S-1 filed 26 October 2026 Nasdaq Global Select Market: ANT Price and size not yet set
Anthropic IPO: the S-1 figures, explained for UK business
Anthropic, PBC filed a preliminary Form S-1 with the US Securities and Exchange Commission on 26 October 2026, applying to list Class A shares on the Nasdaq Global Select Market under the ticker ANT. The price range, share count and offering size are blank.
The filing reports revenue of $16.3bn for the six months to 30 June 2026, against $4.6bn for the whole of 2025, a GAAP operating loss of $693m for the half with $974m of operating income in the second quarter, and a net loss of $80.7bn of which $72.4bn is a non-cash fair-value charge.
Last updated: 28 October 2026 Filing checked: 28 October 2026, no amendment on file
The filing in six figures
Each tile names its period and its line. The six-month figures are unaudited interim results, not full-year estimates, and the three loss measures on this page are different things.
Revenue, six months to 30 June 2026
$16.3bn
Against $4.6bn for the whole of 2025. Unaudited interim figure.
S-1, Results of Operations
Operating income, Q2 2026 (GAAP)
$974m
The first quarter with operating income. The six-month result was still a loss of $693m.
S-1, quarterly results, unaudited
Net loss, six months to 30 June 2026
$80.7bn
Of which $72.4bn is the change in fair value of convertible notes, derivatives and SAFEs, a non-cash accounting charge.
S-1, Results of Operations
Cash and short-term investments, 30 June 2026
$131.7bn
Our sum of $112.4bn cash and $19.3bn short-term investments. Excludes restricted cash.
S-1, consolidated balance sheet
Customers with $100,000 or more of trailing-year revenue
more than 11,900
At 30 September 2026. The filing's proxy for enterprise adoption, not a count of users or seats.
S-1, customers and adoption
Offer price, share count and size
Not set
Preliminary S-1 filed 26 October 2026. Application to list on the Nasdaq Global Select Market as ANT.
S-1, cover page
What is set, and what is still open
A preliminary S-1 establishes the security, the market and the use of proceeds. It does not establish the price, the size or the date. The SEC’s company submissions feed, checked on 28 October 2026, listed no amendment to the 26 October 2026 filing.
| Issue | What the filing establishes | What remains open |
|---|---|---|
| Security and market | Class A common stock; application to list on the Nasdaq Global Select Market under ANT | Admission, and the completed offering |
| Primary and secondary shares | Both the company and selling stockholders are contemplated as sellers | Number of shares, which holders sell, and the final allocation |
| Price and valuation | A preliminary registration statement with the price range left blank | Price range, offer price and the implied market value |
| Use of proceeds | General corporate purposes, working capital, capital expenditure including compute, hiring and possible acquisitions; the company receives nothing from secondary sales | Net proceeds and their exact allocation |
| Timing | Registration process under way | Effective date, roadshow, pricing and first trading day |
| Underwriting and dilution | Offering mechanics are described | Final underwriting economics and complete post-offer dilution |
The $1 billion in the fee table is not a raise or a valuation
The filing-fee exhibit uses a $1.0bn proposed maximum aggregate offering price solely to calculate the registration fee under Rule 457(o). At the stated rate of 0.000087 the fee is $87,000. The figure says nothing about the intended capital raise, the final offer size or the company’s value.
A blank, future-dated figure to watch
The Run-Rate Revenue passage leaves the value blank and refers to 31 October 2026, a date after the filing. Several September financial and commitment fields are also unpopulated. No run-rate figure or September liquidity total has been reconstructed from those placeholders, here or anywhere on this page.
Revenue growth, and the base it grew from
Revenue was $4.6bn in 2025, up 1,087.8% on 2024, and $16.3bn in the first six months of 2026, up 1,300.3% on the $1.2bnof the same period a year earlier. Both percentages agree with the filing’s rounded figures. They describe rapid expansion from a small base; they are not a forecast, and annualising a half-year or a strong quarter would not establish contracted or repeatable annual revenue.
Anthropic revenue by period, as filed
View the data
| Period | Revenue | Audited |
|---|---|---|
| 2023 | 41 | Yes |
| 2024 | 386 | Yes |
| 2025 | 4,585 | Yes |
| H1 2025 | 1,163 | No |
| H1 2026 | 16,286 | No |
Quarterly revenue, Q1 2025 to Q2 2026
View the data
| Quarter | Revenue | Compute and infrastructure |
|---|---|---|
| Q1 2025 | 376 | 944 |
| Q2 2025 | 787 | 1,390 |
| Q3 2025 | 1,382 | 2,047 |
| Q4 2025 | 2,040 | 2,945 |
| Q1 2026 | 4,731 | 3,387 |
| Q2 2026 | 11,555 | 5,558 |
About 80% of first-half revenue was consumption-based, meaning token and usage charges on the Claude Platform and through cloud marketplaces. Subscriptions to Claude Pro, Max, Team and Enterprise were about 20%. For a buyer, that mix is the point: the vendor’s revenue grows when your usage grows, so task design and usage behaviour are financially material on both sides.
Consumption against subscription revenue
- Consumption
- Subscriptions
View the data
| Period | Consumption | Consumption share | Subscriptions | Subscription share |
|---|---|---|---|---|
| 2024 | 288 | 75% | 98 | 25% |
| 2025 | approx. 3,800 | 83% | approx. 789 | 17% |
| H1 2026 | approx. 13,100 | 80% | approx. 3,200 | 20% |
Three different losses
For the six months to 30 June 2026 the filing reports a GAAP operating loss of $693m, a margin of -4.3%; an adjusted operating income of $1.6bn, a non-GAAP margin of 9.9%; and a net loss of $80.7bn. All three are correct. They measure different things, and a headline that quotes one as if it were another is wrong.
Three different losses, six months to 30 June 2026
Values in US$ million as filed. The fair-value line is not cash spent. It is also not irrelevant, because it reflects value transferred to financing counterparties and dilution to come.
View the data
| Line | H1 2026 |
|---|---|
| Operating loss (GAAP) | (693) |
| Fair-value remeasurement | (72,388) |
| Net loss | (80,748) |
What the adjusted measure removes
The company adds $2.2bn of stock-based compensation and $126m of non-cash charitable contributions back to the $693m operating loss to reach $1.6bn of adjusted operating income. In 2025 it also removes the $1.5bnBartz copyright settlement. These are the company’s exclusions. Equity compensation and share donations are not economically irrelevant: both dilute shareholders.
The quarter that turned
The second quarter of 2026 produced $974m of GAAP operating income on $11.6bn of revenue, a margin of 8.4%, and $2.2bn on the adjusted basis, a margin of 19.3%. The half-year as a whole was still a loss. One quarter of operating income is meaningful; it is not yet a durable earnings record, and the filing itself cautions that quarterly results are not indicative of a full year.
Quarterly operating result: GAAP against the company's adjusted measure
- GAAP operating result
- Adjusted operating result (non-GAAP)
Values in US$ million as filed; parentheses are losses. Quarterly figures are unaudited.
View the data
| Quarter | GAAP operating result | Adjusted operating result |
|---|---|---|
| Q1 2025 | (935) | (830) |
| Q2 2025 | (1,200) | (1,019) |
| Q3 2025 | (3,089) | (1,284) |
| Q4 2025 | (2,836) | (1,721) |
| Q1 2026 | (1,667) | (617) |
| Q2 2026 | 974 | 2,235 |
Why there is no gross margin
The company presents expenses by nature, not by function, and recast the comparative periods on that basis. The compute and infrastructure line covers research, training, inference and internal use together, so revenue minus compute is not a cost-of-revenue calculation and the filing offers no gross-profit subtotal. The ratio below is the nearest honest substitute: revenue divided by the quarter’s compute expense rose from 0.40x in Q1 2025 to 2.08x in Q2 2026.
Revenue per dollar of compute and infrastructure expense
Calculation by The AI Consultancy from the filed quarterly figures. The filing presents no gross-profit subtotal, so this is not comparable with a software company’s gross margin.
View the data
| Quarter | Revenue (US$m) | Compute (US$m) | Ratio |
|---|---|---|---|
| Q1 2025 | 376 | 944 | 0.40 |
| Q2 2025 | 787 | 1,390 | 0.57 |
| Q3 2025 | 1,382 | 2,047 | 0.68 |
| Q4 2025 | 2,040 | 2,945 | 0.69 |
| Q1 2026 | 4,731 | 3,387 | 1.40 |
| Q2 2026 | 11,555 | 5,558 | 2.08 |
Cash, and where it came from
At 30 June 2026 the company held $112.4bn in cash and cash equivalents and $19.3bn in short-term investments, which we sum to $131.7bn. A further $2.5bn of restricted cash is excluded from that sum. The balance was financed rather than earned: first-half financing inflows were $110.8bn, including $110.4bn of net proceeds from preferred stock, against $3.9bn from operations.
| Line | 31 December 2024 | 31 December 2025 | 30 June 2026 |
|---|---|---|---|
| Cash and cash equivalents | 3,141 | 17,388 | 112,389 |
| Short-term investments | 0 | 2,889 | 19,262 |
| Restricted cash | 13 | 13 | 2,529 |
| Accounts receivable | 95 | 909 | 5,833 |
| Total assets | 7,752 | 24,661 | 147,358 |
| Convertible notes at fair value | 14,116 | 41,677 | 108,566 |
| Total liabilities | 14,679 | 45,397 | 117,467 |
| Redeemable convertible preferred stock | 2,874 | 29,769 | 157,510 |
| Total stockholders' deficit | (9,801) | (50,505) | (127,619) |
| Activity | 2025 | H1 2025 | H1 2026 |
|---|---|---|---|
| Net cash from operating activities | (2,637) | (1,513) | 3,929 |
| Net cash from investing activities | (3,077) | (49) | (17,181) |
| Net cash from financing activities | 19,960 | 4,749 | 110,774 |
The $17.2bn investing outflow was mostly purchases of investments net of maturities and sales; cash purchases of property and equipment were $482m. Labelling the whole investing line as capital expenditure would be wrong. The $108.6bn of convertible notes on the balance sheet is a fair-value carrying amount, not a cash bill: the principal outstanding on the Amazon notes was $6.2bn. And a June balance is not an October one. The preliminary filing’s blank September fields and its future commitments mean no reliable cash-runway figure can be drawn from it, so this page does not offer one.
Customers, and the approximately 900% figure
The filing uses "customer" for a prosumer or an organisation that uses Anthropic's models, platform or applications, and aggregates accounts belonging to the same organisation when it applies revenue thresholds. It defines no separate enterprise customer category; the $100,000 trailing-twelve-month revenue threshold is its proxy for enterprise adoption.
Customers with $100,000 or more of trailing-twelve-month revenue
View the data
| Date | Customers | Filing's qualifier |
|---|---|---|
| 31 December 2023 | 20 | approximately |
| 31 December 2024 | 200 | approximately |
| 30 September 2025 | 950 | more than |
| 31 December 2025 | 1,500 | approximately |
| 30 September 2026 | 11,900 | more than |
| Metric | Filing's figure | How to read it |
|---|---|---|
| Customers at $100,000 or more TTM revenue | more than 11,900 | From about 20 at 31 December 2023. Rounded counts; the filing's enterprise proxy. |
| Customers at $1 million or more | more than 1,900 | A subset of the $100,000 group. |
| Customers at $10 million or more | approximately 250 | A subset of the $1 million group. |
| Customers below $100,000 | approximately 40,000,000 | Not a count of active people, paid seats or monthly users; the filing gives none of those. |
| Fortune 500 and Global 2000 penetration | over 50% and over 30% | Companies meeting the $100,000 threshold, not all of their employees using Claude. |
| Concentration | No customer above 10% | No single customer represented more than 10% of revenue for the nine months ended 30 September 2026. The filing does not name its largest customers. |
| Revenue outside the top five customers | Q1 2026: 64%; Q3 2026: 79% | Our derived top-five share: about 21% in Q3 2026. A different period from the nine-month concentration test. |
Net revenue retention needs its denominator
The approximately 900% figure compares revenue in the latest three months with the corresponding three months a year earlier, for an eligible prior-year cohort that generated at least $1, and excludes fixed-price Claude Pro and Max subscriptions. It captures extraordinary spend expansion by existing eligible customers. It is not a churn rate, not a share of customers retained and not a trailing-year ratio, and the arithmetic guarantees it falls as the comparison base grows. The filing does not provide a paid-to-free split, a monthly active user total or a seat count, and neither the revenue thresholds nor the approximately 40,000,000 smaller customers should be used to fill those gaps.
Compute commitments, kept apart
This is the section the headlines get wrong most often, usually by adding the rows below into a single “total debt” or “compute bill”. The arrangements are disclosed on different bases and dates. Some are minimum purchases that fall due regardless of usage, some are “up to” figures that either party can cancel after an initial period, some depend on delivery, acceptance or financing, and one is a schedule of equipment lease rentals. The filing leaves its own aggregate September commitment fields blank.
Disclosed compute and infrastructure arrangements, by counterparty
Why there is no total. These obligations are disclosed on different bases and dates. Some are minimum purchases, some are cancellable after an initial period, and some depend on delivery, acceptance or financing. The filing leaves its aggregate September commitment fields blank. Adding the rows into one total would mix a lease schedule with conditional capacity and a cancellable arrangement, so no total is shown.
The amber bar is a schedule of equipment lease rentals, a different kind of obligation from a cloud purchase commitment.
View the data
| Counterparty | Basis | Amount | Term | Qualification |
|---|---|---|---|---|
| Broadcom equipment leases | approximately | $161.2bn | Five-year equipment leases | Aggregate rentals excluding interest. Non-cancellable, with security, guarantees and default remedies. |
| SpaceX.AI | up to | $120.5bn | Terms extending into 2029 | Approximately $3.1bn a month at full capacity, expected in December 2026. After initial periods ending January and March 2027, either party can cancel on 90 days' notice. |
| at least approximately | $114.5bn | April 2026 to July 2033 | True-up regardless of usage. Further 2028 to 2031 TPU capacity, estimated at $125bn to $150bn, sits outside the fixed table and depends on further variables. | |
| Amazon | at least | $110bn | May 2026 to April 2036 | Usage-independent true-up, with breach exceptions. |
| Microsoft | at least | $44.9bn | November 2026 to May 2033 | Usage-independent commitment. Termination principally for uncured breach or undelivered tranches. |
| Nscale | up to | $44.6bn | Through 2033 | Delivery, acceptance and financing conditions, with specified termination rights. |
| Lambda | approximately | $41.3bn | Through 2034 | Delivery and acceptance. An August 2026 arrangement of $34.9bn includes a facility-lease condition. |
| AMD | estimated above | $20bn | Initial capacity from 2027 | Capacity and proposals are conditional. Not a fully fixed commitment. |
The combination of large fixed commitments and delivery conditions creates risk on both sides. Delays can restrict model availability; weaker demand or adverse pricing can leave contracted capacity underused. A large June cash balance does not resolve that by itself. For a UK client the relevant protections are application-level continuity, clear service assumptions and data portability, not the provider’s headline financing.
Funding facilities, conversion and dilution
Beyond its cash, the company describes conditional facilities with its cloud and hardware partners and a secured bank line. None of these is cash, each has conditions, and two of them must not be double counted: Google’s potential IPO participation sits inside its facility rather than on top of it, and the Amazon umbrella and its remaining undrawn capacity are one amount, not two.
| Arrangement | Headline | Conditions and significance |
|---|---|---|
| Google note purchase agreement | $20.0bn facility through December 2030 | Capacity and trailing-revenue conditions, scheduled closings and a 14.99% ownership cap unless waived. Up to $10bn of IPO participation reduces the facility dollar for dollar. |
| Amazon note purchase agreements | $20bn contractual umbrella; $15.0bn undrawn at 30 June 2026 | Milestone-based $5bn tranches and alternative equity funding. The umbrella and the remaining capacity are not two separate amounts. |
| Broadcom convertible notes | Up to $42.0bn tied to specified lease rent | Interest paid in kind at 9%, rising to 13% in specified breach cases; conversion at 90% of a market reference price, or 85% in the relevant breach case; cross-default and acceleration provisions. Revenue and liquidity conditions apply and the numerical thresholds are redacted. |
| AMD equity commitment | $5bn linked to milestones | First milestones expected in 2027; post-IPO Class N pricing linked to market value. Not unrestricted current cash. |
| Senior secured revolving credit facility | $15bn agreement dated 4 September 2026 | Replaces the earlier $3.25bn facility. Normal maturity 4 September 2031 with a springing-maturity mechanism; at least $5bn of liquidity required at quarter ends from 31 December 2026. Undrawn capacity is not cash. |
The Amazon notes carried $6.2bnof principal at 30 June 2026, and the filing’s conversion illustration describes about 1.84 billion shares in aggregate, subject to an ownership cap and contractual mechanics. Future conversion and Class N issuance can still have substantial economic dilution, which is a separate matter from the fair-value carrying amount on the balance sheet. Several numerical thresholds in the Broadcom agreement are redacted, so the cushion against those conditions cannot be verified from the public exhibit.
Who controls Anthropic after the listing
Anthropic is a Delaware public benefit corporation whose directors must balance stockholders’ financial interests, the interests of people materially affected by the company, and its stated public benefit: the responsible development of advanced AI for the long-term benefit of humanity. The IPO governance is designed to preserve that purpose, and founder influence, despite economic dilution.
| Class | Economic rights | Voting and control |
|---|---|---|
| Class A | Ordinary economic participation | One vote per share. Public investors do not control the trust-elected board majority. |
| Class N | Economic participation | Generally non-voting. Used for strategic holders and other stated purposes; generally converts to Class A on qualifying unaffiliated transfers. |
| Class T | Non-economic | Held by the Long-Term Benefit Trust, which generally elects and removes a majority of the authorised directors. |
| Class F | Non-economic | A single share held by the Founder LLC carrying a floating 50.1% of general voting power, excluding the separate trust election rights. |
| Class E | Non-economic | Eligible employees, excluding founders. Limited votes on specified trust changes and specified director decisions; allocation reflects tenure and pledged donations. |
What the structure does and does not do
- Public shareholders do not control the board.The Long-Term Benefit Trust generally elects and removes a majority of directors through its Class T shares, and the founders’ single Class F share carries 50.1% of general voting power. Anthropic expects to qualify as a Nasdaq controlled company and to rely on the exemption that lets its nominating and governance committee not be entirely independent.
- The trust is not unconstrained.Specified Class T director decisions need the concurrence of a committee that includes non-trust directors; if agreement is not reached within a 15-day process, eligible Class E employee voting can decide the outcome. “The trust appoints every director without constraint” would be inaccurate.
- There is no simple sunset. Class F runs on founder eligibility and succession rules rather than a fixed term: fewer than three qualifying original founders for 90 consecutive days starts a 450-day transition, after which the Class F and Class E shares are redeemed without payment. Once Class F has ended, changing the trust arrangements against trustee opposition needs 85% of outstanding Class A shares.
- Philanthropy and control are separate. A public benefit fund initially holds 846,840,960 Class N shares, described as 5% of the fully diluted June base assuming full Amazon conversion, with further grants at the board’s discretion. The founders’ 80% personal donation pledges are a different arrangement and do not remove founder voting control.
Our reading: the structure may protect safety commitments from short-term market pressure, and it reduces public shareholders’ ability to change leadership or strategy. The strength of the purpose protection and the strength of ordinary investor accountability are two different questions, and the filing is candid that it has chosen the first.
The risks the filing itself names
The S-1 devotes roughly a hundred pages to risk factors. The two tables select the ones that bear on a business deciding whether to depend on Claude. The priority labels are ours, based on business relevance and potential severity; they are not the company’s probability estimates.
| Priority | Risk | Filing evidence | What it means for a Claude buyer |
|---|---|---|---|
| Critical | Compute execution and fixed spending | Very large multi-year arrangements with readiness, delivery, acceptance and usage-independent payment terms; the aggregate September commitment fields are blank. | Capacity and price assumptions deserve review. Do not promise availability beyond contracted service terms. |
| High | Financing conditions and dilution | Conditional partner facilities, secured credit, conversion features, equity compensation and charitable shares. | Headline funding is not unconditional liquidity. Watch financing amendments and covenant disclosures. |
| High | Demand and competitive pricing | Rapid growth with consumption exposure, frontier investment, and competing models, cloud platforms and applications. | A client's unit economics must survive lower usage, higher usage and provider price changes. |
| High | Partner interdependence | Cloud suppliers are also investors, distribution channels and potential competitors. | Contract and data portability should cover the model layer and the cloud layer. |
| High | Customer expansion normalisation | Very high cohort revenue retention and rapid spend growth from a small base. | Do not assume a UK SME will replicate large early adopters' spending or return. |
| High | Limited public shareholder influence | Trust board-election rights and the founders' 50.1% voting structure. | Mission continuity may be strong; external accountability and strategic flexibility are constrained. |
| Priority | Risk | What is disclosed | Practical implication |
|---|---|---|---|
| High | Model access and government action | The filing describes export restrictions and conflicting outcomes in government-access litigation. | Check geography, use case and model eligibility at deployment and at renewal. |
| High | Copyright and content claims | A $1.5bn Bartz settlement expense in 2025; other claims and legal uncertainty remain. | A settlement does not establish that all training, output or customer use is free of IP risk. |
| High | Agent and cyber behaviour | Advanced-model misuse, security failures and unexpected autonomous behaviour are named risk factors. | Least privilege, approval for consequential actions, logs and adversarial testing. |
| High | Privacy and confidential data | Sensitive business data, breaches and compliance obligations can affect trust and access. | Map data flows, retention, permissions and contractual controls for each deployment. |
| High | Release or access restrictions | Safety testing can delay releases or restrict capability and availability. | Do not plan around an unreleased model or assume unrestricted access to the most capable one. |
| Medium to high | Reputational exposure | Safety commitments and the public benefit purpose invite scrutiny of implementation. | Keep documented safeguards separate from claims that outcomes are guaranteed. |
Two points deserve care. The government dispute is not one resolved case: the filing describes a district court ruling largely in the company’s favour in August 2026 and a split appeals decision in September denying review of a separate designation, with rehearing sought, and it states that June export restrictions affected access to the Fable 5 and Mythos 5 models worldwide until lifted. And the $1.5bn Bartz copyright settlement, approved in July 2026, is a settlement of one case; it does not establish that all training, output or customer use is free of intellectual property risk.
What this means if your business uses Claude
Nothing in the filing is a reason to buy or sell shares, and nothing in it tells a UK firm whether to adopt a particular product. It does change how a buyer should contract and plan.
- Budget usage separately from implementation. About 80% of Anthropic’s revenue is consumption. Your bill moves with task design, prompt length and volume. Quote implementation as a fixed scope, agree a usage budget with alert limits, and define what triggers a scope change.
- Keep critical workflows portable. Cloud distribution through AWS, Google Cloud and Microsoft Azure gives procurement choices, but model access can still be affected by export rules, supplier capacity and policy changes. A fallback configuration only counts if it has been tested against your real quality and privacy requirements.
- Read the headline numbers carefully. A positive adjusted result is not GAAP net profit; a fair-value accounting loss is not cash burn; the approximately 900% retention figure describes a cohort’s spend growth, not customer loyalty; and the compute commitments do not sum.
- Expect the vendor to reach the work consultancies sell.The filing describes Anthropic’s own applications, an Applied AI function, a partner network, a $100 million Claude Frontier Academy aiming to train 10,000 engineers by the end of 2027, and Ode with Anthropic, a services company with private equity partners for mid-sized companies. Sector knowledge, implementation accountability and ongoing service are what an independent adviser has to offer; general product familiarisation is not.
- Check model eligibility at deployment and at renewal. Geography, use case and the specific model all matter, and the filing shows that access to the most capable models can be gated or restricted at short notice.
Related reading: the Claude Enterprise buyer’s guide, what Claude costs in the UK, Claude data residency for UK firms and our briefing on Anthropic’s and OpenAI’s consulting ventures.
What we are watching
The questions the preliminary filing leaves open, in the order they matter to a buyer. Each is closed in the update log when a filing answers it.
| Priority | Question or event | Why it matters |
|---|---|---|
| P1 | An S-1/A with the price range, share count, dilution and selling holders | Enables an accurate description of the offer economics and the public float. |
| P1 | Completed September financial, liquidity and commitment tables | Tests whether the June liquidity and the Q2 operating improvement persisted. |
| P1 | The Run-Rate Revenue value, and correction or confirmation of its 31 October date | Prevents a blank, future-dated placeholder being repeated as a reported fact. |
| P1 | An annual cash schedule for the compute and infrastructure commitments | Allows a meaningful view of cash needs, cancellation rights and financing overlap. |
| P1 | Revolver borrowing, minimum liquidity and partner-funding conditions | Shows actual usage and whether conditional capacity remains available. |
| P1 | Model export and access restrictions, and government litigation developments | Can change deployability for specific sectors, countries or public bodies. |
| P2 | Paid and free users, active-user definitions and enterprise cohort detail | Improves adoption analysis without mixing accounts, people, seats and spending cohorts. |
| P2 | Gross-profit or inference-cost disclosure with consistent definitions | Would permit a unit-economics comparison with other software companies. |
| P2 | Trust appointments, charter changes and public benefit reporting | Tests the operation of the mission and control protections. |
| P2 | Pricing, partner terms, Academy availability and the Ode expansion | Affects training, procurement and implementation options for UK firms. |
Sources, method and reuse
Every figure on this page is read from the filing’s own text and tables, in US dollars and for the periods the filing states, and rendered from a single data file with the printed page each table came from. No figure is typed into the page by hand. We do not convert currency, annualise, infer a gross margin, a cash runway or a valuation, or sum the compute commitments, because the filing does not support those calculations. Annual figures for 2025 are audited by EY and earlier years by Grant Thornton; interim and quarterly figures are unaudited.
The filing is company-authored evidence. Its customer stories, benchmarks, safety claims and market descriptions were not independently audited by The AI Consultancy. Litigation and policy passages are attributed to the S-1; this page is not a current-law review of each proceeding, and it is not legal due diligence.
Calculations
Each calculation uses the filing's reported values in US dollars and its stated periods. No currency conversion, no annualisation, no substitution of periods.
| Calculation | Expression | Result | Source |
|---|---|---|---|
| 2025 revenue growth | (4,585 / 386 - 1) x 100 | 1,087.8 % | S-1 p. 147 |
| H1 2026 revenue growth against H1 2025 | (16,286 / 1,163 - 1) x 100 | 1,300.3 % | S-1 p. 147 |
| H1 2026 GAAP operating margin | -693 / 16,286 x 100 | -4.3 % | S-1 p. 147 |
| H1 2026 adjusted operating margin | 1,618 / 16,286 x 100 | 9.9 % | S-1 p. 156 |
| Q2 2026 GAAP operating margin | 974 / 11,555 x 100 | 8.4 % | S-1 p. 157 |
| Q2 2026 adjusted operating margin | 2,235 / 11,555 x 100 | 19.3 % | S-1 p. 160 |
| Cash plus short-term investments, 30 June 2026 | 112,389 + 19,262 | 131,651 US$ million | S-1 F-5 |
| H1 2026 adjusted operating reconciliation | -693 + 2,185 + 126 | 1,618 US$ million | S-1 p. 156 |
| Revenue from the top five customers, Q3 2026 | 100 - 79 | 21 % (approximate) | S-1 pp. 134 to 143 |
| Revenue per dollar of compute, Q1 2025 | 376 / 944 | 0.40 x | S-1 p. 157 |
| Revenue per dollar of compute, Q2 2026 | 11,555 / 5,558 | 2.08 x | S-1 p. 157 |
Primary sources
- SEC filing index for accession 0001628280-26-067168
- Form S-1, Anthropic, PBC, filed 26 October 2026
- EX-FILING FEES, filing fee table
- SEC archive directory listing for the filing
Reuse: The AI Consultancy’s compilation, charts and calculations on this page are licensed under CC BY 4.0. Credit The AI Consultancy and link to this page. The filing itself is a public record published by the US Securities and Exchange Commission.
Update log
The page changes when Anthropic files an amendment or a related document, when the offering prices or trades, and when we correct something. Earlier values are superseded here, not silently overwritten.
: Page published
First version, built from Anthropic, PBC's preliminary Form S-1 filed on 26 October 2026 (accession 0001628280-26-067168). The SEC's company submissions feed was checked on 28 October 2026 and listed no amendment. Every figure was read from the filing's own text; the calculations table shows the arithmetic for every derived number.
Frequently asked questions
Has Anthropic gone public?
Not yet. Anthropic, PBC filed a preliminary Form S-1 on 26 October 2026, which starts the registration process. It has applied to list on the Nasdaq Global Select Market under the symbol ANT, but the offer price, the number of shares and the date of the first trading day are not in the filing. As of our check on 28 October 2026, no amendment had been filed.
When is the Anthropic IPO date?
The filing does not give one. A preliminary S-1 is followed by at least one amendment with a price range and share count, then an effective date, a roadshow and pricing. None of those steps had happened at 28 October 2026. This page's update log records each one when it does.
What is Anthropic's valuation?
The filing does not state one, because the offer price is blank. The only dollar figure attached to the offering is the $1.0bn maximum aggregate offering price in the filing-fee exhibit, which exists solely to calculate the 87,000 dollar registration fee under Rule 457(o). It is not the intended raise, the offer size or a valuation.
Is Anthropic profitable?
It depends which line you read, and the filing reports three. For the six months to 30 June 2026 the GAAP operating result was a loss of $693m, although the second quarter alone produced $974m of operating income. The company's adjusted operating income, a non-GAAP measure that adds back $2.2bn of stock-based compensation and $126m of non-cash charitable contributions, was $1.6bn. The net loss was $80.7bn, of which $72.4bn was the change in fair value of convertible notes, derivatives and SAFEs. Operating cash flow for the half was $3.9bn.
What is Anthropic's revenue?
$16.3bn for the six months to 30 June 2026 (unaudited), $4.6bn for the year to 31 December 2025 and $386m for 2024. Growth was 1,087.8% in 2025 and 1,300.3% for the first half of 2026 against the first half of 2025. About 80% of first-half revenue was consumption-based and about 20% came from subscriptions.
Why is Anthropic's net loss so much larger than its operating loss?
Because $72.4bn of the $80.7bn first-half net loss is the change in fair value of convertible notes, derivative liabilities and SAFEs. When the company's valuation rises, the instruments it sold to investors are remeasured upwards and the increase runs through the income statement as a loss. It is not cash leaving the business. It does reflect value transferred to financing counterparties and dilution still to come, so it is not nothing either.
What does 900% net revenue retention mean?
The filing's approximately 900% figure at 30 September 2026 compares revenue in the latest three months with the corresponding three months a year earlier, for an eligible prior-year cohort of customers that generated at least $1. Fixed-price Claude Pro and Max subscriptions are excluded. It says that eligible existing customers spent about nine times as much as a year earlier. It is not a churn rate, not a share of customers retained and not a trailing-year ratio, and it will fall as the comparison base grows.
What is the Long-Term Benefit Trust?
A financially disinterested trust that holds Anthropic's Class T shares and generally has the power to elect and remove a majority of the board. Alongside it, a single Class F share held by the founders' LLC carries 50.1% of general voting power. Public Class A shareholders therefore do not control the board. The trust's director selections are subject to a committee process with an employee-vote fallback, so "the trust appoints every director without constraint" would be inaccurate. Anthropic expects to be a Nasdaq controlled company.
How much compute has Anthropic committed to buy?
The filing lists 8 arrangements, the largest of which is Broadcom equipment leases at approximately $161.2bn. We do not add them into one figure, and we recommend against quoting one: the arrangements are on different bases (minimum purchase, up to, approximately, estimated), some are cancellable after an initial period, some depend on delivery or financing, and the filing leaves its own aggregate September commitment fields blank. The table on this page gives each amount with its basis and its conditions.
What does the Anthropic IPO mean for UK businesses using Claude?
The filing confirms that consumption pricing is the core of Anthropic's business, that it is building its own applications and consulting ventures, and that model access can be affected by export rules and government action. For a UK buyer that means budgeting model usage separately from implementation, testing a fallback configuration against real quality and privacy requirements, and checking model eligibility at renewal. It does not establish that current pricing, model access or the growth rate will continue, and nothing in the filing is a reason to buy or sell shares.
This page changes when Anthropic files again. Subscribe to get the update when it does.
Get new briefings by email
The AI Consultancy newsletter delivers briefings and analysis for UK businesses. We use your address only to send it, and you can unsubscribe at any time.