Anthropic is pacing past $100 billion and could list in November

Chronology that produces the window's story (FACT unless labelled; dates matter — the event date is the date of the reporting, 2026-09-18):

1. End-2025: Anthropic exits 2025 at about a $9B annualized revenue run rate (Reuters, Aug 2026: "according to the company"; internal goal confirmed in Oct 2025 Reuters reporting). (COMPANY CLAIM via Reuters; CONFIRMED as reported context)
2. 2025-10 → 2026-05: run rate rises from nearly $7B (Oct 2025) to the $9B end-2025 target, then $30B in April 2026 and $47B+ by early May — Anthropic publicly stated the $47B figure alongside its $65B Series H at a $965B post-money valuation (Reuters Oct 2025; NYT June 1; Yahoo Finance; Reuters). (COMPANY CLAIM / CONFIRMED-reported)
3. 2026-06-01: Anthropic confidentially files its IPO registration with the SEC (NYT); the filing "gives us the option to go public," company says; revenue run rate had crossed $47B in May. (CONFIRMED)
4. 2026-07-31: run rate reaches $65B (Bloomberg first, Aug 17; CNBC confirmed; Reuters). Reuters (Aug 15) separately reported Anthropic projected Q2 2026 revenue of at least $10.9B — more than double the prior quarter, on track for a first projected quarterly operating profit of ~$559M — and said the company's run rate grew more than 10-fold annually in each of the three years through early 2026. (COMPANY CLAIM-reported)
5. 2026-08-15: Reuters exclusive — Anthropic IPO valuation "hinges on" 2028 revenue forecasts of roughly $190B–$200B.
6. 2026-08-27: Reuters/The Information — Anthropic plans to publicly unveil the IPO prospectus after Labor Day.
7. 2026-09-03: OpenAI releases GPT-6 Astra (context: enterprise competition; via Reuters).
8. 2026-09-04/05: Reuters exclusive — IPO marketing expected to begin mid-October at the earliest, listing days before the US midterm elections in November; ~$15B revolving credit facility being finalized; banks: Morgan Stanley, Goldman Sachs, JPMorgan, Citi; "Anthropic declined to comment."
9. 2026-09-12 (Saturday, pre-window but load-bearing): Dario Amodei publishes "We Must Pace the Frontier" (darioamodei.com, ~3,800 words): "We must slow the pace at which we improve the capabilities of AI models." Three-step plan (embedded evaluators — which Anthropic commits to unilaterally; democratic coordination; global coordination). Cites recursive self-improvement and the OpenAI–Hugging Face incident. Altman ("we will do the same") and Musk ("Dario is right") publicly agree; in a Fortune interview the same day Altman says OpenAI will not go public in 2026 ("I would say not 2026"). (CONFIRMED)
10. 2026-09-16: Bloomberg-reported (via RocketNews) — Anthropic advancing a ~$2T IPO on Nasdaq; Nvidia reportedly considering a ~$10B cornerstone contribution; raised-up-to-$100B possible; MS/GS/JPM lead underwriters.
11. 2026-09-18 (the window's event):
    • NYT: expected to exceed $100B annualized revenue by end of 2026; pursues IPO "despite its own AI-safety warnings"; shares trading "by sometime in November"; ~5 GW compute expected by end-2026, roughly double by end-2027 (per CapitalSignal's reading of the NYT piece).
    • Bloomberg: "Sales Near $100 Billion Before November IPO" — debut "as soon as November."
    • WSJ: "Anthropic Shifts Planned IPO to November" — shift from October; Q3-results rationale; decision predated safety debate/antitrust suit (corroborated via PYMNTS, CapitalSignal, TradingKey, VantageMarkets).
    • Axios: >$100B pace = +50% vs two months ago; >10x vs end-2025; 50-largest-by-revenue / 10-largest-by-market-cap framing; OpenAI-stands-down contrast.
    • Reuters exclusive: considering a new model release ahead of the IPO to counter GPT-6 Astra; balancing safety evaluation with profitability; IPO could slip past the November midterms.
    • AP (filed same day, reported Sept 18/19): four paying users filed a class-action antitrust suit in the U.S. District Court for the Northern District of California against Anthropic, OpenAI, SpaceXAI and Google, alleging an illegal agreement to slow AI development; the complaint centers on the Sept 12 Amodei essay ("coordination largely took place on Sept. 12"); at pleading stage — EARLY RESEARCH. (Reported by AP via US News; also referenced by TradingKey Sep 20.)
12. 2026-09-19..09-22: syndication/follow-up coverage (US News, Rappler, WMBD, Calcalistech, Euronext, TradingKey Sep 20, VantageMarkets Sep 21, CapitalSignal ~Sep 22). Anthropic still has not confirmed any number or date; no public prospectus yet. (CONFIRMED — absence of confirmation)
13. 2026-09-19..09-22 (post-window retail-crowding signal): AI-IPO trackers (e.g., anthropic-stock.com daily updates, Sept 19–21) report retail investors crowding Anthropic pre-IPO funds as the November slip and the Accenture-embedded-evaluator deal (S16) circulate; "Retail Investors Crowd Anthropic Pre-IPO Funds" (Sep 21 update). Lower-grade tracker sources; treated as corroborating context only (RUMOR/EARLY RESEARCH grade).

• The ">$100B annualized revenue" marker entered the public record — a number that, per Axios, puts Anthropic on track to be "one of America's 50 largest companies by revenue and ten largest by market capitalization" by year-end, ~11x its own end-2025 pace, in a company founded in 2021 ($1B run rate at the start of 2025, per Reuters' Oct 2025 reporting).
• The IPO window hardened publicly from "October" to "November" — a sequence documented across Reuters (Sep 4: mid-Oct marketing, pre-midterm listing), WSJ (Sep 18: explicit November shift, Q3-results rationale, pre-dating the safety row), and Bloomberg/Axios ("as soon as November" / "by sometime in November"), with a Reuters-reported possibility of slipping past the midterms.
• The safety paradox became a market story with legal teeth: the same week NYT framed "IPO despite its own AI-safety warnings," the Sep 12 "pace the frontier" coordination was converted into a filed antitrust class action (Sep 18, N.D. Cal.) against Anthropic, OpenAI, Google, and SpaceXAI — the first judicial test of "slow down" advocacy-as-collusion, now running in parallel with the largest IPO in AI history.
• Competitive context shifted mid-roadshow-planning: OpenAI's GPT-6 Astra (Sep 3) took ~13% of enterprise AI spend tracked by Ramp vs ~8% for Anthropic's Claude Fable and led OpenRouter spend for the first time in 2.5+ years (Reuters), while Reuters reported Anthropic weighs an early new-model release to counter it — directly against the CEO's pacing thesis.
• The contrast with OpenAI's stance sharpened: Altman publicly deferred OpenAI's IPO ("not 2026") citing safety work; Anthropic moved toward listing on the same safety debate — the market reads this as either pricing in Anthropic's dominance or as a caution flag (press framing; INTERPRETATION).

• Before (through early September 2026): Anthropic is the private-market champion — $9B ARR end-2025, $47B May, $65B end-July, $965B Series H valuation (May), confidential SEC filing (June 1), prospectus promised after Labor Day, marketing slated mid-October, listing before the midterms; OpenAI (GPT-6 Astra) applying enterprise pressure; Amodei's Sep 12 essay publicly calls for pacing while Altman defers OpenAI's IPO.
• Change (2026-09-18): NYT/Bloomberg/WSJ synchronize the headline: >$100B annualized-revenue pace + November IPO; Reuters adds the new-model consideration; AP records the antitrust class action. The story converts from "when does Anthropic list?" to "how large is the largest AI IPO ever — and what does it mean that it proceeds amid its own safety warnings?"
• After: A public-market referendum on frontier-lab economics is now dated: Q3 results (October), roadshow (mid-October), possible listing days before the November midterms — against a background of the antitrust suit, the GPT-6 Astra share shift, open-weight pressure (S08 Harvey economics, S09 Xiaomi MiMo), California's kill-switch EO (S04), and the UN agent-control brief (S05). Every enterprise buyer and developer contract negotiation now carries a publicly priced reference point for Anthropic's durability.

• Run-rate mechanics: "Annualized revenue run rate" extrapolates current sales pace to a full year. $9B (end-2025) → $47B (May; +5.2x in ~5 months) → $65B (July 31; +38% in 2.5 months) → >$100B (year-end-2026E; +54% in 5 months). Axios' "+50% in two months" (≈$65B→$97-100B) is internally consistent with the NYT year-end figure — a useful sanity check (verified in the lab, Section 19).
• IPO mechanics (U.S. confidential filing path): SEC confidential submission (June 1) → public prospectus (after Labor Day per Reuters/The Information; not yet public as of Sep 22) → analyst meetings → marketing/roadshow (mid-October earliest per Reuters Sep 4; Q3 results expected to be shown first per WSJ Sep 18) → pricing → listing "as soon as November" (NYT via Bloomberg). A ~$15B revolving credit facility was being finalized in parallel (Reuters); banks working the deal: Morgan Stanley, Goldman Sachs, JPMorgan, Citi (Reuters); Bloomberg-reported elements include a ~$2T valuation target, up-to-$100B raise aspiration, Nasdaq listing, and a possible Nvidia ~$10B cornerstone (via RocketNews).
• The pacing/safety frame (why it is "despite safety warnings"): Amodei's Sep 12 essay argues capability growth must be paced (embedded evaluators with employee-like access — Anthropic committing unilaterally; that commitment is separately productized in the Accenture "embedded evaluation" deal, S16, announced Sep 18 in-window). The IPO proceeds during this campaign; the antitrust complaint alleges the campaign itself is unlawful coordination (unadjudicated — EARLY RESEARCH).
• The economics being priced: enterprise-product-led revenue (Claude Code, Cowork; ~80% enterprise mix per Oct 2025 Reuters), heavy infrastructure intensity (~5 GW compute by end-2026, ~2x by end-2027 per NYT via CapitalSignal), profitability now a stated investor focus (Reuters Aug 15: Q2 2026 projected revenue at least $10.9B, more than double prior quarter, on track for a first quarterly operating profit of ~$559M; five quarters of 10x+ annual run-rate growth), and 2028 revenue forecasts of ~$190–200B as the valuation anchor (Reuters Aug 15).

• A $100B+ run-rate company with a ~$2T IPO plan is the largest financial referendum on AI-lab economics ever staged — SpaceX's June debut ($1.77T valuation) becomes the comparator; a successful Anthropic listing would validate premium frontier pricing for the entire industry; a weak debut would undercut every lab's negotiation position.
• It resolves (directionally) the biggest open question in enterprise AI spend: is frontier-API pricing durable? The same week Bloomberg documented Harvey's margin inversion and the open-model exodus (S08), Anthropic's numbers argue the opposite for the model layer: demand is accelerating (Claude Code/Cowork), not commoditizing.
• The safety-paradox subplot — IPO while publicly urging the industry to slow down — forces investors, regulators, and buyers to confront whether "pacing" is strategy, conviction, or both; the antitrust class action makes that question justiciable.
• Calendar collision: a listing days before the US midterms (Reuters), one week after California's kill-switch EO (S04) and a UN brief on losing control of agents (S05), and against OpenAI's stated deferral — the AI-infrastructure political economy and the AI-stock story converge in one pricing event.
• For the window's own narrative: S01 (OpenAI undercutting with GPT-6 Sol/Luna), S03 (Grok 4.7), S09/S10 (open-weights at frontier quality) all landed the same week — Anthropic's $100B figure is the market's answer to "who owns the enterprise wallet," published just as that answer becomes contested.

• The first fully disclosed look at frontier-lab unit economics: the public prospectus and post-IPO reporting will expose revenue mix, gross margins (vs the Harvey −50% story), capex intensity (5 GW), and compute commitments — data that every enterprise buyer and every competitor has been pricing blind.
• A direct public-market arb between closed and open weights: if a >$100B run-rate lab with an 11x YoY pace keeps premium pricing, the open-model tide (S08/S09/S10) faces its strongest counter-evidence; if post-IPO margins disappoint, the Harvey/StepFun/Xiaomi cost logic wins.
• A new shareholder class in AI governance debates: public investors will hold voting stakes in a company whose CEO asks the industry to slow down — "pacing" becomes a shareholder-communication problem, and disclosures about advance-precision AI (per AISI-style commitments) become securities-relevant.
• Embedded-evaluator accountability at scale: Amodei's unilateral commitment (and the Accenture productization, S16) means the company's own safety claims will be verified by third parties with employee-like access — a governance structure no IPO-priced company has had before.

1. When does the public prospectus/S-1 land, and what does it disclose for revenue, gross margin, and compute commitments?
2. Is the Q3-results-before-roadshow sequencing (WSJ) bullish bookbuilding or a sign of reduced visibility into H2 demand?
3. Does the reported "new model" release arrive before listing, and does that collide with Amodei's pacing thesis?
4. Will the antitrust class action survive a motion to dismiss, and does it affect bookbuilding / pricing (e.g., enhanced risk-factor disclosure)?
5. Does Nvidia actually take a ~$10B cornerstone stake, and what does that signal about the compute partnership?
6. Where does pricing land relative to the ~$2T valuation talk — and what does that imply for OpenAI's eventual listing comp?
7. Do the open-weight releases (MiMo-V2.6 Pro AA-46; Step 5; Kimi K3 stacks) dent enterprise share before the roadshow — the exact risk Reuters' investors cite?
8. What do the California EO N-9-26 (S04) and UN brief (S05) imply for the IPO's risk-factor language and investor letter?

• Days–weeks (late Sep–early Oct): public prospectus expected (post-Labor-Day plan, Reuters/The Information); analyst meetings; Q3 preliminary numbers circulating in bookbuilding; antitrust defendants' response deadline; new-model decision (Reuters sources).
• October: Q3 results + roadshow (mid-October at earliest per Reuters Sep 4; WSJ says November shift is expressly to show Q3 first); final pricing range; Nvidia cornerstone decision (Bloomberg-reported talks).
• November: listing "as soon as November" (Bloomberg/NYT) or days before/after the US midterms (Reuters); possible slip past the elections per Reuters' two sources.
• Outlook (PREDICTION, medium confidence): the listing happens in November 2026 with a debut valuation in the $1.5–2.5T band, pricing somewhere between the $965B private round and the ~$2T talk; the S-1 will show accelerating revenue with heavy compute-intensity disclosure, and the "pacing" campaign will be disclosed as a risk factor. The antitrust class action is unlikely to block the IPO (preliminary-injunction odds low) but will follow it into 2027 as a discovery-heavy overhang; the first material post-listing event to watch is Q4 revenue retention vs the GPT-6 Astra/Ramp-share trendline.

This is the week the AI business stopped being a venture narrative and became a public-market fact. On Friday the 18th, three of the world's most careful financial newsrooms — NYT, Bloomberg, WSJ — converged on a single sentence: Anthropic is pacing past $100 billion in annualized revenue and could be public within weeks, raised against its own CEO's plea to slow the frontier down. Both halves of that sentence are unprecedented: no company has ever reached that run rate this young, and no CEO has ever taken a ~$2 trillion company public while asking the industry to hit the brakes — then watched four users sue him for saying so. The prospectus, when it lands, will be the most important document in AI economics this year: the first audited look at what frontier intelligence actually costs to make, to run, and to defend. Whether the debut prices at the top of the ~$2T talk or disappoints, the market will be voting on the same question this week's model releases argued from four different directions — is the frontier actually worth the price of admission, or is the real signal the 1/10-cost open models shipping the same week? Every number in this file is still company-claimed and press-carried; the S-1 is the moment the claims become evidence.

Evidence labels used: CONFIRMED (independently corroborated events and reporting), COMPANY CLAIM (company-sourced financial figures carried by press, unconfirmed by Anthropic), INDEPENDENTLY VERIFIED (multi-outlet confirmation of the reporting itself), EARLY RESEARCH (the Sep 18 antitrust class action — filed, unadjudicated), INTERPRETATION/PREDICTION (analyst views). Discovery-record inaccuracies (Amodei Sep 19 attribution; additions in-window) are corrected in Section 1.