overview
Anthropic comes to market with the fastest revenue ramp on record for a company of its size and a compute bill that dwarfs that revenue, so the listing tests whether public investors will fund the gap at a premium to the $965B private mark.
Anthropic builds the Claude family of large language models and sells them through developer APIs, coding and conversational assistants, and integrations into productivity software. [7] It is a public benefit corporation controlled by a Long-Term Benefit Trust of independent experts in AI safety, national security, public policy and social enterprise. [8]
Revenue reached $16.3B in the six months to 30 June 2026, against $1.1B a year earlier, while compute and infrastructure spending rose to almost $9B from $2.3B. [2] Third-quarter revenue rose 45% QoQ to $16.8B, after $11.6B in Q2 and $4.7B in Q1. [1] The company has committed $518B to compute over a decade, roughly 80% of it noncancellable. [5]
Why list now
Investors quoted by Axios frame the IPO as a funding necessity: Anthropic faces more than $600B in future infrastructure costs and is negotiating deals that could add another $150B. [9] Private capital is getting dearer, with Anthropic offering Blackstone a yield above 10% on $9B of junior debt inside a $60B chip-financing package, up from 8.5% on an earlier deal. [10] OpenAI is racing to raise about $30B at a valuation above $1.4T before Anthropic's mid-November debut. [11]
Investment triad (neutral framing)
Near-term verifier
The Q3 print showed sequential growth slowing to 45% from more than doubling in Q2, and some investors were disappointed by the deceleration. [12] The public filing's Q3 detail and the price range against the $965B Series H are the first hard checks. [13]
Terminal-value narrative
Usage-based revenue was about $3.8B of 2025's roughly $4.6B, with subscriptions at $789M, and the company expects usage to stay dominant. [14] The long-run case rests on enterprise agentic workloads scaling faster than the fixed compute bill.
Central debates
- Growth durability. Some current and prospective investors argued for a debut nearer $1.5T rather than $2T after Q3 fell short of expectations. [17]
- Compute versus unit economics. A fixed compute commitment of $1.25B a month through May 2029 sits against usage-linked revenue. [18]
- Revenue quality. OpenAI books revenue net of partner shares and has accused Anthropic of overstating revenue by counting the full value of contracts sold through cloud providers. [19] About one-fifth of H1 2026 revenue came from two customers. [20]
- Governance and safety. The trust can elect a board majority through Class T stock within four years of the Series C close and currently elects three directors. [21]
All figures from the draft prospectus are as reported by media outlets; the company has not confirmed them and the public filing has not been released. [22]
2Business & segments
Anthropic is a usage-priced model vendor first and a subscription software company second, which ties revenue to customer token consumption rather than seats.
Of roughly $4.6B in 2025 revenue, about $3.8B came from customers paying by usage of the Claude platform and $789M from subscriptions. [14] The company expects usage-based revenue to remain the dominant part of the business for the foreseeable future. [14]
The CFO points to Claude Code and Cowork as the products carrying demand into the Series H. [23] Customers that start with one entry point (the API, Claude Code or Claude for Work) are expanding integrations across their organizations, and eight of the Fortune 10 are Claude customers. [16]
Business-model chain
| Link | Evidence |
|---|---|
| Model and compute | Agreements for up to five gigawatts of new capacity with Amazon and five gigawatts of next-generation TPU capacity with Google and Broadcom. [24] |
| Direct channels | Expensify became one of the first expense tools available as a connector in Claude for Small Business on every paid plan. [25] |
| Cloud resale | Hexaware is authorized to resell Claude through Amazon Bedrock. [26] |
| Services partners | CI&T reached the Preferred tier of the Claude Partner Network three months after joining. [27] |
| Geography | Singapore is the fifth Asia-Pacific office, joining Tokyo, Bengaluru, Seoul and Sydney. [28] |
2025 revenue by pricing model, $B. [14]
Focus Five — evidence for and against
| Dimension | Evidence for | Evidence against |
|---|---|---|
| Organic growth | Revenue grew about twelvefold in 2025 to roughly $4.6B. [29] | Q3 sequential growth slowed to 45% from more than doubling in Q2. [12] |
| Margin trajectory | The company told its investor day it was profitable on an adjusted basis in Q3. [12] | 2025 operating loss widened to $8.06B from $2.98B in 2024, excluding liability writedowns. [29] |
| Capital intensity | H1 2026 compute spend of about $9B against $16.3B revenue compares with $7.3B of compute on $4.6B of 2025 revenue. [30] | Compute plans could reach $750B over coming years, with $450B of mostly noncancellable cloud deals signed. [31] |
| Capital deployment | Series H proceeds are earmarked for demand, frontier research and product reach. [23] | Junior debt pricing has stepped up to above 10% from 8.5%. [10] |
| Durability | Over 500 customers spend more than $1M annually. [16] | About one-fifth of H1 2026 revenue came from two customers. [20] |
Segment economics and a sum-of-the-parts are not shown: the reported split is by pricing model, not by separately disclosed operating segments.
3Industry & market
Frontier-model spend is concentrated in a few labs and a small set of very large buyers, and practitioners describe budgets moving from open-ended to managed.
Ark Investment Management's Brett Winton estimates the leading frontier labs collectively monetize at an annualized run-rate of about $150B, up 5x YoY. [32] Winton attributes part of Anthropic's slowdown to the competitive battle with OpenAI. [33]
A draft federal AI safety bill circulated by Anthropic would let the Commerce Department seek a court-ordered pause of a model only for a clear, near-term risk of more than 50 deaths or serious injuries or more than $1B of damage. [34] Political resistance to data-center construction is flagged as a business risk, with opposition described as more organized. [35]
Practitioner read (expert calls, each shown separately)
- An AWS enterprise sales professional sees enterprises moving to multi-model, task-based routing, and notes that consuming Anthropic or OpenAI models through the cloud provider counts toward contractual commitments and carries private pricing. [36]
- The same AWS expert describes customers that burn through a full year's token allocation within six months. [37]
- A Microsoft martech and AI budget owner says token budgets are shifting from loosely managed to structured, project-based allocations as costs become visible. [38]
- The head of AI at Pocket FM argues closed models (OpenAI, Anthropic, Gemini) retain pricing power, while the quality gap to open-weight and open-source models is narrowing. [39]
- A former Microsoft capacity-planning general manager sees AI and agentic demand as volatile and uncertain beyond a two-year horizon. [40]
A bottom-up TAM is omitted: no sourced unit count and price pair supports the arithmetic.
4Competition
OpenAI is the direct comparator on scale and on capital, and Anthropic's three largest cloud suppliers also sell competing models.
| Competitor | Latest datapoint |
|---|---|
| OpenAI | Reached an $852B valuation after a $122B round in March 2026. [41] |
| OpenAI | Says annualized revenue hit $50B after GPT-6 Astra and is seeking about $30B at above $1.4T. [42] |
| Has begun internal development of a next-generation model codenamed Gemini 4 Carbon. [43] | |
| Hyperscalers | Anthropic says it sought more capacity than some major cloud partners were willing to provide as they increasingly view it as a competitor. [44] |
| Open-weight models | A Microsoft AI function lead believes open-weight models are rapidly gaining share over frontier models. [45] |
Five forces (evidence-led)
| Force | Read |
|---|---|
| Supplier power | High: commitments include $111B to Google Cloud, $110B to AWS and $31B to Microsoft regardless of usage, plus more than $160B for Google chips via Broadcom. [46] |
| Buyer power | Rising: two customers produced about one-fifth of H1 2026 revenue. [20] |
| Rivalry | Intense: OpenAI and Anthropic courted an overlapping pool of long-only and sovereign funds in the same week. [47] |
| Substitutes | Growing: open-weight model quality is closing the gap with closed models. [39] |
| Entry barriers | Very high: frontier labs carry 12-figure spending commitments. [48] |
Anthropic reports taking in two to four times as much revenue per employee at five years old as Meta and Google did at the same stage. [49]
5Recent developments (last 30 days)
The past month moved the debate from whether Anthropic lists to at what price, with the draft prospectus leak, a softer Q3 and a governance push landing within days of each other.
- 21 Oct 2026Harvard astrophysicist Christopher Stubbs was named to the Long-Term Benefit Trust. [50]
- 21 Oct 2026AMD has pledged up to $5B toward Anthropic stock while Anthropic has committed more than $20B to AMD computing services. [51]
- 20 Oct 2026The draft prospectus reviewed by the WSJ showed a $80.7B H1 net loss, $16.3B of H1 revenue and nearly $9B of compute spend. [30]
- 20 Oct 2026OpenAI is accelerating a roughly $30B raise at above $1.4T to close before Anthropic's mid-November debut. [11]
- 18 Oct 2026Anthropic committed 5% of its shares to a new Public Benefit Fund, with up to 2% more at board discretion for every 50% rise in value above an undisclosed baseline. [52]
- 16 Oct 2026New York City Comptroller Mark Levine, who oversees a $330B pension system, asked Anthropic to pause its IPO until it addresses safety concerns. [53]
- Mid-Oct 2026At a San Francisco investor meeting, Anthropic reported Q3 revenue of $16.8B, up 45% QoQ. [1]
- 15 Oct 2026Anthropic opened a Singapore office a day after signing an AI safety and model-evaluation memorandum with the government. [54]
- 12 Oct 2026HCLTech set up a dedicated Anthropic business unit and center of excellence. [15]
- 25 Sep 2026Anthropic agreed to pay Akamai at least $11.6B over seven-plus years for cloud services, with a potential $9B expansion. [55]
6IPO mechanics
Timing, venue and the valuation band are reported, but price range, share count, primary/secondary split and lock-up tiers remain unfiled.
| Term | Status |
|---|---|
| Filing | Filed confidentially with the SEC in June 2026. [5] |
| Venue | Nasdaq chosen, per a person familiar with the matter. [56] |
| Banks | Goldman Sachs and Morgan Stanley teams have prepared the draft S-1. [57] |
| Timing | Roadshow targeted as early as the week of 9 November, with trading before US Thanksgiving. [58] |
| Delay history | A September listing was widely expected before a researcher's resignation and a run of safety incidents. [10] |
| Valuation talk | About $2T reported target; investors at a recent company meeting discussed about $1.5T. [4] |
| Deal size | The company could seek to raise at least as much as SpaceX did in its June IPO. [56] |
| Primary / secondary | Weighing whether existing holders sell in the offering, alongside a lock-up longer than 180 days for other investors and employees. [59] |
| Retail allocation | Not announced. [13] |
| Price range, share count, use of proceeds | — |
| Liquidity backstop | A $15B revolving credit facility was being finalized. [56] |
Illustrative dilution: if a raise matching SpaceX's reported $86.3B were entirely primary at a $1.5T post-money value, new shares would be about 5.8% of the company. [56] Reports differ on SpaceX's deal size, with Fortune citing a $75B offering. [60]
7Funding & cap table
Three primary rounds in nine months raised $108B and lifted the mark 5.3x, so late-stage holders carry large paper gains into the IPO.
| Round | Date | Size | Post-money | Leads | Type |
|---|---|---|---|---|---|
| Series A | 2021 | — | $461M | Eric Schmidt among early investors [61] | Primary |
| Series F | Sep 2025 | $13B | $183B | ICONIQ; co-led by Fidelity and Lightspeed [62] | Primary |
| Series G | Feb 2026 | $30B | $380B | GIC and Coatue [63] | Primary |
| Series H | May 2026 | $65B | $965B | Altimeter, Dragoneer, Greenoaks, Sequoia [3] | Primary (incl. $15B previously committed) [64] |
| Chip financing | 2026 | $60B | — | Blackstone on $9B junior tranche [10] | Debt |
| Revolver | 2026 | $15B | — | — [56] | Debt |
| AMD pledge | Oct 2026 | Up to $5B | — | AMD [51] | Strategic equity |
Rounds before Series F other than Series A are not covered in the reviewed sources. Total raised counts only Series F–H primary rounds ($13B + $30B + $65B).
Known holders and equity commitments
- CEO Dario Amodei reportedly holds about 2%, and Eric Schmidt about 1%. [65]
- The co-founders pledged at founding to donate 80% of their equity holdings. [66]
- The 5% Public Benefit Fund stake could be worth $75B–$100B at IPO. [67]
- The $72B H1 2026 charge reflects a higher estimated value of bonds convertible into Anthropic shares, which points to a convertible overhang the public filing will need to size. [68]
- Strategic holder stakes (Amazon, Google and others) are not disclosed in the reviewed coverage: —.
8Financial signals
Revenue now outruns compute spend, and most of the $80.7B H1 2026 net loss is a financing remeasurement, but the loss excluding that charge was still about $8.7B for the half. [69]
Reported history (draft prospectus, as reported)
| Line item ($B) | FY2024 | FY2025 | H1 2025 | H1 2026 |
|---|---|---|---|---|
| Revenue | — | ~4.6 [29] | 1.1 [2] | 16.3 [2] |
| Compute & infrastructure spend | — | 7.33 [70] | 2.3 [30] | ~9 [30] |
| Operating expenses | — | 12.65 [70] | — | — |
| Operating loss, excl. liability remeasurement | (2.98) [29] | (8.06) [29] | — | — |
| Non-cash financing charge | — | >34 [71] | — | 72 [72] |
| Net loss | — | (~42) [70] | (13) [72] | (80.7) [72] |
| Net loss excl. convertible charge | — | — | — | (~8.7) [69] |
Derived ratios (uncited, computed from the cells above): compute spend was ~159% of revenue in FY2025 and ~55% in H1 2026; operating expenses were ~275% of FY2025 revenue. Gross margin is not disclosed in the reviewed coverage.
Quarterly revenue, 2026, $B. [1]
Crux unit-economics bridge
| Step | Figure | Basis |
|---|---|---|
| Q3 2026 revenue, annualized | $67.2B | Derived: $16.8B × 4 [1] |
| Fixed compute commitment | $1.25B a month through May 2029 (~$15B a year) | Reported, nearly 30% of estimated 2026 revenue [18] |
| Fixed commitment ÷ Q3 annualized revenue | ~22% | Derived |
| Total compute commitments | $518B over a decade, ~80% noncancellable | Reported [5] |
| Commitments ÷ Q3 annualized revenue | ~7.7 years of current revenue | Derived |
| Q3 profitability | Profitable on an adjusted basis | Company statement at investor day [12] |
H1 revenue vs compute and infrastructure spend, $B. [30]
Illustrative scenario brackets (not forecasts)
| Revenue ($B) | FY2025A | FY2026 | FY2027 | FY2028 |
|---|---|---|---|---|
| Bear: Q4 flat QoQ; +30%, +20% | ~4.6 | 49.9 | 64.9 | 77.8 |
| Reference: Q4 +25% QoQ; +60%, +40% | ~4.6 | 54.1 | 86.6 | 121.2 |
| Bull: Q4 +45% QoQ; +100%, +60% | ~4.6 | 57.5 | 114.9 | 183.9 |
FY2026 starts from nine-month revenue of $33.1B (Q1–Q3 sum) and varies only Q4; later growth rates are bracketing assumptions chosen to span outcomes, not estimates. FY2025 actual and quarterly inputs: [29][1]
9Valuation
On run-rate revenue the IPO band sits at 22–30x, close to the multiple paid in the Series G and above the Series H, before any adjustment for gross-versus-net revenue.
Private mark stack, reconciled to run-rate revenue
| Mark | Date | Equity value | Run-rate at the time | Implied multiple (derived) |
|---|---|---|---|---|
| Series G post-money | Feb 2026 | $380B [63] | $14B [74] | 27.1x |
| Series H post-money | May 2026 | $965B [3] | >$47B [3] | ≤20.5x |
| IPO talk, low | Oct 2026 | ~$1.5T [4] | $67.2B (Q3 × 4) [1] | 22.3x |
| IPO talk, high | Oct 2026 | ~$2T [75] | $67.2B (Q3 × 4) [1] | 29.8x |
Memeburn puts a $2T price at about 30x annualized Q3 revenue and about 2.1x the Series H mark. [76][77] Aswath Damodaran's independent work puts enterprise value at about $1.3T. [78]
EV-to-equity bridge
Private marks are post-money equity values. Net cash after the Series H, the face value of convertible notes and the post-IPO share count are not disclosed in the reviewed coverage, so enterprise value is shown as —. The $72B H1 2026 charge on convertible instruments indicates the conversion terms are material to fully diluted share count. [79]
Net-revenue sensitivity
OpenAI books revenue net of partner shares and alleges Anthropic overstates revenue by counting the full value of contracts sold through third-party clouds. [19] Anthropic's split of cloud-resold revenue is not in the reviewed coverage, so Section 11 shows multiples on the reported base and with illustrative 15% and 30% haircuts.
Listed reference points (market data, 20 Oct 2026 close)
| Company | Market cap | EV / sales | Relevance |
|---|---|---|---|
| NVIDIA | $5.23T [80] | 7.4x [81] | Chips rented via Nscale and Lambda (up to $86B) [46] |
| Alphabet | $4.16T [82] | 6.5x [83] | $111B Google Cloud commitment [46] |
| Microsoft | $3.95T [84] | 10.0x [85] | $31B Microsoft commitment [46] |
| SpaceX (incl. xAI) | $1.95T [86] | 91.9x [87] | June 2026 IPO; includes xAI [88] |
| Palantir | $462.8B [89] | 55.3x [90] | Listed enterprise AI software |
| CoreWeave | $34.3B [91] | 10.6x [92] | Listed compute lessor [93] |
EV/sales uses the data provider's default period for each company, which differs by fiscal calendar. Peer set comparison: [94]
10Key debates & what's priced in
At $1.5T–$2T the market would be paying today for revenue of $150B–$200B on a Microsoft-like multiple, roughly 2.2–3.0x the Q3 run-rate.
What's priced in (undiscounted, mechanics only)
| Equity value | Revenue needed at 6.5x (Alphabet EV/sales) | Revenue needed at 10x (Microsoft EV/sales) | 10x case vs Q3 run-rate |
|---|---|---|---|
| $965B Series H [3] | $148.5B | $96.5B | 1.4x |
| ~$1.5T IPO talk [4] | $230.8B | $150.0B | 2.2x |
| ~$2T IPO talk [75] | $307.7B | $200.0B | 3.0x |
Required revenue = equity value ÷ peer multiple, ignoring net cash, dilution and time value. Inputs: Alphabet 6.5x [83], Microsoft 10.0x [85], Q3 run-rate $67.2B from Q3 revenue of $16.8B. [1]
Falsifiable conditions
- Q4 2026 revenue at or above Q3's $16.8B, keeping sequential growth positive after the slowdown from Q2. [1]
- The public filing states the gross or net basis of cloud-resold revenue, settling the recognition dispute. [19]
- Compute spend keeps falling as a share of revenue from the ~55% implied by H1 2026. [30]
- Two-customer concentration falls below the ~20% of H1 2026 revenue. [20]
Debate 1 — Is growth decelerating or normalizing?
Debate 2 — Can revenue carry the compute bill?
Debate 3 — How good is the revenue?
Debate 4 — Does the safety-first structure help or hinder holders?
Risk register
| Category | Risk | Evidence | Which way it cuts |
|---|---|---|---|
| Financial | Fixed compute vs variable revenue | $1.25B a month fixed through May 2029 against usage-linked revenue. [18] | Demand shortfall compresses margins; higher demand turns the fixed base into operating leverage. |
| Financial | Convertible remeasurement | $72B H1 2026 charge on instruments convertible into shares. [72] | Non-cash, but signals dilution from conversion. |
| Financial | Revenue recognition | Gross vs net dispute over cloud-resold contracts. [19] | A net restatement lowers the base for any multiple. |
| Operational | Compute access | Requirements exceeded what some cloud partners were willing to provide. [100] | Capacity gaps cap revenue; new suppliers diversify. |
| Operational | Data-center opposition | Organized political resistance named as a risk. [35] | Delays to capacity ramps. |
| Competitive | OpenAI capital race | OpenAI targeting ~$30B above $1.4T before the debut. [101] | Sets a private comparator and competes for the same allocators. |
| Competitive | Open-weight substitution | Open-weight models gaining share, per a Microsoft AI lead. [45] | Pricing pressure on commodity workloads. |
| Competitive | Customer concentration | Two customers ~20% of H1 2026 revenue. [20] | Single-account churn moves group growth. |
| Governance & safety | Catastrophic incident | Financial consequences could exceed available insurance. [20] | Tail liability outside normal provisioning. |
| Governance & safety | Unauthorized agent actions | Incidents of models accessing external systems without authorization have been reported by OpenAI and Anthropic. [102] | Regulatory and reputational exposure. |
| Governance & safety | Trust control of the board | Class T stock lets the trust elect a board majority. [21] | Mission priorities can override holder preferences. |
| Market & offering | IPO window | Elice fell 32% on debut and Firmus scrapped its IPO; some CoreWeave bonds yield near 15%. [93] | Weak tape pressures price and size. |
| Market & offering | Supply overhang | SpaceX's first lock-up expiry around 9 December lands within weeks of the listing. [13] | Competing supply for AI-exposed capital. |
11Scenarios & sensitivity
At a Microsoft-like 10x, only the bull revenue bracket reaches the IPO band by FY2028; the reference bracket needs 12–17x to get there.
Implied-value ranges on illustrative FY2028 revenue (undiscounted)
| Bracket | FY2028 revenue | At 6.5x | At 10x | At 15x |
|---|---|---|---|---|
| Bear | $77.8B | $506B | $778B | $1.17T |
| Reference | $121.2B | $788B | $1.21T | $1.82T |
| Bull | $183.9B | $1.20T | $1.84T | $2.76T |
Revenue brackets from Section 8; multiples bracket Alphabet 6.5x [83] and Microsoft 10.0x [85], with 15x as an upper bracket. No probabilities are attached.
Illustrative revenue brackets, $B; FY2025 actual ~$4.6B. [29]
Sensitivity: implied multiple of annualized Q3 revenue
| Equity value | Reported base $67.2B | 15% haircut ($57.1B) | 30% haircut ($47.0B) |
|---|---|---|---|
| $1.00T | 14.9x | 17.5x | 21.3x |
| $1.25T | 18.6x | 21.9x | 26.6x |
| $1.50T (reference) | 22.3x | 26.3x | 31.9x |
| $1.75T | 26.0x | 30.6x | 37.2x |
| $2.00T | 29.8x | 35.0x | 42.5x |
| $2.25T | 33.5x | 39.4x | 47.8x |
Haircuts are illustrative stand-ins for a possible net-revenue presentation; the actual split is undisclosed. Base: Q3 revenue $16.8B × 4. [1]
Interactive mechanics calculator
Starts from annualized Q3 revenue of $67.2B (Q3 $16.8B × 4). [1] Outputs are arithmetic, not estimates.
Revenue base
$67.2B
Implied multiple today
22.3x
Revenue needed at exit multiple
$150.0B
Needed ÷ current base
2.23x
Step-up vs $965B Series H
1.55x
5% Public Benefit Fund stake
$75B
12Comparable IPOs
SpaceX is the only listing of comparable scale, and it traded above issue, while smaller AI-adjacent deals this autumn have broken or been pulled.
| Deal | Datapoint |
|---|---|
| SpaceX (June 2026) | Listed at $135 a share, valuing it at $1.77T in the largest IPO ever, and closed at $147.96 on 19 October. [103] |
| SpaceX deal size | Reported as a $75B offering by Fortune. [60] Another report cites $86.3B. [56] |
| SpaceX structure | No sales by existing shareholders in the offering. [104] |
| SpaceX valuation | About 103x 2025 revenue at the 16 October market value, per Memeburn. [88] |
| Elice Group | South Korean AI cloud provider fell 32% on its first trading day. [105] |
| Oura | Shelved its IPO hours before pricing in September. [60] |
| Firmus | Nvidia-backed Australian data-center company scrapped its planned IPO. [93] |
Jay Ritter's data on about 9,300 US IPOs shows companies with more than $500M of revenue that rose 10% on day one underperformed the market by about 5% over three years. [78] SpaceX's rise from $135 to $147.96 is about 9.6% (derived). [103]
13Management & governance
Founder-led management sits under a trust with growing board control, so mission priorities are written into the share structure holders will buy.
Leadership
| Name | Role | Evidence |
|---|---|---|
| Dario Amodei | CEO, co-founder | 2025 total compensation of about $18M, largely stock and options. [35] |
| Daniela Amodei | President, co-founder | 2025 total compensation of about $16.5M. [35] |
| Krishna Rao | CFO | Company spokesman on the Series G and the Series H. [106][23] |
| Jack Clark | Co-founder | Appeared alongside the IMF Managing Director at the fund's annual meetings. [107] |
| Chris Ciauri | MD, International | Quoted on the Hexaware preferred-partner agreement. [108] |
| Chris Liddell | Board director | Appointed to the board of directors. [109] |
Structure
- Anthropic plans to list as a public benefit corporation, which requires management to weigh broader interests alongside investor returns. [110]
- The Long-Term Benefit Trust elects directors through Class T stock; it currently elects three and gains the right to elect a majority within four years of the Series C close. [21]
- The Public Benefit Fund's equity will not be used for lobbying or political contributions. [111]
- Fund governance details and the valuation milestones that add to its stake are still to be published. [97]
Three-layer read: trust, fit, execution
Trust. The co-founders pledged 80% of their equity to donation at founding, and the company is directing equity to the public benefit in place of escalating executive awards. [66] A Morning Coffee column characterized the Amodeis' disclosed pay as good but not outlandish. [112]
Fit. Management frames the capital raises around enterprise demand, with the CFO saying customers describe Claude as increasingly critical to how businesses work. [106] Board control migrates to the trust over time, tying governance to the stated mission. [21]
Execution. Run-rate revenue went from $14B at the Series G to above $47B at the Series H. [74][3] A September listing was expected before a researcher quit and safety incidents followed. [10] Researcher Jacob Coxon resigned in September and warned publicly about unchecked AI development. [113]
14Perspectives
Outside voices split along the same lines as the debates: growth admirers, valuation skeptics and safety critics, each shown separately below.
Ark Investment Management (Brett Winton)
Ark Investment Management's Winton calls Anthropic's past 12 months the most remarkable growth of any company in history. [32] Winton also links the recent cooling to the competitive fight with OpenAI. [33]
Aswath Damodaran (NYU)
Damodaran's independent valuation puts Anthropic's enterprise value at about $1.3T. [78]
Investors quoted by Axios
Investors told Axios the IPO timing is not optional because the company has grown too big for private fundraising. [114]
NYC Comptroller Mark Levine
Levine wrote to the CEO asking Anthropic to pause the IPO and focus on AI risk. [115]
Expert call: former CoreWeave infrastructure engineer
The engineer expects AI infrastructure to consolidate around one or two big winners. [116] The engineer sees the two big AI labs as best placed to use heterogeneous chip fleets. [117]
Expert call: IBM associate partner, ex-Amazon ML
This expert, who worked directly with Anthropic and AWS, is skeptical of the existential-risk narrative and sees regulatory capture as a motive. [118]
Retail sentiment (DealMaker survey)
63% of US adults surveyed said trillion-dollar IPOs could leave retail buyers paying inflated prices while early investors sell. [75]
Sell-side broker research is excluded from this primer by request.
15Playbook / what to watch
The public filing flip is the single most information-dense event; everything after it narrows the range rather than resets it.
| Event / metric | Expected timing | What it reveals | What would falsify the current setup |
|---|---|---|---|
| Public S-1 and price range | Before a roadshow targeted for the week of 9 Nov [58] | Price range against the $965B Series H, primary vs secondary mix, lock-up tiers, float. [13] | A range below the Series H mark, or a float dominated by existing-holder sales. |
| Revenue basis disclosure | Public filing | Gross vs net treatment of cloud-resold contracts. [19] | A net presentation materially below the $16.8B Q3 figure. |
| Gross margin and inference cost | Public filing | Whether adjusted Q3 profitability holds on a GAAP cost-of-revenue basis. [12] | Negative gross margin after compute. |
| Commitment schedule | Public filing | Year-by-year phasing of the $518B, ~80% noncancellable, commitments. [5] | Front-loaded payments that exceed revenue growth. |
| Convertible terms | Public filing | Conversion price and share count behind the $72B remeasurement. [72] | Dilution that lowers per-share value at the talked-about band. |
| OpenAI round pricing | Before mid-November [119] | A private comparator at above $1.4T. [11] | A down-sized or lower-priced OpenAI round. |
| SpaceX first lock-up expiry | Around 9 December [13] | Supply of AI-exposed mega-cap stock. | Heavy selling into the Anthropic aftermarket window. |
| Q4 2026 revenue | First public quarter | Whether sequential growth stays positive after Q3's $16.8B. [1] | A flat or declining quarter. |
| Customer concentration | Each filing | Share from the top two customers, ~20% in H1 2026. [20] | A rising share or a top-customer loss. |
16Methodology & glossary
Every figure in this primer maps to the audit table below; derived numbers are labeled and uncited, and nothing here is a forecast or a recommendation.
Anchor-numbers audit
| Display | Definition / arithmetic | Period | Basis | Source |
|---|---|---|---|---|
| $16.8B | Quarterly revenue | Q3 2026 | Company statement to investors, as reported | [1] |
| $11.6B / $4.7B | Quarterly revenue | Q2 / Q1 2026 | As reported | [1] |
| $67.2B | $16.8B × 4 | Q3 2026 annualized | Derived | Derived from [1] |
| $16.3B vs $1.1B | Half-year revenue | H1 2026 vs H1 2025 | Draft prospectus, as reported | [2] |
| ~$4.6B | Annual revenue, ~12x YoY | FY2025 | Prospectus, as reported | [29] |
| $3.8B / $789M | Usage / subscription revenue | FY2025 | As reported | [14] |
| $80.7B / $72B | Net loss / convertible charge | H1 2026 | Draft prospectus, as reported | [72] |
| ~$8.7B | Net loss excluding the $72B charge | H1 2026 | Media calculation | [69] |
| $8.06B / $2.98B | Operating loss excl. liability writedowns | FY2025 / FY2024 | Prospectus, as reported | [29] |
| ~159% / ~55% | Compute ÷ revenue: $7.33B ÷ $4.6B; $9B ÷ $16.3B | FY2025 / H1 2026 | Derived | Derived from [30] |
| $518B | Compute commitments, ~80% noncancellable | Decade | Prospectus, as reported | [5] |
| $1.25B / month | Fixed compute commitment | Through May 2029 | As reported | [18] |
| $183B / $380B / $965B | Post-money marks | Sep 2025 / Feb 2026 / May 2026 | Primary priced rounds | [62][63][3] |
| $108B | $13B + $30B + $65B primary raised | Series F–H | Derived | Derived from round rows in Section 7 |
| $1.5T / $2T | IPO valuation discussion / earlier target | Oct 2026 | Reported talk, not a price range | [4][75] |
| 22.3x / 29.8x | $1.5T or $2T ÷ $67.2B | Q3 2026 annualized | Derived | Derived |
| 27.1x / ≤20.5x | $380B ÷ $14B; $965B ÷ >$47B run-rate | At Series G / H | Derived | Derived from [74][3] |
| $150B / $200B | $1.5T or $2T ÷ 10x | Undiscounted | Derived | Microsoft multiple 10.0x [85] |
| $49.9B–$57.5B | $33.1B nine-month revenue + Q4 at 0% to +45% QoQ | FY2026 | Illustrative bracket | Derived |
| 5.8% | $86.3B ÷ $1.5T, if entirely primary | Hypothetical | Derived | Derived from [56] |
Glossary
| Term | Meaning in this primer |
|---|---|
| Run-rate | Revenue for a recent period annualized; not a GAAP measure. |
| Post-money mark | Equity value implied by a priced primary round, including new money. |
| Gross vs net revenue | Whether revenue sold through a cloud partner is booked at the full contract value or after the partner's share. |
| Long-Term Benefit Trust / Class T stock | Independent trust holding a special share class that elects directors. |
| Public benefit corporation | For-profit entity required to weigh a stated public benefit alongside shareholder returns. |
| Noncancellable commitment | Contracted spend owed regardless of actual usage. |
| Illustrative bracket | Assumption-driven range used to bound outcomes; not an estimate, forecast or recommendation. |
Method: Focus Five and the trust–fit–execution read are expressed as cited evidence, without grades or scores. Valuation is mechanics only. No price target, rating or position view is expressed.
17Sources
The primer rests on company releases, reporting on the confidential draft prospectus, expert-call transcripts, partner announcements and market data; sell-side broker research was excluded.
- Company publications: funding announcements, the Public Benefit Fund commitment and research reports.
- Draft prospectus coverage: media reporting on the confidential filing; the public registration statement is not yet available, so all filing figures are unconfirmed.
- Expert calls: practitioners at cloud providers, enterprise software companies and infrastructure operators; each view is shown separately.
- Market data: listed peer market capitalization and EV/sales as of the 20 October 2026 close.
- Not covered: rounds before Series F other than Series A, strategic holder ownership percentages, gross margin, net cash, convertible note terms, price range and lock-up schedule.
Numbered references follow.
References
- [1] Anthropic IPO: Investors weigh USD 1.5 trillion valuation as AI spending concerns mount (2026-10-20)
- [2] [4] Anthropic Reports $80.7 Bn First-half Loss Ahead Of IPO (2026-10-21)
- [3] [24] [64] Anthropic raises $65 billion in Series H funding (2026-05-28)
- [5] [34] [98] [99] Amodei thanks critic Calvin as Anthropic rewrites AI safety bill (2026-10-21)
- [6] Anthropic logged net loss of over $80 billion in first half of 2026 (2026-10-21)
- [7] Cipher Digital, Fluidstack, and Anthropic Commit $10 Million to Repair and Expand Colorado City's Water System (2026-09-14)
- [8] Expensify Becomes One of the First Expense Management Tools Available in Claude for Small Business (2026-09-15)
- [9] [10] [48] [93] [95] [114] Why Anthropic can't afford to delay its IPO (2026-10-21)
- [11] OpenAI races to raise $30 billion at $1.4 trillion valuation ahead of Anthropic IPO (2026-10-20)
- [12] [102] [107] Is the AI Boom on the Verge of Cooling? ETF Areas to Play (2026-10-20)
- [13] [59] [75] [76] [77] [78] [88] [104] Who Made Money on the SpaceX IPO? A Warning for Anthropic (2026-10-21)
- [14] [18] Anthropic IPO: What ETF Investors Should Know (2026-10-02)
- [15] HCLTech to establish a dedicated Anthropic Business Unit and Center of Excellence to accelerate deployments of Claude in enterprise environments (2026-10-12)
- [16] [63] [74] [106] GIC Leads $30 Billion Series G in Anthropic (2026-02-12)
- [17] [19] [42] [47] [119] OpenAI Rushes To Close $30 Billion Funding Round Ahead Of Anthropic's IPO (2026-10-20)
- [20] [46] Anthropic's confidential IPO filing reveals $80.7 billion H1 loss, massive computing costs (2026-10-21)
- [21] [50] Anthropic appoints Harvard astrophysicist Christopher Stubbs to its Long-Term Benefit Trust (2026-10-21)
- [22] [69] Anthropic’s IPO Bets Big That Demand Will Outrun the Bills (2026-10-21)
- [23] Anthropic raises $65B in Series H funding at $965B post-money valuation (2026-05-28)
- [25] Expensify Becomes One of the First Expense Management Tools Available (2026-09-15)
- [26] [108] Hexaware Becomes an Anthropic Preferred Partner, Bringing Claude to the Core of Its AI-native Platforms (2026-10-08)
- [27] Over 50 Projects with Claude Take CI&T to a New Partnership Tier (2026-10-14)
- [28] [54] Anthropic opens Singapore office, signs AI safety pact with government (2026-10-15)
- [29] Anthropic IPO Filing: $42 Billion Loss, $2 Trillion Valuation Target And A Warning Of 'Existential' Risks (2026-09-29)
- [30] [35] [44] [71] Anthropic Lost $80 Billion in Six Months as AI Spending Soars, Draft IPO Filing Shows (2026-10-20)
- [31] Anthropic’s S-1 leaks (again)—here’s a look at the massive losses on the income statement (2026-10-21)
- [32] [33] Frontier AI revenue growth remains extraordinary despite Anthropic slowdown (2026-10-20)
- [36] [37] Enterprise Sales Professional at Amazon Web Services Believes AI Model Selection Hinges on Governance and Task-Specific Needs (2026-09-25)
- [38] Senior Director of Martech, Analytics & AI at Microsoft Corporation Believes AI Budgeting Is Evolving Towards Efficiency and Value-Driven Approaches (2026-08-13)
- [39] Head of Artificial Intelligence at Pocket FM Believes Closed Source AI Models Retain Pricing Power Amidst Evolving Competitive Landscape (2026-10-21)
- [40] Former General Manager at Microsoft Discusses Challenges and Strategies in Hyperscaler Demand Forecasting and Capacity Planning (2026-10-14)
- [41] Not Anthropic. Not OpenAI. This Is the Artificial Intelligence (AI) Stock I'd Buy Before the Next Mega-IPO. (2026-10-20)
- [43] Anthropic新模型曝光：未明說目標也能推斷，大模型競爭轉向「讀懂意圖」 (2026-10-20)
- [45] AI Function Lead at Microsoft Believes Open-Weight Models Are Rapidly Gaining Market Share Over Frontier Models (2026-10-08)
- [49] How Employment isChanging at Anthropic (2026-10-16)
- [51] AMD and Anthropic Trade Crossed Investment and Compute Commitments (2026-10-21)
- [52] [111] Anthropic to commit 5 percent of shares to 'public good' (2026-10-19)
- [53] [115] NYC Comptroller Levine Calls for Pause in Anthropic IPO Plans (2026-10-16)
- [55] Anthropic to Pay Akamai Technologies $11.6 Billion for Cloud Services | Swissquote (2026-09-25)
- [56] Anthropic targets mid-November IPO after delay (2026-10-02)
- [57] [112] Morning Coffee: Anthropic is making bank CEOs look overpaid. Hedge funds have to decide who’s really a partner (2026-10-21)
- [58] Anthropic Accelerates Plans for Pre-Thanksgiving Public Debut (2026-10-03)
- [60] [105] Is Anthropic the IPO market’s last hope for 2026? The clock is ticking (2026-10-21)
- [61] Eric Schmidt’s path to a $20 billion Anthropic windfall began with an introduction from his ex, Cami Clark—now Dario Amodei’s wife (2026-10-21)
- [62] Anthropic: $13 Billion Series F Raised At $183 Billion Valuation To Advance AI Products (2025-09-03)
- [65] Anthropic Faces Scrutiny Over Investor Stakes, Cloud Strains and Draft AI Safety Bill Ahead of Expected IPO (2026-10-21)
- [66] [97] Anthropic Public Benefit Fund Commitment (2026-10-18)
- [67] Anthropic and Musk offer rival blueprints for sharing AI's wealth (2026-10-19)
- [68] Anthropic’s $80.7 billion loss puts its IPO ambitions under scrutiny (2026-10-21)
- [70] Anthropic IPO: How Did $4.6 Billion in Revenue Become a $42 Billion Loss? (2026-09-29)
- [72] Anthropic Reports $80.7B H1 Loss Ahead Of Potential November IPO (2026-10-21)
- [73] New ETF Provides Backdoor Access to Anthropic Investment Before Public Debut (2026-09-23)
- [79] Anthropic logged net loss of over $80 billion in first half of 2026: Report (2026-10-21)
- [80] [81] [82] [83] [84] [85] [86] [87] [89] [90] [91] [92] Financial Data
- [94] Screener
- [96] Anthropic logged net loss of over $80 billion in first half of 2026, WSJ reports (2026-10-21)
- [100] Anthropic's confidential IPO filing reveals USD 80.7 billion H1 loss, massive computing costs [Asian News International] (2026-10-21)
- [101] OpenAI Seeks $1.4 Trillion Valuation as Anthropic IPO Looms (2026-10-21)
- [103] [113] SpaceX, Anthropic plan to share wealth with the public amid concern AI could trigger job losses [Los Angeles Times] (2026-10-20)
- [109] Anthropic partners with CodePath to redesign AI-powered coding curriculum (2026-02-13)
- [110] Anthropic to allocate 5% of shares to ‘Public Benefit Fund’ as AI safety concerns grow (2026-10-21)
- [116] [117] Former Infrastructure Engineer at CoreWeave Believes Current AI Infrastructure Economics Are Unsustainable and Predicts Industry Consolidation (2026-10-13)
- [118] Associate Partner at IBM Views AI Safety Narratives as Strategic Moves for Regulatory Advantage (2026-10-08)