Anthropic IPO S-1 Explained: Claude’s Revenue, Profits and Biggest Risks

Claude’s business is growing fast enough for Anthropic to have earned an operating profit in the second quarter of 2026. Yet the same company reported an $80.748 billion net loss for the first half of the year.

Both statements are true. The explanation lies in the difference between running an AI business and accounting for the financing that helped build it.

On October 26, Anthropic publicly filed its S-1 registration statement for an IPO. The company behind Claude intends to list Class A shares on the Nasdaq Global Select Market under ANT. Its final offer price, size and trading date remain unconfirmed in this filing.

The prospectus makes a serious business visible beneath the AI excitement. It also exposes the commitments, ownership rules and unanswered pricing questions that will determine how much of that business’s success can reach public shareholders.

Anthropic IPO 2026: Date, Price, Ticker and Filing Status

An S-1 is the company’s detailed introduction to prospective public investors. It describes what the business does, its financial history, its risks and the rights attached to its shares.

Anthropic’s confidential draft submission on June 1 was an earlier step in the review process. The October public filing is the event that opens the prospectus to everyone. The offering contemplates both newly issued shares and shares from existing owners. Only the company’s portion would raise capital for Anthropic. Proposed uses include computing infrastructure, operations, hiring and certain employee-equity tax obligations.

How Anthropic Makes Money From Claude AI

Anthropic sells access to Claude’s capabilities: writing and analysing information, developing software and carrying out tasks across business systems. The model is the underlying technology. Claude Platform gives developers tools to build with it, while applications give customers a ready-made way to use it.

These are connected layers of one business. Anthropic reports one operating and reportable segment; the filing does not provide separate revenue or profit accounts for models, Platform, Claude Code and Claude Cowork.

Most revenue follows usage. In H1 2026, consumption-based revenue was $13.064 billion, while subscription revenue was $3.222 billion. That is approximately 80% and 20% of the $16.286 billion total, respectively.

Consumption pricing is metered in tokens: small units into which the system converts information it processes. A token can represent part of a word or other data; it is not a fixed number of words. The model chosen and the amount processed influence the bill.

The familiar comparison is a metered utility. More work passing through the service generally produces a larger bill. But Claude has different models at different prices, and an extended task may require several rounds of processing. Counting users alone therefore tells investors much less than understanding how deeply customers use the service.

Subscriptions charge for a specified level of access and usage, generally over terms of a year or less. They are not unlimited promises of computing power. Customers can move to a higher tier or pay for additional consumption, so subscriptions and usage revenue can coexist within a customer relationship.

Claude Code helps people develop software. Cowork extends the approach to broader knowledge work. Their strategic importance is that they make Claude useful inside everyday workflows, giving an organisation more reasons to return and spend. Their individual revenue and margins are not disclosed.

The customer-case images illustrate the commercial logic. AIG’s case describes Claude supporting insurance underwriting and reports a reduction of more than 90% in submission-review time where its system was deployed. Shopify describes using Claude among the models behind its Sidekick merchant assistant, alongside internal use of Claude Code.

Customers can access Claude directly or through cloud marketplaces. About 48% of H1 2026 revenue came through those marketplaces. 

Anthropic says it contracts directly with customers, sets pricing and remains responsible for providing the service. It recognises marketplace sales on a gross basis and pays channel fees. That accounting detail matters when judging its margins.

Anthropic Revenue Growth: How Fast Is Claude Growing?

The revenue history shows how quickly the business has changed scale.

Source: S-1 statements of operations and quarterly results. The rows cover different lengths of time; H1 includes the two quarters above it.

For a like-for-like comparison, H1 revenue rose from $1.163 billion in 2025 to $16.286 billion in 2026, an increase of a whopping 1,300%. Q2 alone generated substantially more revenue than the whole of 2025.

The important question is what customers are paying for. An occasional chatbot subscription is one relationship. Embedding Claude in coding, underwriting or a merchant assistant creates repeated demand tied to actual business activity. The cases suggest that broader use is possible; the revenue data shows that spending has already expanded considerably.

There is still a distinction between evidence of adoption and evidence of permanence. Consumption can fall when customers reduce activity, improve their own software or switch models. Useful technology does not automatically give its supplier permanent pricing power.

What does approximately 900% net revenue retention mean?

Anthropic’s approximately 900% NRR at September 30, 2026 compares revenue from the same qualified customer cohort across two three-month periods, one year apart.

The base cohort consists of customers with at least $1 of revenue in the earlier quarter. Subscribers to the fixed-price Claude Pro and Max plans are excluded because those subscriptions do not capture the expansion and contraction the measure is intended to show. Revenue from that cohort in the current quarter is divided by its revenue in the year-earlier quarter.

For illustration, a cohort generating $1 million in the earlier quarter and $9 million in the later quarter would produce 900% NRR. That represents nine times the revenue, or an 800% increase, after expansion, contraction and churn within that cohort. It is not a customer-retention percentage, and newly acquired customers outside the base cohort do not explain the result.

This is strong evidence of spending expansion from a small starting base. It is not a promise that the same customers will multiply their spending ninefold every year.

The organisation-level customer count reinforces the adoption story. As of September 30, more than 11,900 customers generated at least $100,000 each over the trailing 12 months, up from more than 950 a year earlier. The higher-spending groups, more than 1,900 at $1 million or more and approximately 250 at $10 million or more, sit inside that broader group. Adding the bands would double-count customers.

Concentration is easing, but remains relevant. For the nine months ended September, the top five customers generated approximately 25% of revenue, and no individual customer exceeded 10%. That is a different period from H1, when one customer contributed 11%.

At June 30, remaining performance obligations, or RPO, were $10.8 billion, with approximately 75% expected to become revenue over the following 12 months. RPO covers revenue still to be earned under non-cancelable commitments, including deferred revenue. It provides some visibility, but does not capture all future consumption or guarantee future customer expansion.

One qualification deserves attention: Anthropic estimates that customers subsequently judged to violate its usage policy or commercial terms generated 5-7% of nine-month revenue. Stronger enforcement could affect growth, but the filing does not establish a certain future loss of that percentage. This is the company’s estimate, and detection and enforcement timing matter.

Finally, revenue earned, a run rate and a forecast answer different questions. Anthropic’s defined run rate combines trailing 28-day consumption fees multiplied by 13 with monthly subscription fees multiplied by 12. It expresses a recent pace as an annual figure. It is neither a completed year’s revenue nor a forecast of the next year.

Is Anthropic Profitable? Reconciling the S-1’s Contradictory Numbers

Anthropic earned an operating profit in Q2, recorded an operating loss across H1 and reported net losses in both periods. The apparent contradiction disappears once the periods and accounting measures are separated.

Where the $80.748 billion H1 loss came from

Operating profit measures revenue after operating expenses. Net profit also includes financing-related gains and losses, investment income, other non-operating items and tax.

Here is the complete bridge for January-June 2026:

Source: S-1, MD&A p. 147 and consolidated statements, p. F-7. Amounts converted from millions into billions.

The largest charge came from revaluing financing instruments, primarily convertible notes. These are funding arrangements that can turn into shares under agreed conditions. When the value of what the investor is entitled to increases, the liability’s accounting value can rise too. Anthropic records that increase as an expense under its chosen accounting treatment.

Consequently, a higher estimated company valuation can produce a larger reported loss. The $72.388 billion charge was a non-cash remeasurement, rather than a bill for that amount of chips or electricity during H1.

But removing it does not finish the analysis. The resulting $8.360 billion loss is neither operating loss nor cash burn. It still includes the $8.757 billion net other expense, interest income and tax.

That other-expense line was itself largely driven by recognition and valuation of financial instruments, including note-purchase arrangements. The company’s adjusted-net-income reconciliation excludes another $8.720 billion of specified mark-to-market items. Calling everything left after one adjustment “the real operating loss” would therefore be wrong.

These accounting charges also have economic meaning. Financing rights can ultimately create additional shares or cash obligations. A charge can be non-cash today while the underlying agreement still affects what existing shareholders own.

Q2 shows the distinction particularly clearly. On $11.555 billion of revenue, Anthropic earned $974 million in operating income, yet recorded a $66.309 billion net loss, largely because of the financing-related items below operating profit. The positive operating result was reported under standard accounting, rather than existing only after management adjustments.

What the 85.4% gross margin does and does not show

The main income statement groups expenses by their nature, such as computing and personnel. Note 8 also presents them by function, including cost of revenue and research and development.

That table gives H1 cost of revenue of $2.380 billion. Our calculation is:

Gross margin = ($16.286B − $2.380B) ÷ $16.286B = 85.4%.

In ordinary language, roughly $85 of each $100 of revenue remained after the costs classified as delivering the service. It still had to fund research, distribution and administration.

Those remaining costs were substantial. Functional R&D was $9.741 billion and sales and marketing was $3.808 billion in H1. The latter includes channel-partner costs. General and administrative expenses were another $1.050 billion. Together, they explain why the high calculated gross margin did not produce a half-year operating profit.

It would be misleading to subtract the entire $8.945 billion computing-and-infrastructure expense from revenue and label the result gross profit. That computing pool supports model development, research, customer service and internal work. It spans several functions.

The 85.4% figure is a consolidated calculation using the filing’s classifications. It does not establish Claude Code’s margin, the profitability of an individual token or how much cash shareholders can eventually receive.

Adjusted earnings and cash flow provide different checks

For H1, Anthropic reported adjusted operating income of $1.618 billion, adjusted net income of $2.671 billion and positive operating cash flow of $3.929 billion. These are three different measures.

Adjusted operating income excludes stock-based compensation, non-cash charitable contributions and the specified settlement charge where applicable. Adjusted net income makes further financing-related adjustments. Equity compensation remains a cost to shareholders even when excluded from an adjusted measure.

Operating cash flow follows cash receipts and operating payments, including changes in receivables, customer advances and supplier-payment timing. It does not include every capital investment or financing payment. Its improvement is encouraging, but does not establish that Anthropic can fund its entire future build-out internally.

There is another reason to avoid declaring a permanent profitability turning point: new computing capacity can arrive in large steps, bringing expense before the associated customer revenue fully develops. The IPO will also trigger recognition of compensation for certain restricted stock units: employee share awards whose conditions include a qualifying public-market event.

Why Anthropic’s Compute Contracts Matter More Than a Headline Total: Amazon, Google and SpaceXAI Deals

Compute is the combination of specialised chips, memory, networks and supporting infrastructure used to develop models and answer customers’ requests. Training develops the model. Inference runs it when a customer uses Claude.

The commercial challenge resembles reserving factory capacity before knowing the exact volume of orders. Too little capacity can mean missed demand. Too much can leave the company paying for equipment or services it cannot use profitably.

Anthropic says it typically procures long-term compute 12-18 months before availability. Current margins therefore partly reflect past procurement decisions. They cannot tell us what the next wave of capacity will cost relative to the revenue it supports.

Three disclosed arrangements show why the contractual wording matters:

Source: S-1, MD&A pp. 170–171. These are selected examples.

Delivery timing matters too. Some fees begin when capacity is delivered and accepted. Under certain Broadcom equipment leases, rent can begin earlier in specified circumstances, including a facility-readiness delay. Capacity that arrives late can constrain revenue; payments that start before productive use can strain cash.

Anthropic is also changing how it obtains capacity. Under Anthropic Direct, it leases chips and uses dedicated facilities operated on its behalf, alongside conventional cloud arrangements. Separating long-lead-time buildings from later equipment commitments may improve flexibility and economics. It also adds lease, funding and execution responsibilities that a conventional software subscription business would not carry to the same extent.

The partner relationships are unusually intertwined. Amazon, Alphabet’s Google and Microsoft provide combinations of investment capital, infrastructure, distribution and customer demand while also competing in AI. Their support helps Anthropic expand, but changes in their priorities can affect several parts of its business simultaneously.

Broadcom’s financing mechanism shows how infrastructure can reach the share count. A conditional facility permits up to $42 billion of convertible notes to finance specified rent. It was undrawn at June 30. Normal terms include 9% interest paid by adding it to principal and potential conversion at 90% of a recent average share price. For notes issued when specified revenue conditions are not met, those become 13% and 85%. Deferring a cash payment can therefore create a larger claim and future dilution; the facility is not cash already raised.

Anthropic nevertheless had substantial historical resources: $112.389 billion in cash and equivalents plus $19.262 billion in short-term investments at June 30, totalling $131.651 billion. That is a dated combined balance, not current cash alone. H1 financing inflows were $110.774 billion, showing how much the balance depended on capital raised.

The company says its resources can support at least the next 12 months, subject to its stated assumptions. It also expects its infrastructure build-out to require significant cash. Both can be true. The unresolved question is the timing of future payments, customer collections and additional financing, not simply whether June’s balance looked large.

What ANT Shareholders Would Own and What They Would Not Control

Public investors would own Class A equity in a Delaware public-benefit corporation. The board must balance shareholders’ financial interests with the interests of those materially affected by the company and its public-benefit purpose.

This has practical consequences. Anthropic may restrict a model release, direct resources toward public-benefit work or accept lower financial results when it believes its wider duties justify that choice. Its founders’ letter explicitly prepares investors for decisions that depart from pure profit maximisation.

Economic ownership, ordinary shareholder votes and board-election rights are separate questions:

The Long-Term Benefit Trust generally elects four of seven directors. One of those seats has additional nomination and removal safeguards, including an employee vote when the relevant board committee and trust disagree. The founders’ voting entity can determine elections for the other three seats. The trust’s appointment power does not make it the day-to-day management team.

Founder LLC’s 50.1% voting power is not 50.1% economic ownership. Its voting share adjusts to preserve that general voting majority after future issuances. Public shareholders can therefore suffer economic dilution without acquiring greater influence over general shareholder decisions.

The structure has sunset provisions, rather than a simple promise that control ends after a fixed number of years. For example, fewer than three qualifying Continuing Founders for 90 consecutive days starts a 450-day transition toward the Class F sunset. The trust has its own separate termination rules.

The Public Benefit Fund also affects the ownership analysis. Its initial allocation is based on 5% of fully diluted shares as of June 30, assuming full conversion of Amazon’s notes. It is not 5% of the IPO’s freely traded shares.

The board may make later grants of up to an additional 2% at sustained 50% share-price milestones, at its discretion. The initial fund shares are legally outstanding but excluded from consolidated accounting outstanding shares while retained by the wholly owned subsidiary. Transfers outside the group can change the relevant accounting share count. An immediate, automatic dilution calculation would miss that qualification.

Separately, the founders pledged 80% of their personal Anthropic equity to philanthropy. They did not pledge 80% of the entire company.

Other potential sources of new shares include fundraising, employee awards and financing conversions. A secondary sale of an existing stake is different: it transfers ownership and sends proceeds to the seller, without itself creating new company shares. Final offering terms are needed to quantify either effect.

The Risks That Could Change Anthropic’s Business

The filing contains many risks. The most important ones have a direct route into revenue, margins, financing needs or shareholder ownership.

1. Competition can shorten the life of the advantage. A cheaper model that is good enough can win routine workloads even if Claude remains stronger at difficult tasks. Larger competitors can bundle AI into existing software or subsidise pricing. Anthropic may then need to cut prices while continuing to fund research and compete for scarce technical talent. Models released in the six months ended September accounted for more than 80% of Q3 revenue, according to the company.

2. Customer and channel dependence can amplify a slowdown. Large accounts can optimise usage, negotiate terms or shift providers. Some younger customers depend on outside funding, which can constrain their spending when capital becomes scarce. Cloud marketplaces broaden distribution but concentrate access and collections in powerful partners. The risk is a revenue slowdown while infrastructure obligations remain.

3. Chips, electricity and facilities must arrive together. An equipment order is not the same as usable capacity. Power connections, cooling, networking, construction and financing all have to work. Delays can force Anthropic to obtain more expensive short-term compute or restrict customer access. The filing describes a March 2026 outage and degraded performance, with demand growth and compute constraints among the contributing factors. A service interruption is an existing event; any future damage to retention is a risk rather than a reported outcome.

4. Copyright and data disputes can change the cost of the product. Anthropic recorded a $1.5 billion Bartz settlement charge in FY2025. It also discloses continuing publisher, author and data-related claims. Allegations in those proceedings are not findings of liability. Adverse outcomes could require damages, licensing payments, changes to training data or model retraining. That could consume cash and weaken model economics even if customer demand stays strong.

5. Safety restrictions can postpone revenue, while safety failures can threaten it. The company has already limited model access and delayed development steps for safety or security reasons. Its founders also describe choosing to forgo revenue from a broad release of Mythos until safeguards were improved. Separately, the S-1 says June export restrictions on Fable 5 and Mythos 5 were subsequently lifted. Those episodes demonstrate a commercial constraint; they do not establish that the models remain globally prohibited. A serious future misuse or security incident could lead to claims, tighter controls and reduced customer trust.

6. Government disputes can reach beyond government sales. Direct government-agency contracts represented approximately 1% of revenue in the periods presented, but procurement restrictions can affect commercial contractors too. The D.C. Circuit’s September 25 decision upheld the Department of War’s exclusion under the supply-chain-security statute it reviewed. The S-1 says Anthropic requested rehearing. That decision concerns a particular legal route, rather than resolving every related lawsuit. The business risk includes customers delaying adoption because they also serve government agencies.

7. First-party applications can compete with platform customers. A software company may use Claude to build its product while Anthropic develops an application addressing the same task. Successful applications can deepen Anthropic’s customer relationships, but may also encourage developers to use rival models. The company must capture more value without weakening the ecosystem that brings it demand.

What Would Make Anthropic’s IPO Valuation Reasonable?

The public filing establishes a business to analyse. It does not yet establish the price investors will be asked to pay.

Take the $2 trillion target valuation purely as an illustration. Annualising the disclosed Q2 revenue gives:

$11.555 billion × 4 = $46.220 billion.

$2,000 billion ÷ $46.220 billion = approximately 43.3 times Q2 annualised revenue.

This uses a completed quarter multiplied by four. It is not a forward-sales multiple, Anthropic’s separately defined run-rate metric or an enterprise-value multiple adjusted for cash and debt. Substituting a much larger forecast underneath the same valuation would make the multiple look lower while adding the risk of achieving that forecast.

A price at that illustrative level would demand more than rapid adoption. Anthropic would need to sustain material growth, retain customers as alternatives improve and turn that activity into durable cash after research, distribution and infrastructure costs. It would also need to do so after accounting for future equity issuance and its public-benefit commitments.

The evidence to watch is specific:

• Final offer terms: price, issuer proceeds and the split between new and existing shares.
• Complete Q3 financials: revenue, standard and adjusted earnings, and their reconciliations.
• A current commitment ladder: minimum payments, delivery dates, cancellation terms and variable obligations.
• Cash generation: customer collections and cash remaining after infrastructure investment and financing payments.
• Fully diluted shares and tradable supply: conversions, employee awards, fund transfers and early unlocks.

If revenue keeps scaling faster than the fully loaded cost base and cash generation withstands new capacity deliveries, the case for a premium strengthens. If growth slows while fixed payments and new share issuance accelerate, the same headline valuation becomes harder to support.

Can Indian investors access Anthropic shares yet?

The proposed ANT listing does not yet create a publicly traded Anthropic share. Once Anthropic lists on a US stock exchange, its shares will be available to Indian investors through INDmoney US Stocks, just like other US-listed stocks. 

Author’s Take

Anthropic’s strongest evidence is already in the business: recognised revenue, rapidly expanding customer spending and a positive Q2 operating result. Those facts deserve more weight than either a spectacular loss headline or a distant revenue ambition.

The unresolved issue is how durable that operating improvement will be as new capacity arrives. Anthropic must repeatedly develop models customers prefer, earn enough from them to support long-lived obligations and leave value for each share after financing and other issuances.

My assessment is that the filing makes a credible case for an unusually strong growth business. It does not yet make a complete investment case at an unknown IPO price. At the illustrative $2 trillion level, the disclosed economics require sustained expansion and cash conversion, with limited room for expensive execution mistakes. The final price, complete Q3 accounts and current payment-and-share schedules could materially change that assessment.