The largest IPO in history updates its prospectus, with Q3 revenue increasing by 1127% year-on-year, and a net income retention rate of approximately 900%

Author: iponews Studio, Unicorn Early Insights

On November 2, Anthropic submitted an updated S-1/A filing to the U.S. SEC. This document provides a complete quarterly snapshot up to September 30, 2026, for the upcoming Nasdaq IPO, bringing the core financial tensions of this AI giant to the forefront.

Within the same financial report, two opposing stories are told. According to non-GAAP measures, the adjusted operating profit for the third quarter was $2.5 billion, with a profit margin of 15%, indicating that the operational level has crossed the breakeven line. However, under GAAP measures, the operating loss for the quarter was $567 million, and the net loss was $23.3 billion, nearly doubling from $12.1 billion in the same period last year.

At the same time, a note in the filing hides a number far exceeding the visible liabilities on the balance sheet: as of September 30 and beyond, the total commitment expenditures under purchase commitments and lease agreements amount to approximately $895.4 billion, almost entirely directed towards computing power and infrastructure.

The question that the IPO pricing needs to answer is not "Is Anthropic profitable?" but rather which set of accounts the market is willing to pay for: the already positive adjusted operating profit or the computing power bill that will continue to be realized in the coming decade, measured in hundreds of billions.

The $23.3 billion quarterly net loss, in absolute terms, is enough to plunge any traditional company into crisis. However, for Anthropic, the way this loss is composed means it cannot be equated with a deterioration in operations.

The bulk of the loss comes from the fair value changes of convertible notes and derivative liabilities. The company's valuation in the private market continues to rise, causing the book value of early-issued convertible bonds and preferred stocks to inflate, which must be accounted for in the current profit and loss according to accounting standards, resulting in massive non-cash expenses. The net losses of $6.63 billion in the second quarter and $23.3 billion in the third quarter are largely "accounting reflections of valuation increases," rather than actual cash outflows for the quarter.

What truly deserves attention is the operational efficiency turning point corresponding to the adjusted operating profit of $2.5 billion. The third-quarter revenue was $16.96 billion, a year-on-year increase of 1127%, continuing to grow approximately 47% from the previous quarter's $11.56 billion. The gross profit margin has risen to a high level due to scale effects, and the computing power cost ratio has systematically declined from extremely high levels a year ago.

In the second quarter, Anthropic achieved a GAAP operating profit of $974 million, and the GAAP operating loss of $567 million in the third quarter is more a reflection of the rhythm of expenses and accounting recognition fluctuations between quarters, rather than a reversal of operational trends.

In other words, Anthropic's operating model has reached a scale range capable of generating positive operating profits. The issue is not whether it can make money, but whether the money earned is sufficient relative to the already locked-in expenditure commitments.

The $895.4 billion in committed expenditures is the part of this S-1/A that is most easily overlooked by ordinary investors but is most worth reading word by word.

These commitments include leases that have not yet begun, signed but undelivered computing power purchase agreements, and minimum payment obligations for cloud services over the coming years. Among them, the amount for leases that have not yet begun is approximately $451.2 billion; the leasing commitments from Broadcom alone reach $161.2 billion; and the amount of newly signed computing agreements after September 30 is about $100.2 billion. By time distribution, the amount of commitments due within the next 12 months is approximately $69.8 billion, while the commitments expected to incur by 2031 are about $447.5 billion.

David Trainer, CEO of investment research firm New Constructs, characterizes this as a massive scale of "hidden debt," stating, "They must invest heavily in training, inference, and energy."

When these figures are placed alongside operating cash flow, the tension becomes clearer. In the first half of the year, Anthropic's cash flow from operating activities was positive at $3.929 billion, with cash and short-term investments totaling approximately $131.6 billion, indicating that the cash reserves are not weak. However, relative to the long-term commitments measured in hundreds of billions, operating cash flow and existing reserves can only cover a portion of obligations for the next few years. One of the core uses of the IPO proceeds is to provide an equity-level financial buffer for this computing power bill.

From a business logic perspective, the $895.4 billion in commitments is not irrational expansion. In the context of ongoing tight supply of AI computing power, locking in chip, electricity, and data center capacity for the next few years in advance is a form of "insurance" for the certainty of model training and inference purchases.

Amazon, Google, and Microsoft are not only shareholders and cloud service providers for Anthropic but also direct competitors in the large model market. Locking in computing power supply through long-term contracts has strategic defensive significance. However, the cost of this strategy is that the company's operational freedom is deeply coupled with the strategic priorities of a few infrastructure suppliers, and the rigidity of committed expenditures will amplify profit fluctuations when demand growth slows.

If we only look at the $895.4 billion in committed expenditures, Anthropic's IPO appears to be a gamble backed by future cash flows. But supporting this gamble is another equally extreme set of data.

As of September 30, 2026, Anthropic's net revenue retention rate (NRR) is approximately 900%. This means that the revenue contributed by the same batch of enterprise customers after one year is about nine times the initial amount. The number of customers contributing over $100,000 annually has increased from 950 a year ago to over 11,900; the number of customers contributing over $1 million annually exceeds 1,900, and those contributing over $10 million are about 250. More than 50% of high-value customers simultaneously use the Claude Platform and at least one of Anthropic's proprietary applications, while this proportion was about 30% at the end of 2025.

A 900% NRR is unprecedented in the SaaS industry, revealing a mechanism: when enterprises embed Claude into core workflows such as coding, research, finance, and law, token consumption will exponentially increase with the depth of use.

The top 5% of high-spending enterprise users using Claude Code in the third quarter have monthly spending approximately 20 times that of users not using Code. This "pay-as-you-go model, automatically scaling with deeper penetration" gives Anthropic's revenue curve a flexibility that traditional software subscription models do not possess.

This also explains why the $23.3 billion GAAP net loss did not deter underwriters and cornerstone investors. What the market is truly pricing is whether the 900% NRR can continuously convert enterprise customers' AI budgets into Anthropic's revenue within the payment cycle of the $895.4 billion in commitments. The underwriting syndicate led by Goldman Sachs and Morgan Stanley, and the market discussions around a valuation range of $1.5 trillion to $2 trillion, are all built on this logic.

The S-1/A presented by Anthropic shows a company that has successfully established an operational profitability model but has yet to achieve cash flow consistency.

The $2.5 billion adjusted operating profit in the third quarter proves that scale effects are real; the $23.3 billion GAAP net loss reminds the market that the numbers on the books under accounting standards are profoundly influenced by capital structure; the $895.4 billion in committed expenditures lays the long-term bet of this IPO on the table: in the foreseeable future, every dollar of operating profit earned by Anthropic will need to race against a computing power bill measured in hundreds of billions.

The Nasdaq pricing will not provide a simple "yes" or "no," but will seek a market-accepted risk premium between adjusted profits and net losses, between 900% NRR and $895.4 billion in commitments.

For investors focused on the commercialization of AI, the most important information in this document may not be any single quarterly figure, but rather a structural judgment: the competition among frontier AI companies has shifted from "who can create better models" to "who can use sustainable cash flow to lock in enough computing power and sufficiently deep enterprise workflows for the next generation of models." Anthropic has demonstrated its capability in the first aspect with a $2.5 billion adjusted profit, and the IPO will test its performance in the second aspect.