The editorial view: read it as a buyer's manual, not a stock pitch
Anthropic made its S-1 registration statement public on Oct. 26, after confidentially submitting a draft to the SEC on June 1, with a planned Nasdaq listing under the ticker ANT and Goldman Sachs and Morgan Stanley leading the underwriting syndicate (Source: 证券之星; Anthropic). The 335-page filing has been called the largest IPO in history by some accounts - and the most loss-making AI company by others - but for B2B technology buyers the more useful reading is what the filing reveals about how enterprise AI spending is consolidating, and what it means for their own procurement (Source: 36氪; 每日经济新闻).
The headline numbers are striking. Revenue reached US$16.286 billion in the first half of 2026, roughly 14 times the year-earlier period and 3.5 times full-year 2025 revenue of US$4.585 billion, with Q2 alone generating US$11.555 billion in a single quarter (Source: 金融界; CodingCat.dev). The S-1 reports net revenue retention of approximately 900% as of Sept. 30, 2026 (Source: FourWeekMBA). Customers spending at least US$100,000 on a trailing-twelve-month basis exceeded 11,900 - versus just over 950 a year earlier - with more than 1,900 spending over US$1 million and about 250 over US$10 million (Source: 金融界). Revenue splits roughly 80.2% consumption-based (API usage) in the first half, about 48% flowing through cloud marketplaces, and 28% from customers billed outside the United States (Source: 36氪; FourWeekMBA).
The customer economics explain the strategy. Analysts describe it as selling Claude as a foundation model, expanding into developer tools like Claude Code, attaching first-party applications that deepen everyday use, distributing through cloud partners without forcing a single sales path, and encouraging expansion so existing accounts spend far more over time (Source: Superintelligence News). The Teardown estimates 70-75% of revenue is pay-per-token API and roughly 80% comes from business customers - making Anthropic structurally a B2B infrastructure business rather than a consumer app company, a distinction SemiAnalysis says separates it from OpenAI's consumer-led model (Source: The Teardown; druckfin.com/SemiAnalysis).
The capital side is the part B2B buyers should watch
The filing also quantifies the bet behind the product: future compute commitments of roughly US$542 billion to US$567 billion, spanning AWS Trainium, Google TPUs, Nvidia GPUs and AMD GPUs, with Google committed to at least US$114.5 billion in cloud and compute services (2026-2033), Amazon at least US$110 billion (2026-2036), and an arrangement with SpaceX's AI entity for up to about US$120.5 billion in Nvidia GPUs (Source: 36氪; 每日经济新闻/CNBC; CodingCat.dev). These commitments are the reason the company's funding rounds kept climbing: roughly US$25 billion at first investment in December 2023, US$61.5 billion in the Series E (February 2025), US$183 billion in Series F (September 2025), US$300 billion in Series G (February 2026), and a US$65 billion Series H at a US$965 billion valuation in May 2026 - at the time surpassing OpenAI's value (Source: SEC filing; Fortune; Anthropic).
That scale explains the optics of the income statement: a GAAP net loss of US$80.7 billion in H1 2026, of which US$72.4 billion stems from accounting charges tied to changes in the fair value of liabilities, with an operating deficit of US$693 million - a much smaller, cash-based picture (Source: Xinhua; AI Flare). For buyers, the implication is structural, not sentimental: the largest AI suppliers are increasingly tied to cloud providers through commitments so large that neither side can walk away, and that mutual dependency is what stabilizes - and concentrates - the market (Source: CodingCat.dev).
What this means for procurement
Three practical signals emerge for B2B buyers. First, vendor concentration is real: a supplier whose economics are anchored to two hyperscalers and a handful of compute deals can change pricing, capacity allocation or roadmap terms with limited market pushback. Second, the 48% share of revenue through cloud marketplaces means enterprises are already buying AI through their cloud commitments - procurement should treat AI spend as part of cloud negotiation, not a separate line item (Source: FourWeekMBA). Third, the 900% net revenue retention indicates that AI usage inside customer accounts expands fast once embedded; contract governance - seats, tokens, spend caps - will matter more than model choice (Source: FourWeekMBA).
Valuation skepticism should temper any extrapolation from the S-1. PitchBook's analysis of the leaked financials concluded they reflect fast growth but not a US$2 trillion valuation, and New Constructs called the offering the most overpriced IPO of 2026, modeling roughly 43% downside to the targeted valuation even under aggressive long-run scenarios (Source: Morningstar/PitchBook; New Constructs). SemiAnalysis separately modeled Anthropic reaching around US$1 billion in quarterly profit, framing the company as an infrastructure utility rather than a consumer application (Source: druckfin.com/SemiAnalysis). Investor expectations have ranged up to US$2 trillion, which would put Anthropic above SpaceX's roughly US$1.77 trillion valuation at its June listing (Source: 新浪财经).
None of that changes the buyer's core takeaway: the S-1 documents how enterprise AI demand has moved from pilot budgets to core operating workloads, with AI Flare arguing the US$16.3 billion first-half figure demonstrates frontier-model demand has moved beyond experiments into core operations (Source: AI Flare). Procurement teams that plan for consolidation, lock-in and marketplace-based buying are positioning for the market the filing describes - regardless of where the IPO prices.
