Read Anthropic's S-1 like an acquirer would

On 11 November 2026, the Financial Times reported that underwriters for Anthropic are preparing to raise the company's IPOIPOThe first sale of a private company's shares to public investors on a stock exchange, converting private ownership into publicly traded stock.View full definition → valuation target to $1.5 trillion or higher. A week earlier the mood ran the other way. After a post-election AI selloff, potential buyers were discussing about a $1.2 trillion to $1.3 trillion valuation range for the company, and some were weighing a value as low as $1.1 trillion. In between, Anthropic set a price range of $70 to $78 per share in an updated filing with the Securities and Exchange Commission, and CNBC reported that the order book is already covered at the proposed range.

Within a few days the price talk moved by hundreds of billions of dollars. The facts that decide whether that money is earned hardly moved, and they sit where acquiring CFOs tend to stop reading: in how the headline metric is defined, in how fixed obligations compare with variable revenue, and in who actually controls the asset. Anthropic's filing is the most useful due diligence teaching document of 2026 because it states all three in plain language. Most private targets you will buy state none of them unless you ask.

How Anthropic's S-1 puts three red flags in plain sight

Anthropic did not hide its risks. CNN counted 90 pages of warnings and caveats in its more-than-300-page filing. What matters for a CFO is the structure of what was disclosed, because those same three structures turn up in mid-market targets with far less candour.

A run-rate built on 28 days times 13

Anthropic's non-GAAPGAAPThe standard set of accounting rules companies follow to prepare consistent, comparable financial statements, dominant in US reporting.View full definition → metric annualises the previous 28 days of consumption revenue by multiplying it by 13, then adds monthly subscription fees multiplied by 12. The company says plainly that Run-Rate Revenue is not a substitute for revenue prepared in accordance with GAAP or a forecast of revenue. It also warns that seasonal patterns, such as dips around holiday and vacation periods in the summer and winter, may differ significantly from the annualised figure. Your target's version of this is the "exit-month annualised" number on page three of the information memorandum. It usually comes with no such warning.

$900bn of fixed commitments against floating revenue

The 2 November amendment quantifies approximately $895.4 billion of aggregate committed spending, substantially all related to compute and infrastructure. On top of that, the filing estimates $150-175 billion of payments under certain arrangements without fixed dollar commitments, excluded from the aggregate. A further amendment on 10 November puts the total at about $900.9 billion. The original filing had less: the October 26 filing left aggregate committed spend blank. That is a red flag in its own right. The most important number in the document only appeared after a revision.

The mismatch itself is spelled out in the risk factors: revenue is primarily consumption-based, and much of it is not subject to long-term contractual commitments, while compute and infrastructure commitments generally extend over multiple years. One analysis of the first filing noted that 80% of first-half 2026 revenue was consumption-based and can fall if customers cut back or shift to cheaper models.

The third quarter already showed the strain. Revenue grew 46.7% sequentially, while compute expense grew 77.5%. Anthropic attributes the pressure to $5 billion of compute under short-term commitments, which currently costs more than comparable multi-year capacity, and approximately $2 billion of non-cash donation-matching charges.

Control sits with a trust and a single founder share

According to the filing, the Long-Term Benefit Trust generally elects four of seven directors. Its Class T shares carry no economic interest. Separately, a founder-held share carries 50.1% of voting power on matters submitted to common stockholders. Anthropic itself notes that interpreting this structure increases the risk and unpredictability of governance-related litigation. Public investors are buying the economics without the control. In M&A terms, that is a minority stake priced like a majority one.

Anthropic's valuation moved more than its filing did

The price history shows how far sentiment can run ahead of the documents. On a fully diluted basis the company could be worth up to $1.405 trillion, up from the $965 billion valuation it announced in a funding round in May. In September, it had been eyeing a valuation of around $2 trillion, according to an earlier draft filing. If it prices at the midpoint, Anthropic would raise $61 billion and have a fully diluted valuation of $1.33 trillion. At the top of the price range, it would raise $64.4 billion and be valued at $1.4 trillion. Including a concurrent private placement of about $26 billion, with investors including Google, Nvidia, and AMD, Anthropic expects net proceeds of about $84.6 billion at the midpoint.

The macro backdrop explains some of the swing. Bloomberg reported that the Democratic sweep of Congress raised the prospect of greater scrutiny over data center expansion, and the Cboe Volatility Index rose to the highest since April 2025. The final price will not be known until the deal prices, expected on November 18. Any figure above $1.4tn is still banker preparation, not a result.

The bull case has real numbers behind it. Compute and infrastructure expenses fell from 177% of revenue in Q2 2025 to 48% in Q2 2026. The bear case has numbers too. For the first nine months, compute expense was $18.8 billion, or 57% of $33.2 billion of revenue, and $6.5 billion of that was short-term compute.

Worked example: three ratios from the filing

Using figures from the filing and its analysts:

• Q3 GAAP revenue of $16.96bn times four gives $67.8bn of annualised recognised revenue.
• MLQ.ai puts roughly $81 billion of Run-Rate Revenue reported for Oct. 31. $81bn divided by $67.8bn is about 1.19, so the headline sits roughly 19% above last quarter annualised.
• Commitments of $900.9bn divided by $81bn come to about 11 times run-rate. Against recognised revenue of $67.8bn the multiple is about 13 times.

A labelled hypothetical shows how the first ratio plays out in a private deal. Suppose a target books $50m of usage revenue in its best 28 days. Multiplied by 13, that is a $650m run-rate. If trailing twelve-month GAAP revenue is $480m, the $170m gap is what the seller is asking you to pay for on faith. At Anthropic's growth rates a 19% gap is plausible. For a flat business, a gap that size means the number was manufactured.

Which due diligence red flags do CFOs miss most often?

The red flags CFOs miss most often sit outside the income statement: how metrics are defined, fixed obligations that are not yet on the balance sheet, and concentration of control or talent. Quality-of-earnings work catches the accrualsaccrualsAccrual accounting records revenue and expenses when they are earned or incurred, not when cash changes hands, giving a more accurate picture of financial performance.View full definition →. It rarely catches these three. Three other companies in this week's news show each one.

CoreWeave shows the fixed-versus-floating problem and one way to manage it. Going into its results, its 9.625% notes due 2032 yielded more than 16%. Its answer is a planned chip-backed loan that is expected to cost about 7% to 7.5%, which it described as less than half the yield on its unsecured debt. On the earnings call, management said it has collected prepayments on 70% of its deals with customers this year, funding up to 100% of the contract value upfront. By design, loan tenors are linked to initial customer contracts; when the initial contracts expire, the debt is fully repaid. That tenor matching is the model to copy. The leftover risk is timing. Finimize notes that prepayments typically sit as deferred revenuedeferred revenueCash a company has collected for goods or services it has not yet delivered. It sits on the balance sheet as a liability until earned.View full definition → while chip and data-center invoices have to be paid now, and that the $9 billion prepayment wave still trails its $14.5 billion quarterly capexcapexCapital Expenditure (CapEx) is money spent to acquire, upgrade, or extend long-lived assets like equipment, property, or software that deliver value over multiple years.View full definition →.

Nvidia shows the key-person flag that disclosure documents leave out. Its board hired Wachtell, Lipton, Rosen & Katz to investigate whether company secrets or national security was compromised, after Huang admitted to having an affair with a woman who has been connected to Chinese intelligence. Yet Nvidia's annual report does not name Huang as a key person. The board's tenure is a data point as well: it has a combined 161 years of service. A shareholder class action has been filed against Nvidia and Jensen Huang in federal court in San Jose. When a risk factor is missing, that does not mean the risk is.

Weil, Gotshal & Manges shows what happens when the asset is people who can leave. Partners at the 95-year-old law firm voted to approve a wind-down plan, after a series of partner defections that ultimately led to a loss of confidence in its ability to continue operating. Rescue talks with Vinson & Elkins and Willkie Farr didn't go anywhere. The liabilities left behind include capital contributions owed to some former partners and ongoing obligations to retirees. Anyone buying an accounting, consulting or advisory firm in 2026 faces the same structure at smaller scale.

The three-ratio test

Write this on the whiteboard before the data room opens:

• Fixed commitments divided by contracted revenue (contracted, not run-rate), over the life of the commitments.
• Headline metric divided by trailing GAAP revenue.
• Voting control divided by economic ownership, for every holder and every key individual.

The rule:any ratio far from 1 gets more diligence hours than the income statement. My working thresholds are a heuristic, not an industry standard. Above 1 on the first ratio means you are buying a financing problem. Above 1.2 on the second means you need a month-by-month bridge. On the third, any holder with control and no economics, or any one person whose exit moves revenue, gets a dedicated workstream. Whatever you find should end up inspecific indemnities, escrows or price adjustments, not in a footnote of the investment memo.

Your context differs from Anthropic's in one important way. An S-1 carries securities-law liability, which forces disclosure. A private seller owes you only what the SPA requires, so all three ratios have to be requested, rebuilt and warranted. The general method is covered inthe financial due diligence playbook. The three ratios tell you where to point it.

What to do on Monday before your next deal

• Ask for the exact formula behind every annualised or "run-rate" figure in the information memorandum, then rebuild it from monthly GAAP revenue for the last 24 months.
• Request one schedule of all purchase commitments, take-or-pay contracts and leases signed but not yet commenced. Anthropic's split between roughly $320 billion in existing contractual obligations and about $451 billion in signed leases that have not commenced yet is a good template.
• Line up each obligation's end date against the end date of the customer contracts paying for it, and flag every obligation that outlasts its revenue.
• List every share class, trust, golden share and side letter in the cap table, with its votes and its economics side by side.
• Name the ten people whose departure would move revenue, check whether the risk disclosures mention any of them, and note how long each board member has served.
• For professional services targets, record notice periods and client concentration for every partner who brings in more than 5% of revenue.

Anthropic's bankers can raise the price because demand is strong. Demand does not change the $900.9bn commitment schedule. Before you agree a multiple on your next deal, find the target's equivalent of that line.