OFFTAKE DIAL · ANTHROPIC + OPENAI + XAI RUN-RATE
Lab ARR
is the offtake dial.
After AllVentured flagged apparent ARR growth deceleration, Bubble Burster tracks sourced annualized revenue / ARR prints for Gerstner's three labs — Anthropic, OpenAI, and xAI (owned by SpaceX, SPCX) — company disclosures and SEC filings first, then Reuters / Bloomberg / Axios / NYT. Annualized run-rate is often month×12, not GAAP booked revenue; the labs may not account alike. xAI counts only Grok-type model sales — no X ads or compute leasing — to match what OpenAI and Anthropic sell. Gerstner's Flight Path hurdles sit as offtake framing on the chart (stall / takeoff / YE combined), not as disclosed ARR.
01 · COMBINED LINE + GERSTNER OVERLAY
Prints over time. Hurdles labeled.
As-of Nov 11, 2026 · 20 sourced points (7 company · 5 reported · 2 gross-comparable). Horizontal bands annualize Gerstner monthly offtake (×12) for chart framing only. Compare those hurdles with the Combined Anthropic + OpenAI + xAI lines — Gerstner's $180B counts all three labs — not any lab alone. The headline combined is company-reported; the ghost line is gross-comparable.
DEFINITIONAL GAP Anthropic counts run-rate revenue gross, including sales through cloud partners (AWS, Google Cloud); OpenAI counts net, excluding partner revenue. OpenAI told investors ~$50B at end-September (FT, CNBC, Oct 8); the widely reported ~$68–70B (and August’s ~$40B) were investor adjustments grossing OpenAI up to Anthropic’s method. Gerstner’s ≥$180B framing does not specify a basis, so both combined bases are shown — company-reported as the headline line, gross-comparable as a ghost line.
THREE LABS + COMBINED · SOURCED RUN-RATE PRINTS
Anthropic + OpenAI + xAI annualized revenue
Dated ARR / annualized run-rate prints plus a forward-filled Combined line. COMPANY disclosures, press-reported figures, third-party rumors, and modeled YE checkpoints are labeled separately so a TickerTrends track is never confused with an Anthropic blog print — or a Gerstner offtake hurdle. xAI counts only Grok-type model revenue (no X ads, X subscriptions or compute leasing) to match what OpenAI and Anthropic sell; SpaceX discloses no Grok-only line, so its last print is Bloomberg's Q3 2025. Gerstner horizontals are compared with the three-lab combined lines, not any lab alone: the solid-dash headline is company-reported (OpenAI net ~$50B); the faint ghost is gross-comparable (OpenAI grossed up to Anthropic's cloud-partner method, ~$70B).
Series starts near YE 2024 / early 2025 disclosed run-rates. Horizontal bands are Gerstner offtake hurdles (monthly × 12), not GAAP ARR.
SOURCE ROSTER · EVERY POINT LINKED
Hover the chart or read the table · newest first · on small screens, tap + for kind, note, and source. Digests should append to app/lab-arr/data.ts.
- ANTHROPIC$110B◇ MODELEDInvestor modeled YE 2026 · $100–120B (mid $110B) — MODELED only — investor expectations for YE 2026 run-rate in the $100–120B band (FT/Axios era reporting). Not an actual print. Midpoint charted for geometry; The Sep NYT projection (top $100B in 2026) sits at the floor of that band. Oct 15 reports conflict: Reuters says analysts who met execs this week estimate the run rate could reach ~$110B by YE (CNBC cites WSJ investors expecting $110B), while Euronews/Yahoo says annualized revenue is now reportedly expected to end 2026 below $100B.FT / Axios (investor modeled) ↗
- OPENAI$70B◇ MODELEDProjected (Bloomberg, Oct 8): reach or exceed ~$70B annualized by YE 2026 (net) — PROJECTED, not an actual print — Bloomberg Oct 8 (Torrence/Ghaffary, people familiar): OpenAI expects to reach or exceed $70B in annualized revenue by end of 2026, driven largely by enterprise; ~$50B at end-September on the same (net) basis. First sourced OpenAI YE run-rate target. Net basis — excludes cloud-partner sales Anthropic counts; no gross-comparable YE figure is sourced, so the same $70B is used for both combined projections (understates the gross-comparable view). Charted at YE for geometry.Bloomberg ↗
- XAI · SPCX$2B◇ MODELEDProjected (Bloomberg, Jun 2025): xAI told investors 2026 revenue “north of $2B” — PROJECTED, not an actual print — Bloomberg Jun 17, 2025: xAI revenue expected at $500M in 2025, “rising to north of $2 billion next year,” investors were told. Full-year 2026 revenue target (not a YE run rate), set before the SpaceX merger and the compute-leasing business; Bloomberg later reported xAI had only >$200M through Sep 2025 vs the $500M goal. Charted at YE for geometry.Bloomberg ↗
- ANTHROPIC$81B◆ COMPANYS-1/A · run-rate ~$81B at Oct 31 — Company, in its amended S-1 (filed Nov 10, 2026): “our Run-Rate Revenue was approximately $81 billion as of October 31, 2026, up $10 billion from September 30, 2026.” That implies ~$71B at end-September, above the $67.2B quarter-average annualization. Anthropic counts gross, incl. cloud-partner sales. Latest Anthropic company print.Anthropic (SEC S-1/A No. 2) ↗
- ANTHROPIC$74B○ RUMOR / 3PSemiAnalysis 3P estimate ~$74B annualized (via Euronews) — Third-party estimate, not a primary print — SemiAnalysis (via Euronews/Yahoo, Oct 15) puts Anthropic’s current annualized revenue at ~$74B, vs ~$65B in late July. Sits above the $67.2B Q3 quarter-average annualization, consistent with an exit run rate above the quarter average; treat as RUMOR / 3P.SemiAnalysis (via Euronews) ↗
- ANTHROPIC$67.2B● REPORTEDQ3 2026 revenue $16.8B × 4 ≈ $67.2B (told investors Oct 14) — CNBC (Oct 14, people familiar; FT, Bloomberg, Reuters, BBC also reported): Anthropic told investors at an Oct 14 pre-IPO meeting that Q3 2026 revenue was $16.8B, up from ~$11.6B in Q2 (some outlets say $11.5B) and $4.7B in Q1 — ~46% QoQ, slower than the >2× Q1→Q2 jump. Also adjusted-profitable and cash-flow positive (~11% adjusted operating margin; $1.70 revenue per compute dollar). Bubble Burster annualizes quarter × 4 (same method as xAI): a quarter-average, not an exit run rate, so it understates the end-Sep pace; no new company run-rate figure was disclosed. Not GAAP-audited (pre-S-1). Anthropic counts gross, incl. cloud-partner sales. Superseded as the latest Anthropic print by the S-1/A’s ~$81B run-rate at Oct 31.CNBC / FT / Bloomberg ↗
- ANTHROPIC$71B◆ COMPANYImplied by S-1/A · run-rate ~$71B at Sep 30 — Implied by S-1/A No. 2 (filed Nov 10, 2026): run-rate of “approximately $81 billion as of October 31, 2026, up $10 billion from September 30, 2026” → ~$71B at Sep 30. Derived, not a stated figure. Anthropic counts gross, incl. cloud-partner sales.Anthropic (SEC S-1/A No. 2), implied ↗
- OPENAI$50B● REPORTEDCompany-reported (net) ~$50B at end-Sep (FT + CNBC, Oct 8) — FT (Oct 8): OpenAI told investors annualized revenue was nearing $50B at the end of September — ~$20B below the ~$70B previously signalled. CNBC independently confirmed ~$50B (https://www.cnbc.com/2026/10/08/open-ai-revenue-nvidia-oracle-coreweave.html), plus 77% total run-rate growth in Q3 and 107% enterprise run-rate growth. CFO Sarah Friar put Q3 2026 revenue at about $10B (up nearly 50% QoQ) on a run rate of roughly $50B on Nov 10 (https://finance.yahoo.com/technology/article/openai-cfo-q3-was-a-blockbuster-quarter-revenue-jumped-50-225023234.html) — no new as-of date, so no new point is charted. OpenAI’s own metric is net: it excludes sales through cloud partners, which Anthropic counts. Headline company-reported OpenAI print.FT + CNBC ↗
- OPENAI$70B◎ GROSS-COMPARABLE (INVESTOR-ADJ.)Gross-comparable ~$70B (investor-adjusted to Anthropic’s method incl. cloud-partner sales; Axios) — NOT a company-reported figure. Axios Sep 29 reported a run rate of almost $70B (widely cited as ~$68–70B). Per the FT and CNBC (Oct 8), that figure included gross revenue from OpenAI’s partners — investors grossing OpenAI up to Anthropic’s method so the two compare directly. Kept as the gross-comparable OpenAI print ($70B used); not in the company-reported combined.Axios ↗
- ANTHROPIC$76.8B○ RUMOR / 3PTickerTrends 3P track ~$76.8B (low confidence) — Third-party tracker (also cited in deceleration chatter). Sits between the >$65B end-July print and the NYT >$100B 2026 projection — treat as low-confidence RUMOR, not a primary print.TickerTrends ↗
- ANTHROPIC$100B◇ MODELEDProjected (NYT via Bloomberg, Sep 18): top $100B in 2026 — PROJECTED, not an actual print — Bloomberg Sep 18 citing NYT: Anthropic “is expected to generate more than $100 billion in annualized revenue this year.” Axios’ same-day summary called it a >$100B “pace.” Latest company print is the S-1/A’s ~$81B run-rate at Oct 31 (after Q3 2026 revenue of $16.8B, ≈$67.2B annualized quarter-average). Oct 15 reports conflict on YE: Reuters says analysts who met execs estimate the run rate could reach ~$110B by YE (CNBC cites WSJ investors expecting $110B), while Euronews/Yahoo says annualized revenue is now reportedly expected to end 2026 below $100B.Bloomberg (NYT) ↗
- OPENAI$40B◎ GROSS-COMPARABLE (INVESTOR-ADJ.)Gross-comparable >$40B (investor-adjusted to Anthropic’s method incl. cloud-partner sales; Bloomberg) — NOT a company-reported figure. Bloomberg Aug 13: on track for >$40B annualized (Axios Aug 17 also cited ~$40B). Per the FT (Oct 8, via Investing.com), the August ~$40B came from investors adjusting OpenAI’s revenue up to Anthropic’s method (incl. sales through cloud partners) — gross-comparable, not OpenAI’s own net metric. Used only in the gross-comparable combined basis.Bloomberg ↗
- ANTHROPIC$65B● REPORTEDEnd-July run-rate >$65B (Reuters, Aug 17) — Reuters Aug 17 (source familiar; Bloomberg first reported): annual revenue run rate topped $65B by end of July, up from $47B in May and ~$9B at YE 2025. Last Anthropic run-rate print before the Q3 2026 revenue disclosure (Oct 14).Reuters ↗
- ANTHROPIC$47B◆ COMPANYSeries H era · crossed $47B (earlier May) — Series H blog (May 28): run-rate revenue “crossed $47 billion earlier this month.” Date pinned mid-May.Anthropic ↗
- ANTHROPIC$30B◆ COMPANYAmazon compute post · >$30B run-rate — Company: “Our run-rate revenue has now surpassed $30 billion.”Anthropic ↗
- OPENAI$25B● REPORTEDEnd-Feb ~$25B annualized (The Information) — Reuters Mar 4: The Information said OpenAI topped $25B annualized as of end-Feb (~17% above a ~$21.4B YE figure). Reuters could not independently verify.Reuters (The Information) ↗
- ANTHROPIC$14B◆ COMPANYSeries G disclosure · $14B run-rate — Company: “Today, our run-rate revenue is $14 billion,” growing >10× annually for three years.Anthropic ↗
- ANTHROPIC$9B◆ COMPANYYE 2025 run-rate ~$9B — Company Amazon compute post (Apr 20, 2026): run-rate “up from approximately $9 billion at the end of 2025.”Anthropic ↗
- OPENAI$20B◆ COMPANYYE 2025 annualized >$20B (Friar) — CFO Sarah Friar blog (via Reuters Jan 19, 2026): annualized revenue surpassed $20B in 2025 (from $6B in 2024). The Information later cited ~$21.4B YE run-rate.Reuters (OpenAI / Friar) ↗
- XAI · SPCX$428M● REPORTEDQ3 2025 xAI revenue $107M × 4 ≈ $428M (Bloomberg; reported quarter annualized) — Bloomberg (Jan 8–9, 2026, via investor documents; Reuters relayed, could not verify): xAI revenue “nearly doubled quarter-over-quarter to $107 million” in the quarter ended Sep 30, 2025; >$200M through September. This is xAI’s own AI revenue, ex-X (SpaceX’s S-1 puts the full X-inclusive AI segment at $3.2B for 2025). Bubble Burster annualizes quarter × 4. Latest Grok-only print — stale; it is forward-filled into the combined line.Bloomberg (internal documents) ↗
02 · DECELERATION COLOR (NOT GOSPEL)
Second derivative is interesting. One X post isn't.
AllVentured highlighted apparent MoM ARR slowing via third-party trackers (e.g. TickerTrends-style Anthropic ~$76.8B with ~1.8% sequential). That second-derivative read is worth watching — especially if company-adjacent leaks corroborate. It is not, by itself, proof that “decline is near.”
Counterweight: OpenAI told investors it was nearing ~$50B annualized at end-September on its own net basis (FT, CNBC, Oct 8), with 77% run-rate growth in Q3 and 107% enterprise growth; the widely reported ~$70B was investors grossing OpenAI up to Anthropic's method. CFO Sarah Friar put Q3 revenue at about $10B on a roughly $50B run rate (Nov 10). Anthropic's amended S-1 (Nov 10) puts its run-rate at about $81B at Oct 31, up $10B in a month and up from Reuters' >$65B at end-July, and xAI's last Grok-only print is Bloomberg's Q3 2025 revenue of $107M (~$428M annualized). Company-reported three-lab combined actual: ~$131.4B, ~$48.6B short of Gerstner's YE ≥~$180B check; gross-comparable (OpenAI at ~$70B): ~$151.4B, ~$28.6B short of it. xAI counts only Grok-type model sales (subscriptions, API / enterprise), not X ads, X subscriptions or Colossus compute leasing. On projections — NYT (via Bloomberg, Sep 18) has Anthropic topping $100B in 2026, Bloomberg reported xAI told investors 2026 revenue would be north of $2B, and OpenAI is held at its latest print — the projected combined is ~$172B reported (Anthropic $100B proj. + OpenAI $70B proj. + xAI (SpaceX) $2B proj.), ~$8B short of the $180B line, or ~$172B gross-comparable (Anthropic $100B proj. + OpenAI $70B proj. + xAI (SpaceX) $2B proj.), ~$8B short of it. Neither projection is an actual print. Treat tracker-only MoM prints as low-confidence until a company blog, Bloomberg, Reuters, Axios, NYT, or FT confirms.
03 · GERSTNER OFFTAKE FRAMING
Monthly offtake × 12 on the chart.
Credit: Brad Gerstner / Altimeter on All-In “Flight Path: Balance of 2026,” episode Sep 17 2026; All-In Summit remarks via Benzinga (Sep 18 2026).
Three labs, one hurdle. Gerstner’s ≥~$180B YE check counts three labs: Anthropic + OpenAI + xAI (owned by SpaceX, SPCX). He put the three at ~$100B combined run rate on July figures and said they need “at least $180 billion by year-end … just to keep the AI trade intact.” Benzinga via Webull, Sep 18 2026 ↗
- STALL ~$4B/mo → ~$48B/yr combined framing
- TAKEOFF ~$8B/mo → ~$96B/yr combined framing
- YE CHECK Anthropic + OpenAI + xAI need ≥~$180B YE 2026 (from ~$100B on July figures)
- LABEL Offtake hurdles — not disclosed lab ARR / GAAP
Takeoff ~$8B/mo AI lab offtake; stall ~$4B/mo. Three-lab (Anthropic + OpenAI + xAI) path ~$100B on July figures → need ≥~$180B YE 2026. Overlay annualizes monthly offtake (×12) for chart framing only — not GAAP booked revenue.
Timeline already carries EST Lab ARR print windows and Gerstner markers — see the Plan Timeline.
S-1 · CUSTOMER CONCENTRATION
Anthropic's top customers: who's guessed
As of Oct 26, 2026. Source: Anthropic public S-1, filed Oct 26, 2026 (EDGAR ↗).
WHAT THE FILING SAYS
- No single customer was more than 10% of revenue in the nine months to Sep 30, 2026. The top five customers were about 25%. The S-1 names none of them; US accounting rules (ASC 280) don't require it to.
- Customers at 10%+ of revenue: Customer A 12% of FY2025. Customer B 12% of FY2025 and 18% of H1 2025. Customer C 43% of FY2023 and 11% of H1 2026. Customer D 11% of FY2023.
- Accounts receivable: Customer A 13% (Dec 2025 and Jun 2026), Customer C 10% (Jun 2026).
- Cloud marketplaces (AWS, Google Cloud, Azure) were 48% of H1 2026 revenue.
WHO'S GUESSED
- REPORTEDCursor (Anysphere)
Bankers on the deal told IFR the two 12% customers are most likely Cursor and GitHub. IFR ↗
- REPORTEDGitHub Copilot (Microsoft)
Same IFR source. VentureBeat put Cursor plus GitHub at about $1.2B of about $5B, per PYMNTS ↗
- OUR INFERENCEAWS or Google as “Customer C” (marketplace resale)
Not reported. Customer C is 43% of 2023 revenue and 10% of receivables, and marketplaces are 48% of revenue (S-1).
- SPECULATIONMeta, AWS, US government, Jane Street
From The Ringer's guess list, via Daily Ferment ↗. The S-1 never mentions Meta.
- SPECULATIONSalesforce (Jon / this site)
The S-1 names it only as a partner building products on Claude (with Legora) and as a former office sublessor.
Why it matters:the reported top two are coding tools without long-term contracts that can switch providers (the S-1 says “many of our top customers are not subject to long-term contracts”), and OpenAI is offering big discounts to poach them.
04 · CROSS-LINKS
Same bet, other dials.
Offtake next to CapEx financing, token demand, private marks, and the dating board.
FINANCING
Bank of NVDACrack paths on the financed demand stack — CoreWeave as canary.DEMAND
Token runwayEvercore base-case token stack — physical twin of offtake.MARKS
Valuation raceAnthropic / OpenAI / SpaceX sourced valuation marks.DATING BOARD
Plan TimelineGerstner Flight Path markers + EST Lab ARR print windows.05 · METHODOLOGY / CAVEATS
Disclosed vs estimate. Always.
Prefer primary company blogs, then Reuters / Bloomberg / Axios / NYT / FT / Sacra. Kind labels are mandatory on every row.
- Run-rate ≠ GAAP Annualized figures often annualize a recent month or short window. Booked fiscal revenue looks different in an S-1: Anthropic's filing defines Run-Rate Revenue from trailing-28-day consumption and monthly subscriptions, and reports about $81B at Oct 31, 2026 (S-1/A, Nov 10).
- Not apples-to-apples Anthropic counts run-rate gross, including sales through cloud partners (AWS, Google Cloud); OpenAI counts net and excludes partner revenue. OpenAI's own end-September figure is ~$50B (FT, CNBC, Oct 8); the ~$40B (Aug) and ~$68–70B (Sep) prints were investor adjustments to Anthropic's method and are tagged GROSS-COMPARABLE (hollow ring markers) — kept visible, not company-reported. Gerstner's $180B does not specify a basis.
- OpenAI ~$70B: Gerstner's view (source note) Oct 9, 2026 (CT): Brad Gerstner disputed FT's framing, saying there's ‘near zero chance’ OpenAI finance called $70B net revenue and that $70B was ‘so obviously end of year not current run rate’; he later said he phrased it poorly but kept the point (X post, follow-up). This is Gerstner's view; FT and CNBC reported ~$50B net at end-September. Chart points unchanged.
- Leak quality RUMOR / third-party trackers (TickerTrends-style) stay on the roster with low confidence when they circulate in deceleration debates — they do not override COMPANY or top-tier REPORTED prints.
- Modeled YE Investor $100–120B Anthropic YE 2026 expectations are MODELED only (midpoint charted). The Sep NYT projection that Anthropic tops $100B in 2026 is also charted as a MODELED / projected point at the band floor — not an actual print. xAI's MODELED YE point is Bloomberg's reported >$2B 2026 revenue target (Jun 2025) — a full-year revenue goal, not a run rate, and set before the SpaceX merger. OpenAI's MODELED YE point is Bloomberg's Oct 8 report that it expects to reach or exceed ~$70B annualized (net) by year-end.
- YE / 2026 guidance check (updated Oct 9) OpenAI: expects to reach or exceed ~$70B annualized revenue by year-end, up from ~$50B at end-September, driven largely by enterprise (Bloomberg, Oct 8; people familiar, not public guidance; net basis). Now used in the projected combined. Earlier leaked figures were full-year revenue projections (not run rates), not public guidance: ~$36B for 2026 in a July company presentation (FT via Reuters, Sep 18) and $30B shared with investors (The Information, Feb). SpaceX: CFO Bret Johnsen guided to “$100 billion of ARR … by the end of this year based on our expected revenue in the month of December” (Q2 call, Aug 4; “on track” again at Goldman's conference in Sep). That covers all of SpaceX (Starlink, launch, Colossus cloud leasing, Cursor), not Grok, so it isn't charted. Grok only: nothing newer than Bloomberg's Jun 2025 >$2B.
- xAI = Grok model revenue only To match what OpenAI and Anthropic sell, xAI counts only Grok subscriptions (SuperGrok tiers; Grok-attributable X Premium if ever broken out) and Grok API / enterprise. EXCLUDED: X advertising, X subscriptions, data licensing, and Colossus compute leasing (e.g. Anthropic's $1.25B/mo lease, which would also double count Anthropic's spend). SpaceX's S-1 and Q2 10-Q do not break out Grok: they report “Grok and X subscription revenue” together inside “AI solutions & infrastructure,” alongside data licensing and cloud services. Stripping the $1.6B of Q2 cloud revenue leaves ~$594M for the quarter, but that is still mostly X — an upper bound, not charted. The latest Grok-only figure is Bloomberg's Q3 2025 xAI revenue ($107M, × 4), forward-filled and stale; it likely understates Grok today.
- Three-lab combined The headline combined line forward-fills each lab's latest COMPANY / REPORTED print and sums Anthropic + OpenAI + xAI only on dates where all three have data. The ghost gross-comparable line does the same but swaps in OpenAI's investor-adjusted gross-comparable prints where they exist (earlier OpenAI prints are company figures on both lines). Gerstner source: Benzinga via Webull, Sep 18 2026.
- Gerstner overlay Stall / takeoff / YE bands annualize monthly offtake (×12) for framing. They are not company-disclosed ARR lines.
- Disclaimer Annualized run-rate / ARR prints are not GAAP booked revenue (often latest month × 12). Anthropic counts cloud-partner sales (gross); OpenAI does not (net) — not apples-to-apples, so the combined is shown on a company-reported and a gross-comparable basis. xAI counts only Grok-type model revenue (subscriptions, API / enterprise) to match what OpenAI and Anthropic sell — no X advertising, X subscriptions, or Colossus compute leasing. SpaceX does not disclose a Grok-only figure, so xAI’s latest print is Bloomberg’s Q3 2025 number (stale). Leak quality varies; S-1 filings supersede leaks (Anthropic’s is public; OpenAI’s is not). Not investment advice.