Anthropic filed a confidential draft S-1 with the SEC on Monday June 1, the first formal step toward an initial public offering reportedly targeted for October 2026 at a valuation that begins where the $65 billion Series H closed four days earlier — a post-money mark of $965 billion. The filing itself stays confidential until fifteen days before any roadshow, but the company is now publicly attaching numbers to the trajectory: annualised revenue run rate crossed $47 billion earlier in May, total revenue is approximately $10.9 billion and more than doubling year over year, and management expects to post its first operating profit during the June quarter with a break-even target of 2028 — two years ahead of OpenAI’s 2030 self-disclosed timeline. The S-1 itself is procedural. The $47 billion is the data point.

That number sits inside an income-statement comparator set it reshapes by being part of it. Salesforce’s full-year fiscal 2026 revenue was $41.5 billion. ServiceNow’s annualised run rate at the end of Q1 2026 is in the $15 billion range. Workday’s FY27 subscription revenue is guided to $9.925–9.950 billion. Snowflake’s most recent quarterly product revenue annualises to roughly $5.3 billion. Datadog cleared a $4 billion run rate at its first $1 billion quarter. The AI-native vendor that did not exist seven years ago is now reporting an ARR run rate larger than the four-year revenue of any other named SaaS vendor on the Inferential ledger except Salesforce, and is converging on Salesforce within a margin that the next two quarterly prints will close. The framing question that has driven the H1 hypothesis since the seat-compression evidence first arrived in March is whether value lost at the application layer would accrue back to AI-native challengers or be re-captured by the same SaaS incumbents at the layer above them. The May prints from Workday, Salesforce, Snowflake, and the rest of the Seven-Vendor Week cohort argued for the latter. Anthropic’s $47 billion run rate argues, hard, for both being true at once.

The unit economics implied by the filing are the part that complicates the v16 framing most directly. Hypothesis v16 reads: “Software loses value at the vendor that cannot re-derive customer-specific context cheaply; value expands where context is held by incumbents who can monetise it through AI-substitution layers above it; cloud-AI infrastructure moats remain conditional on two cash-burning private labs honouring contracts that account for half of the hyperscaler backlog.” Two of those three clauses now have to bend. Anthropic is no longer a cash-burning private lab in the Borrowed Demand sense — a vendor with $47 billion ARR generating a first operating profit in Q2 2026 and a 2028 break-even target is, by the most conservative accounting, a business whose contracts are not conditional on continued investor patience. The $200 billion Anthropic commitment to Google Cloud over five years from 2027, which Borrowed Demand framed as roughly 40% of Alphabet’s $462 billion Q1 2026 backlog and a key conditional liability for the cloud-AI fortress thesis, is now backed by a real ARR run rate that is rising into the implied $40 billion-per-year spend rather than against it. The “circular financing” framing — Alphabet’s $10–40 billion equity in Anthropic feeding the $200 billion contract back to Alphabet — looks different when the vendor receiving the equity is independently generating cash at the contract scale. Not entirely different, because the equity loop is still real, but no longer the central interpretive frame.

The second clause that bends is “value expands where context is held by incumbents.” The incumbent-substitution-capture pattern is verified — the May earnings season produced direct evidence at Workday, Salesforce, Snowflake, and the Dell-Marvell-Okta-MongoDB cohort — but it is not the only pattern in the data anymore. The challenger-capture pattern is also producing observable financial signature, at scale that exceeds any single incumbent’s agent line. Salesforce’s Agentforce ARR is $1.2 billion. Workday’s agentic AI annualised revenue is approximately $500 million. ServiceNow’s Now Assist is on track to $1.5 billion in 2026 ACV. The largest single incumbent-substitution-capture number we have observed is Microsoft’s $2 billion Copilot ARR disclosed yesterday — and Microsoft owns the operating system and the most widely deployed productivity suite in enterprise software. Anthropic’s $47 billion ARR, captured from selling Claude API access and Claude Cowork into the same enterprises those incumbents are selling agent SKUs into, is twenty-three times the largest incumbent agent-line number we have on the ledger. Both patterns are producing revenue at scale; the relative magnitudes are now what the hypothesis has to weight.

A few qualifications, written in service of the strict source-tier rule. The $47 billion figure is Anthropic-disclosed, not yet subject to SEC-grade audit; the full audited financials will not be public until the company elects to flip the S-1 from confidential to public, which by company guidance happens fifteen days before a roadshow. The Series H round was first reported by the Wall Street Journal and confirmed by the company; the $10.9 billion revenue figure and the 2028 break-even guidance also come through WSJ reporting. The October 2026 roadshow window is described as a target by multiple outlets but is not a company-confirmed date. The reported figures are large enough that the qualitative read does not change if any individual line item is restated in audit — but the headline ARR number specifically is the one to watch when audited financials land. If the $47 billion holds in the audited filing, the hypothesis revision below stands; if it is materially restated downward, the revision tightens.

Hypothesis 52% → 52%. The deterministic score holds because the cumulative evidence base is now large enough (109 items) that even a weight-5 single-event update produces less than a full-point ratio change; the qualitative read is complicate by roughly three points, captured in the version-bump scope statement below. Complicate, weight 5 (the AI-native challenger value-capture pattern is now documented at a scale — $47 billion ARR, $10.9 billion revenue, Q2 2026 first operating profit, $965 billion Series H valuation, confidential S-1 with October target — that is larger than the largest incumbent-substitution-capture number we have observed; the “value loss at the substitution-vulnerable vendor accrues to incumbents who hold context” reading of v16 cannot absorb a single non-incumbent vendor whose ARR exceeds the four-year revenue of all but one SaaS incumbent on the ledger). Verify, weight 2 (the existence of $47 billion in AI-native revenue does verify that software is in fact losing budget share to AI substitution at scale — the spend is real and the substitution is delivering — and the cloud-AI infrastructure conditionality clause in v16 retains a residual cash-flow risk through the OpenAI side of the duopoly, where the comparable disclosures have not yet landed). Net complicate by three points, driven by the magnitude of the disclosed number relative to the incumbent comparator set rather than by any new mechanism.

Hypothesis v16 → v17 is now in scope and likely warranted, but holding the revision until two adjacent data points land: OpenAI’s own forthcoming S-1 (the Information has reported October targeting for OpenAI too, and the comparable ARR disclosure will determine whether this is a two-vendor or one-vendor capture phenomenon) and the Mythos GA event Anthropic has flagged for “the coming weeks” (which will resolve the capability-frontier question the H2 thesis is built around in the same window). Both are expected inside June. A version bump that incorporates both lands the revised hypothesis on data rather than on a single vendor’s disclosure.

What to watch: the public S-1 flip and audited financial disclosure, expected fifteen days before any roadshow; the OpenAI S-1 if it lands on the reported October roadshow timeline; first quarterly disclosure from a comparable AI-native that prices its ARR run rate publicly (xAI, Mistral, DeepSeek none of which have indicated near-term IPO plans); whether the $47 billion ARR composition skews to one or two enterprise contracts vs. broad enterprise distribution (the KPMG and PwC deals from earlier in May suggest the latter, but the audited filing will say which); and whether Salesforce, Workday, or ServiceNow update full-year guidance in Q2 prints to reflect any directly-attributed pressure from Anthropic-mediated substitution rather than the agent-line capture they reported in May.

Sources: Anthropic — confidential S-1 SEC announcement; Washington Post — Anthropic files for IPO; How2Shout — Anthropic IPO at $965B; Bitcoin News — Anthropic $965B valuation target; SiliconAngle — $65B Series H; Yellow — October IPO target; Anthropic — KPMG partnership; Anthropic — PwC partnership.