Anthropic S-1 and OpenAI Funding: 2026 Financials

Anthropic filed its S-1 on October 26, 2026 (amended November 2), disclosing hypergrowth in usage-based revenue alongside a massive locked-in compute bill and mostly non-cash losses; OpenAI’s parallel October funding round was cut from a $30B target at ~$1.4T to ~$20B at $1.25T amid market caution, with UAE investors anchoring via equity paired to regional compute capacity.[1][2]

These disclosures provided the first detailed public view of frontier AI economics, highlighting rapid scaling (revenue per compute dollar rising sharply) but also rigid future obligations versus pay-as-you-go revenue. Bubble skeptics emphasized the asymmetry and high multiples; bulls pointed to operating leverage, retention, and strategic compute deals as moats. Most Anthropic numbers are now SEC-filed; OpenAI details remain press-reported.

Anthropic S-1 Revenue, Growth, and Margins (Filed/Confirmed)

Anthropic’s core business—metered token usage of Claude models plus subscriptions—scaled explosively but showed clear deceleration in Q3. The S-1 and amendment make these figures official.[3][4]

• H1 2026 revenue reached $16.286B (up ~1,300% YoY from $1.163B), with Q2 alone at $11.555B; Q3 added $16.956B (~$17B), up 47% QoQ but 1,127% YoY, for nine-month 2026 revenue of $33.24B (vs. ~$2.55B prior year). FY2025 revenue was $4.585B.[5]
• Net revenue retention stood at ~900% as of Sept. 30, 2026, with >1,900 customers paying $1M+ annually.[5]
• Implied gross margin (cost of revenue, largely inference compute) reached ~85% in H1 2026 ($2.38B cost on $16.286B revenue), up from ~40% in FY2025 and ~12% in H1 2025.[6]
• Compute efficiency improved: revenue per compute dollar rose from ~0.40x in Q1 2025 to ~2.08x in Q2 2026 before slipping in Q3 due to expensive short-term capacity.[7]

Implication: The mechanism—usage-based pricing auto-scaling with customer activity—drove retention and early operating profit in Q2 ($974M GAAP operating income on $11.555B revenue, first profitable quarter reported). However, Q3 swung to a $567M operating loss, partly from $5B in costly short-term compute. Competitors entering this space must match or exceed this efficiency curve or risk margin compression as capacity ramps.[8]

Losses, Cash Flow, and Non-Cash Items (Filed)

The headline net losses are dominated by accounting adjustments rather than cash burn.[9]

• H1 2026 operating loss: $693M (narrowed from $2.135B YoY); nine-month operating loss ~$1.26B on $33.2B revenue. Q2 was profitable; Q3 was not.[4]
• H1 net loss: $80.7B, of which ~$72.4B was non-cash fair-value charges on convertible notes, derivatives, and SAFEs (valuation-driven markups). Nine-month net loss: $104B (similarly mostly non-cash).[10]
• Operating cash flow turned positive: $3.9B in H1 2026. Cash/short-term investments: ~$132B at June 30 (plus later raises).[3]

Implication: The non-cash charges reflect rising private valuations feeding back into liability accounting—common in high-growth private companies with complex instruments. This structure inflates reported losses without draining cash, but it signals dilution risk for new investors. Anti-bubble voices note the gap between GAAP and adjusted metrics (e.g., adjusted operating income of $1.618B in H1).

Compute Commitments, Structure, and Risks (Filed)

The S-1 quantified the scale of future obligations for the first time.[1]

• Total commitments: $895.4B (substantially all compute/infrastructure), with ~$69.8B due in the next 12 months, $447.5B–$448B by end-2030, and the balance to 2045. Major partners include Broadcom (~$161B), Google (~$121B), Amazon (~$110B), SpaceX AI (up to $120.5B), and others.[8]
• Many are take-or-pay or minimum commitments payable regardless of usage; some leases had not yet commenced at filing. Short-term purchases added costs in Q3.[11]
• Anthropic notes risks if revenue expectations are not met or funding is unavailable; it typically books capacity 12–18 months ahead.

Implication: This creates a rigid cost base against flexible, usage-driven revenue. Efficiency gains (higher revenue per dollar) have helped, but new capacity could compress margins unless utilization scales. For entrants or competitors, securing similar long-term deals is essential but capital-intensive; vendor financing or sovereign partnerships (see OpenAI below) may be required.

OpenAI October 2026 Funding Round (Reported/Leaked)

OpenAI (not yet public) pursued a bridge round positioned as pre-IPO capital amid overlapping investor pools with Anthropic.[2]

• Target cut from ~$30B at $1.4T valuation (pre-money) to ~$20B at $1.25T due to tech selloff, tepid institutional interest, and Anthropic’s filing. ~$11B committed by late October; first close expected below $20B, with later retail/family office tranche.[12]
• Lead/anchor: UAE consortium (MGX etc., largest participant) + Coatue; participants include Founders Fund, Sequoia, a16z, SoftBank. UAE portion includes option/right of first refusal on new regional compute capacity (e.g., Stargate UAE cluster).[13]
• Context: Follows March 2026 $122B round at $852B post-money. Company cited ~$50B annualized revenue end-Sept (reported), with plans for $600B compute spend by 2030.[14]

Implication (Circular/Vendor Financing): The UAE structure ties equity to compute supply, reducing pure financial risk for investors while giving OpenAI capacity access. This hybrid resembles vendor financing and highlights how sovereign or infrastructure players de-risk frontier spend. It pressured the round size/price as institutions weighed Anthropic’s disclosed obligations.

Valuations, IPO Plans, and Commentary

• Anthropic: Reports peg IPO target at $1.5T–$2T (potentially raising $100B+); analysts vary (e.g., PitchBook ~$1.42T base). Multiples cited around 20–30x annualized Q3 run-rate (~$68B). IPO targeted mid-November under ANT.[15]
• OpenAI: $1.25T pre-money in the revised round (post-money higher with new capital).

Bubble commentators used the S-1 to highlight rigid $895B commitments vs. slowing growth (47% QoQ), pay-as-you-go concentration risk, founder control (50.1% voting via LLC), and high multiples needing sustained hypergrowth. They flagged potential margin compression, funding needs post-IPO, and whether demand justifies the bill (e.g., if revenue doesn’t scale to cover ~$100B+ annual run-rate infrastructure).[16][4]

Anti-bubble/bullish voices emphasized the first operating profit, 900% retention, cash generation, improving efficiency, and strategic scale as evidence of durable demand and moat. The non-cash losses were dismissed as accounting artifacts; compute deals were seen as securing advantage ahead of competitors.[17]

Disclosures fueled debate precisely because they quantified both the upside (growth, leverage) and downside (locked costs, execution risk) in one document. For new entrants: matching compute access and proving utilization at scale are now table stakes; valuation discipline will depend on demonstrating sustainable margins beyond the hype cycle.

Filed/Confirmed vs. Leaked/Estimated:
- Filed/Confirmed (Anthropic S-1): All revenue figures, operating results, net losses and components, compute commitment totals and timelines, retention, customer counts, cash position, and governance (e.g., founder control).[3]
- Leaked/Estimated/Press-Reported: Pre-filing revenue rumors (e.g., exact Q3 before amendment), precise IPO raise/valuation targets and timing, OpenAI revenue run-rate and round size/commitments (all from FT, WSJ, CNBC, Bloomberg reporting), specific partner commitment breakdowns in some analyses, and commentator interpretations of multiples or risks.[18]

These public filings and reports set a new benchmark for transparency in private AI, forcing the market to price tangible (if aggressive) economics rather than narrative alone.

---

Recent Findings Supplement (November 2026)

Anthropic filed its initial S-1 on October 26, 2026 (amended November 2 with Q3 results), providing the first detailed public financials; OpenAI’s October 2026 round saw targets cut amid market volatility and competition for capital ahead of Anthropic’s IPO.[1][2][3]

These disclosures (filed numbers confirmed in SEC documents or direct reports from them; others leaked/estimated via investor updates or press) reveal hyper-growth paired with enormous fixed compute obligations, non-cash accounting losses, and slowing momentum—fueling debates on sustainability.

Anthropic S-1: Revenue, Growth, and Profitability (Filed)

Anthropic’s usage-based Claude API business scaled dramatically, with revenue compounding from near-zero to tens of billions in under two years, but quarterly growth decelerated sharply in Q3 while efficiency metrics improved then partially reversed.[1][2]

• Filed revenue (S-1 and amendment): FY2023 $41M → FY2024 $386M → FY2025 $4.585B; H1 2026 $16.286B (up 1,300% YoY from $1.163B H1 2025); Q2 2026 $11.555B; Q3 2026 $16.95–17.0B (up 47% QoQ from Q2, 1,127% YoY); 9 months 2026 $33.2B.[1][2][4]
• Growth details (filed): QoQ growth slowed from 132–144% in Q1/Q2 2026 to 47% in Q3; TTM revenue implied ~$35B+ post-amendment. Net revenue retention ~900% as of Sept 30, 2026 (filed). >1,900 customers paying $1M+ annually.[4][5]
• Margins and losses (filed): Compute/infra costs fell as % of revenue (e.g., 48% in Q2 2026 vs. 251% Q1 2025; ~58% Q3); derived gross margins expanded (e.g., toward 86% in later periods per analyses of filed data). H1 2026 operating loss $693M (Q2 operating profit $974M, first positive); Q3 operating loss $567M. Adjusted operating income positive in profitable quarters (~10–19% margins).[2][6]
• Net losses (filed): H1 2026 $80.7B (mostly $72.4B non-cash fair-value charges on convertible notes/derivatives/SAFEs tied to rising valuation); 9 months 2026 $104B (Q3 net loss $23.29B). Operating cash flow positive in H1 ($3.9B filed).[2][7]

Implication: The filed numbers confirm a usage-driven model achieving operating leverage faster than expected, but pay-as-you-go revenue lacks long-term contracts, making it vulnerable to customer pullbacks (e.g., top customers reducing usage). Competitors must match both scale and efficiency or risk losing share to Anthropic’s data moat.

Anthropic Compute Commitments and Structure (Filed)

The S-1 (especially amended) details massive, largely non-cancelable obligations that dwarf current revenue and cash, highlighting dependency on continued financing and growth.[1][4]

• Filed commitments: $895.4B total purchase/lease obligations (substantially all compute/infra), with ~$69.8–70B due in next 12 months, $447.5–448B by end-2030 (rest to 2045). Earlier draft figures were lower (~$637B–$750B range in reports). Breakdown includes Broadcom ~$161B, Google ~$114.5–121B, SpaceXAI up to $120.5B, Amazon $110B, Microsoft ~$31–45B, plus others (Nscale, Lambda, AMD).[1][8][9]
• Structure notes (filed/related): Many cloud deals payable “regardless of usage”; leases and buildouts front-loaded. Cash/short-term investments ~$112–132B at mid-2026; $15B undrawn revolver. Notes with Google/Amazon tied to compute milestones. No explicit “circular” financing detailed, but vendor-tied notes and capacity deals create interdependencies.[7][10]
• Risk disclosures (filed): Company warns revenue may not materialize to cover obligations; additional risks if funding dries up or partners ration capacity.

Implication: These locked-in costs (far exceeding current ~$35B TTM revenue) require sustained hyper-growth or further raises; new entrants face similar infrastructure barriers but without Anthropic’s scale or partner leverage.

OpenAI October 2026 Funding Round (Leaked/Estimated, Not Filed)

OpenAI (private) pursued a bridge round positioned as pre-IPO capital, but faced pushback, resulting in scaled-back terms amid tech selloffs and Anthropic’s IPO timing.[3][11]

• Target and valuation (reports): Initially pitched ~$30B raise at $1.4T (pre- or post-money references vary); cut to ~$20B target at $1.25T valuation after tepid interest. First close expected < $20B (~$11B committed as of late Oct); prior March 2026 round was $122B at $852B post-money.[3][12][13]
• Investors and structure: UAE consortium (incl. state-backed MGX) leading/largest; Coatue co-lead; others incl. SoftBank, Founders Fund, Sequoia, a16z. Includes option/right of first refusal for OpenAI to lease new UAE compute capacity (vendor-partnership element). Retail/family office follow-on at same valuation. Positioned as “opportunistic bridge” to 2027 IPO.[3][13]
• Revenue context (estimated/leaked): ~$50B annualized run-rate at end-Sept 2026 (some earlier leaks cited ~$70B; distinction between gross partner sales vs. net). Not confirmed in a public filing.[14]

Implication: The round’s compression signals investor caution on AI spend sustainability; vendor compute tie-ins mirror Anthropic’s model but at lower disclosed scale.

Bubble vs. Anti-Bubble Commentary on Disclosures

Commentators split on whether the numbers validate a durable moat or expose bubble risks from rigid costs vs. flexible revenue.[15][16]

• Anti-bubble/bearish views: Emphasize $895B commitments vs. revenue (implying ~$100B+ annual infra bill if spread), slowing growth, pay-as-you-go concentration (top customers ~25% revenue), non-cash losses masking cash needs, and high multiples (20–30x+ annualized revenue at $1.5T+ targets). Risks include funding shortfalls, partner rationing, or growth not scaling to commitments. Some analyses flag valuation compression potential to $500–800B.[6][17][16]
• Bubble/bullish or neutral views: Highlight revenue surge (overtaking peers in some estimates), first operating profits, improving compute efficiency (revenue per compute dollar rising to >2x), 900% NRR, positive cash flow, and TAM expansion via agents/coding. Argue growth trajectory justifies scale despite risks, with filings providing transparency over prior leaks.[5][18]

Implication: Public data arms both sides with specifics (e.g., exact commitment sizes and non-cash loss breakdowns), shifting debate from speculation to quantifiable trade-offs; competitors can use these as benchmarks for their own infrastructure bets.

These updates (primarily Oct 26–Nov 2, 2026 filings and round adjustments) represent the core new disclosures; earlier private estimates (e.g., revenue run-rates) have been partially superseded or refined by filed figures.