Anthropic Revenue and Burn Rate: What the IPO Filing Shows

Yes. Anthropic, PBC publicly filed a Form S-1 with the SEC on October 26, 2026. The filing is on EDGAR, the company announced it the same day, and Reuters, the Wall Street Journal, Bloomberg, the Financial Times, and CNBC all reported it that day. The registration statement is not yet effective, and the share count and price range are still blank.

The filing itself

The primary record is EDGAR accession 0001628280-26-067168, accepted at 09:00:11 Eastern on October 26, 2026, filing date 2026-10-26, Securities Act file number 333-299633, CIK 0002130357. The cover states it was “As filed with the Securities and Exchange Commission on October 26, 2026.” The submission contains 87 documents, including the S-1 HTML and exhibits (charter documents, material agreements, consents, and graphics). [1] [2]

Registrant details on the index and cover match the company’s identity: Anthropic, PBC, a Delaware public benefit corporation, SIC 7372, EIN 86-1696045, principal executive offices at 548 Market Street, PMB 90375, San Francisco. Agents for service are CEO and co-founder Dario Amodei and president and co-founder Daniela Amodei. Counsel includes Wilson Sonsini and Freshfields. [1] [2]

The company published a matching announcement the same day: it filed a Form S-1 for a proposed IPO of Class A common stock, intends to list on the Nasdaq Global Select Market under ANT, and has not yet set the number of shares or the price range. The release names Goldman Sachs and Morgan Stanley as the places to request the preliminary prospectus and states that the registration statement has been filed but is not yet effective. [3]

That is a public flip of an earlier confidential process, not a first contact with the SEC. On June 1, 2026, Anthropic said it had confidentially submitted a draft Form S-1, under Rule 135, and that an IPO would depend on market conditions and completion of SEC review. Share count and price were unset then as well. [4]

Key sections and dates

The prospectus table of contents in the filed S-1 (prospectus pagination, not the EDGAR HTML line count) is:

Section	Prospectus page
Select Defined Terms	1
Prospectus Summary	4
Risk Factors	26
Special Note Regarding Forward-Looking Statements	120
Industry, Market, and Other Data	123
Use of Proceeds	124
Dividend Policy	125
Capitalization	126
Dilution	129
Management’s Discussion and Analysis	132
Business	178
Our Governance Structure and Its Goals	242
Management	254
Executive Compensation	264
Certain Relationships and Related Person Transactions	282
Principal and Selling Stockholders	288
Description of Capital Stock	292
Shares Eligible for Future Sale	307
Material U.S. Federal Income Tax Consequences to Non-U.S. Holders	311
Underwriting	316
Legal Matters; Change in Independent Accountant; Experts	334
Where You Can Find Additional Information	335
Index to Consolidated Financial Statements	F-1

[5]

Risk Factors alone runs from page 26 to just before page 120 — roughly 90-plus pages — which is why secondary coverage treats the risk section as unusually long. The cover also describes a multi-class structure that is not standard dual-class tech: Class A (one vote, economic rights, the listed stock), Class N (generally non-voting, mainly strategic partners, converts on transfer to unaffiliated third parties), Class T (held so the Long-Term Benefit Trust can elect a majority of the board; no economic rights), a single Class F share issued to a Founder LLC of the seven co-founders that carries 50.1% of general voting power and no economic rights, and Class E for employees excluding co-founders with limited voting and no economic rights. The company says it will be a Nasdaq “controlled company.” [2]

Lead bookrunners named on the cover are Goldman Sachs & Co. LLC and Morgan Stanley, with a syndicate that also includes J.P. Morgan, Citigroup, Barclays, Wells Fargo Securities, and others. Offering price, share count, underwriting discount, and proceeds are left blank. The legend is the standard one: securities may not be sold until the registration statement is effective. [2]

Secondary outlets describe the document as roughly 300 to 400 pages, or nearly 240,000 words, depending on whether graphics and the financial-statement appendix are counted. Those length claims are reporting characterizations, not a single figure printed on the EDGAR index. [6] [7] [8]

What the public filing actually discloses — and what it does not

The mechanism of the October 26 event is a public S-1 after confidential review, not a priced IPO. Confidential submission on June 1 started SEC staff review without putting financials on EDGAR. The October 26 filing is the public registration statement. It still does not set terms, and it is not effective, so there is no offering yet.

What major outlets consistently attribute to the public prospectus, not to leaks:

• First-half 2026 revenue of about $16.3 billion, versus about $1.16–$1.2 billion in the year-earlier half (one secondary transcription of the statements puts the half at $16.286 billion). Full-year revenue cited from the filing: about $41 million in 2023, $386 million in 2024, and about $4.6 billion in 2025. [9] [10] [11]
• First-half 2026 net loss of about $80.7 billion, mostly a non-cash charge from revaluing convertible notes or similar instruments as the private valuation rose (reports put that charge above $72 billion). Operating loss for the half narrowed to about $693 million, from about $2.1 billion a year earlier. Reuters also reports adjusted operating income of $1.62 billion for the half, versus a loss of $1.85 billion a year earlier. [9] [7] [10]
• Cash and short-term investments as of June 30, 2026, reported in secondary coverage of the filing as about $112.4 billion in cash and cash equivalents plus about $19.3 billion in short-term investments. [12] [13]
• Customer concentration metrics as of September 30, 2026: more than 11,900 customers at $100,000 or more of trailing-twelve-month revenue, more than 1,900 at $1 million or more, and about 250 at $10 million or more; net revenue retention of about 900%, excluding Claude Pro and Max subscribers. [14] [12] [15]
• Large multi-year compute commitments. Reported figures are not identical across outlets (different cuts of “named deals,” “future obligations,” and “including unpriced capacity”). Examples: The Verge cites about $3.1 billion per month to SpaceX for Nvidia GPU access once planned capacity is delivered, up to $120.5 billion by 2029, plus $44.9 billion with Microsoft through 2033, at least $114.5 billion with Google through 2033, and $110 billion with Amazon over ten years. CNBC describes about $417 billion in future payment obligations for compute, plus additional ordered capacity expected to add $125–$150 billion. Other write-ups cite totals in the $600 billion to $800 billion-plus range depending on what is included. Treat those as secondary readings of contract tables, not as a single number printed in the company’s press release. [16] [14] [7]
• Governance and risk language that is operational, not boilerplate: founder voting control via one Class F share, a Long-Term Benefit Trust that can elect a board majority, and risk factors covering misuse, loss of human control, bioweapons, models that could behave in ways “interpreted as coercive, deceptive, or manipulative,” including behavior resembling blackmail, and the possibility that catastrophic AI harm may be uninsurable. [2] [9] [6]

What is not in the public filing, and should not be treated as disclosed by Anthropic:

• Share count, price range, proceeds, and a target valuation. Those fields are blank. [2] [3]
• A listing date. Reports that the company is aiming for the week of November 16, or before Thanksgiving, are attributed to people familiar with the matter, not to the S-1. The Wall Street Journal and Bloomberg are the outlets most often cited for that calendar. [6] [10] [17]
• A $100 billion raise or a $1.5–$2 trillion valuation. Those figures circulate as investor and media expectations. Reuters separately reported that some investors were marking the company nearer $1.5 trillion, down from about $2 trillion, after a recent investor event — again, not an offering term. The last disclosed private mark widely cited is the May 2026 Series H at a $965 billion post-money valuation after a $65 billion raise. [9] [10] [18]

The real gap is between a public preliminary prospectus and an effective, priced deal. Blank price, blank share count, and the “not yet effective” legend mean the October 26 filing starts the public clock; it does not close an IPO.

How the market got here before October 26

The public S-1 confirmed, rather than invented, a trail of earlier disclosures.

Anthropic’s own June 1 announcement is the primary record of the confidential draft. It created an option to go public after review; it did not commit to a date or size. [4]

Before the public flip, Reuters reported that it had reviewed the confidential prospectus. A September 30 Reuters Q&A described a 2025 net loss of about $42 billion, of which roughly $34 billion was financing-related, leaving an operating loss a little over $8 billion, and a Founder LLC structure under which seven co-founders would hold 50.1% of voting power through Class F shares. Reuters later said the October 26 filing confirmed a series of its earlier exclusives on financials, growth, and IPO plans. Those September figures are leak-based secondary reporting of a then-confidential draft, not the public S-1. [19] [9]

As of early October, at least one market note still described the public S-1 as missing from EDGAR and treated a late-October public filing as the scheduling requirement for a mid-November roadshow, citing a reported marketing start in the week of November 9. That note is now superseded by the October 26 filing. [20]

Source-quality table

Source	Type	What it establishes	Limit
SEC EDGAR index, accession 0001628280-26-067168	Primary filing record	Form S-1 filed and accepted Oct. 26, 2026, 09:00:11 ET; file no. 333-299633; CIK 0002130357; 87 documents	Index does not summarize financials; the S-1 HTML does
S-1 cover and table of contents, anthropicpbc-sx1.htm	Primary prospectus	Filed date, ticker ANT, Nasdaq Global Select Market, blank price and share count, multi-class governance, underwriters, section map through F-1	Preliminary; not effective; many offering numbers blank
Anthropic news post, Oct. 26, 2026	Company primary	Company confirms the public S-1, ticker, and that size and price are unset; states the registration statement is not effective	Does not attach financial exhibits
Anthropic news post, June 1, 2026	Company primary	Confidential draft S-1 submitted that day; IPO optional after SEC review	No financials; not a public registration statement
Reuters, Oct. 26, 2026	Major-outlet secondary	Same-day confirmation of the public filing; H1 revenue growth, $80.7B net loss, narrower operating loss, risk language; says filing confirmed earlier Reuters exclusives	Valuation and timing still sourced to investors or expectations, not the prospectus terms
WSJ, Oct. 26, 2026	Major-outlet secondary	Public paperwork unveiled Monday; path to a listing as soon as the week before Thanksgiving; risk and growth framing	Timing is reported, not a filed date; paywalled detail beyond the lead
Bloomberg, Oct. 26, 2026	Major-outlet secondary	Public filing on Oct. 26; Nasdaq listing expected in November; explainer of the S-1	Deal size and calendar are explanatory/secondary
Financial Times, Oct. 26, 2026	Major-outlet secondary	Prospectus published Monday; $16.3B H1 revenue; ~$81B net loss with >$72B from note revaluation; operating loss $693M; compute commitments described as more than $800B	Rounds some figures; valuation not in the filing
CNBC live updates, Oct. 26, 2026	Major-outlet secondary	Confirms Form S-1, ticker ANT, June confidential filing, ~900% NRR, long risk-factor section, compute obligation ranges	Live blog; some figures are reporter summaries of the prospectus

[1] [2] [5] [3] [4] [9] [6] [17] [7] [14]

What this means if you are underwriting, competing, or marking the stock

The fact pattern is settled on the narrow question: a public S-1 exists, dated and accepted October 26, 2026. Anyone still describing Anthropic as “only confidentially on file” is using a fact that expired that morning.

The investable gap is the blank term sheet. A controlled-company PBC with founder super-voting stock, a trust that can elect a board majority, an ~$80.7 billion accounting loss that is mostly mark-to-model on convertibles, and hundreds of billions of compute obligations is now a public document. The price, the primary versus secondary split, and the week it actually prices are not. Those will show up in an amendment, not in another leak. Until that amendment, reports of a $100 billion raise or a $1.5–$2 trillion valuation are market chatter sitting on top of a preliminary prospectus, not terms Anthropic has filed.

---

Recent Findings Supplement (October 2026)

Yes, Anthropic, PBC publicly filed a Form S-1 registration statement with the SEC on October 26, 2026 (Accession No. 0001628280-26-067168; File No. 333-299633), accepted at approximately 9:00 a.m. ET.[1][2]

This was the public version of a draft confidentially submitted on June 1, 2026, with subsequent amendments (e.g., July, August, September, and early October 2026). The filing prepares for a Nasdaq listing under ticker “ANT” (Class A common stock), with Goldman Sachs and Morgan Stanley as lead underwriters among a group of 22 banks. No share count, price range, or exact raise amount is disclosed yet. Expected timing is around mid-November 2026 (e.g., week of Nov. 16 or before Thanksgiving), subject to change.[3][4]

Key elements from the ~400-page prospectus (consistent across reports):
- Financials (H1 2026, ended June 30): Revenue $16.3 billion (up ~1,300% YoY from ~$1.2 billion in H1 2025; Q2 alone ~$11.6 billion). Operating loss $693 million (improved from $2.1 billion prior year; Q2 showed first operating profit of ~$974 million). Net loss $80.7 billion, driven primarily by non-cash accounting charges on convertible notes/financing (not operational cash burn). Strong cash position reported (~$112 billion cash equivalents + short-term investments as of June 30 in related disclosures).[5][6]
- Business metrics (as of ~Sept. 30, 2026): >11,900 customers with ≥$100k TTM spend; >1,900 with ≥$1M; ~250 with ≥$10M. Net revenue retention (NRR) ~900%.[3]
- Compute/infrastructure commitments: Hundreds of billions over years (non-cancelable elements noted), including ~$110 billion with Amazon (10 years), ≥$114.5 billion with Google (to 2033), $44.9 billion with Microsoft (to 2033), and up to ~$120.5 billion with SpaceX by 2029 (e.g., ~$3.1B/month at full capacity for Nvidia GPUs).[5][7]
- Governance: Seven co-founders (Dario Amodei, Daniela Amodei, et al.) via “Founder LLC” retain 50.1% voting control via Class F shares to preserve AI safety mission.[3]
- Risk factors: Extensive section (~90+ pages) covering AI safety/misuse (e.g., model self-preservation, “blackmail”-like outputs, bioweapons, geopolitical destabilization, loss of human control), execution/financial risks, competition, regulation, and more.[8][9]
- Other: Executive compensation details (e.g., Dario Amodei ~$18M in 2025, mostly equity; Daniela Amodei ~$16.4M). Emphasis on enterprise growth, partnerships (AWS, Google Cloud, Azure), and long-term AI vision (e.g., “revolutionize every area of human endeavor,” potential for curing diseases vs. existential risks).[5]

No dedicated Anthropic press release was prominently reported; coverage centered on the SEC filing itself. Prior leaks (e.g., Reuters on confidential draft in Sept. 2026) aligned closely with public details.[10]

Source-quality table (primary vs. secondary; all post-4/28/2026, focused on Oct 26+ reporting):

Primary (official filings)

- SEC EDGAR S-1 (10/26/2026 public filing; prior confidential DRS June 2026 + amendments): Full prospectus, exhibits, financials, risks. Direct source for all numbers/sections.[1][2]

Secondary (major outlets)

- The Verge, TechCrunch, WSJ, WIRED, CNBC, Reuters, Bloomberg, etc. (10/26–10/27/2026): Consistent reporting on filing date, key financials, compute deals, governance, risks, and IPO timeline. Multiple independent corroborations of core figures (e.g., $16.3B revenue, $80.7B net loss).[5][3][11]

Gap flagged: No company-issued press release or detailed Anthropic statement beyond the filing was identified in coverage. Valuation targets ($1.5–2T range, >$100B raise potential) and exact IPO timing remain analyst/investor estimates, not in the S-1. Earlier confidential leaks provided previews but were not public. The public S-1 closes the prior gap on detailed, verified disclosures.[12]

This represents the most recent development as of Oct 28, 2026.

The S-1 is public. Next I’ll pull the filed quarterly numbers, the company’s own run-rate definition, and how cloud-reseller sales are recognized.
The S-1 is on EDGAR. I’ll extract the revenue tables and the run-rate and marketplace accounting language directly.
Anthropic’s public S-1, filed October 26, 2026, locks in recognized revenue through June 30, 2026. Everything after that — including the third quarter and the company’s own run-rate metric — is either a blank in the filing, a figure told to investors and reported by the press, or an outside estimate. The gap between those measures is the story, not a rounding error.

What the S-1 actually reports

The registration statement (Nasdaq ticker ANT) is the first audited look at Claude’s books. Revenue is recognized gross, including cloud-marketplace sales, and the growth is almost entirely back-loaded: the June 2026 quarter alone was larger than all of 2025.

The summary statement of operations in the filing shows:

Period	Revenue
Year ended Dec. 31, 2023	$41 million
Year ended Dec. 31, 2024	$386 million
Year ended Dec. 31, 2025	$4,585 million
Six months ended June 30, 2025	$1,163 million
Six months ended June 30, 2026	$16,286 million

That is +841% in 2024, +1,088% in 2025, and +1,300% in the first half of 2026. Full years are audited; the six-month periods are unaudited. The same filing leaves the September 2026 dollar figure and “Run-Rate Revenue as of October 31, 2026” blank. [1] [2] [3]

Readers of the prospectus who extracted the quarterly line get a series that adds back to those totals exactly:

Quarter	Recognized revenue	QoQ	YoY	Status
Q4 2024	$197 million	—	—	Reported in S-1 narrative
Q1 2025	$376 million	+91%	—	Reported
Q2 2025	$787 million	+109%	—	Reported
Q3 2025	$1,382 million	+76%	—	Reported
Q4 2025	$2,040 million	+48%	—	Reported
Q1 2026	$4,731 million	+132%	+1,158%	Reported
Q2 2026	$11,555 million	+144%	+1,368%	Reported
Q3 2026	See below	—	—	Dollar amount blank in the S-1

Q1–Q4 2025 sum to $4,585 million. Q1–Q2 2026 sum to $16,286 million. Q2 2026 was the first GAAP operating profit the filing reports, $974 million, after a $1,667 million operating loss in Q1. [4] [5] [6]

The filing says third-quarter 2026 revenue was 86 times fourth-quarter 2024 revenue. At the $197 million base that multiple implies about $16.9 billion. It does not print the dollar amount. [6] [2]

Q3 2026: investor-day number vs. the blank in the prospectus

On October 14, Anthropic told prospective investors that third-quarter revenue was $16.8 billion, up about 45% from the second quarter ($11.6 billion in Reuters’ and CNBC’s telling; the Financial Times used a $11.5 billion base and “nearly 50%”). The company also said it was profitable on an adjusted basis for a second straight quarter. Reuters, the Financial Times, and CNBC all attribute the $16.8 billion figure to people who were in the room, not to the S-1. [7] [8] [9]

The two Q3 readings sit about $100 million apart: $16.8 billion from the investor day, roughly $16.9 billion from 86 × $197 million. Treat $16.8 billion as what management told investors and ~$16.9 billion as arithmetic on a filing disclosure, not as two independent audits.

Adding the investor-day quarter to the filed halves gives about $33.1 billion of recognized revenue for the nine months ended September 30, 2026 (my sum of $4,731 million + $11,555 million + $16.8 billion). That figure is not in the S-1. Q4 2026 has not been reported. Today is October 28, 2026.

Run-rate is a different number, and it ran far ahead of recognized revenue

Anthropic does not report ARR. The S-1 defines Run-Rate Revenue as trailing 28-day consumption fees times 13, plus monthly subscription fees times 12. That is a four-week snapshot, not contracted recurring revenue, and the filing says it can include usage later found to violate policy. The October 31, 2026 figure is blank. [1] [10]

Company statements and credible press, dated:

As of	Run-rate	Label	What it is
Jan 2024	~$87 million	Company-cited	Trajectory Anthropic has described; VentureBeat, May 8, 2026
Dec 2024	~$1 billion	Reported	Later confirmed by the company to CNBC; also The Information
Early Mar 2025	~$1.4 billion	Publicly estimated	The Information, sources
May–Jun 2025	$3 billion	Reported	Company told CNBC; +$1 billion from March to May
Jul 2025	~$4 billion	Press-reported, sources	Bloomberg, via Techmeme’s series
Aug 2025	More than $5 billion	Reported	Company statement, via Reuters
Mid-Oct 2025	Approaching $7 billion	Reported	Company told Reuters, Oct 15, 2025
End 2025	~$9 billion; Amodei said nearly $10 billion	Both in the record	Bloomberg sources: $9 billion+. Amodei, Bloomberg TV, Dec 9, 2025: from $1 billion at the start of the year to $10 billion annualized at the end. Do not collapse these into one number
Feb 12, 2026	$14 billion	Reported	Company blog, Series G
Mar 2026	$19 billion	Press-reported, sources	Bloomberg series
Apr 2026	$30 billion	Reported	“Anthropic says,” VentureBeat, May 8, 2026
Early May 2026	~$44 billion	Publicly estimated	Secondary accounts ahead of the Series H disclosure; not the figure in the company post
Earlier in May 2026	Crossed $47 billion	Reported	Company blog, May 28, 2026, with the $65 billion Series H
End of Jul 2026	More than $65 billion	Company-disclosed to investors	Investor update; Bloomberg first, CNBC confirmed. Not in the S-1
Jul 22, 2026	$74.3 billion	Third-party, inconsistent	Dealroom. Conflicts with the end-July investor figure of $65 billion+. Do not use as a company number
Oct 2026	~$70 billion (Sacra); ~$74 billion (SemiAnalysis)	Publicly estimated	S-1 line is blank. FT said Anthropic did not update the $65 billion July figure at the October investor meeting; SemiAnalysis annualized the new quarterly data to about $74 billion
Year-end 2026	$110 billion	Forward estimate, not a result	Analysts who met executives, via Reuters / WSJ. Not a reported run-rate

[11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [8] [7]

The mechanism that makes run-rate so much larger than recognized revenue is intra-period acceleration. Annualizing a quarter averages a ramp; the 28-day metric annualizes the exit week. That is why Q4 2025 recognized revenue of $2.04 billion annualizes to about $8.2 billion while the year-end run-rate was about $9–10 billion, and why Q2 2026 revenue of $11.555 billion annualizes to $46.2 billion while the end-July investor run-rate was above $65 billion. July sat after the quarter closed, on a still-rising curve. [5] [19]

Q3 broke that pattern. A $16.8 billion quarter annualizes to about $67 billion, only a few billion above the end-July run-rate of $65 billion+, after run-rate had gone from $14 billion in February to $65 billion in July. Sequential recognized growth slowed from +144% in Q2 to about +45% in Q3. Sacra’s October read is roughly $70 billion of annualized revenue, up only about 8% from the July figure. [22] [21] [8]

Gross versus net: the marketplace gross-up is real, and smaller than the headline share

Claude is sold direct and through Amazon Bedrock, Google’s cloud marketplace, and Microsoft Foundry. The S-1 treats Anthropic as principal: it has the customer contract, sets price, and delivers the model. Cloud providers are described as collection agents. They bill the end customer and remit cash net of fees. Anthropic books the full sale as revenue and the fee as an expense, not as contra-revenue. [23] [4] [24]

Marketplace share of reported revenue, from the S-1:

Period	Share through cloud marketplaces
2023	~11%
2024	~32%
2025	~47%
H1 2025	~44%
H1 2026	~48%

On $16.286 billion of first-half 2026 revenue, 48% is about $7.8 billion booked through Amazon, Google, and Microsoft. [23] [25]

The fee, not the 48%, is the gross-versus-net difference. Reuters’ read of the confidential draft put 2025 marketplace sales at about $2.16 billion — consistent with 47% of $4.585 billion — and distribution fees at about $351 million, roughly 16 cents per marketplace dollar and about 8% of total 2025 revenue. Anthropic told Reuters it follows principal accounting. OpenAI has told investors the gross method inflates Anthropic’s reported revenue by billions; that claim and the $351 million 2025 fee can both be in the record without being the same calculation. [24] [26]

The public S-1, as read by Blockspace, does not give a first-half 2026 commission dollar figure. One secondary note (Dealroom) put distribution fees at about 17% of first-half revenue. That would be roughly $2.8 billion, far above 16% of the marketplace slice. I would not use it. The clean inference from the 2025 fee rate is that netting marketplace fees would cut reported revenue by high single digits, not by the 48% channel share.

Mix inside the gross number, also from the filing: consumption (token) revenue was $288 million in 2024 (75%), $3.8 billion in 2025 (83%), and about $13.1 billion in the first half of 2026 (80%). Subscriptions were $98 million, $789 million, and about $3.2 billion over those same periods. [27] [28] [29]

How to read the three definitions against each other

Recognized revenue is what the income statement will show once a quarter closes. It is gross of cloud fees, usage-heavy, and not contracted. Remaining performance obligations were only $10.8 billion against a Q2 revenue base that annualizes to $46.2 billion, so backlog covers well under a quarter of the exit pace. [5]

Run-Rate Revenue, as Anthropic defines it, multiplies the last 28 days. In a business that was adding several billion dollars of run-rate per month from February through July 2026, that metric systematically overshoots the quarter you just closed and the year you will actually recognize. It is the right metric for “what is the exit pace,” and the wrong one for “what did they earn.”

A third, informal metric — last quarter times four — sits between them. Q2 × 4 is $46.2 billion. End-July run-rate was above $65 billion. Q3 × 4 is about $67 billion, which finally caught the July run-rate because sequential growth slowed. Anyone comparing Anthropic to OpenAI, or to a software ARR multiple, has to say which of the three they are using. Bloomberg has already noted the two labs may not measure run-rate the same way. [30]

Internal targets from late 2025 are now a footnote, not a forecast. In October 2025 the company was aiming for a $9 billion run-rate by year-end 2025 (it hit that) and a 2026 base case above $20 billion with a best case of $26 billion. Recognized revenue in the first half of 2026 alone was $16.3 billion, and the nine-month sum is already above $30 billion. The Information’s January 2026 report that Anthropic had raised its 2026 sales forecast to as much as $18 billion was already stale by the time Q1 closed at $4.7 billion. [13] [31]

What this means if you are underwriting the IPO or competing with Claude

The comparable that matters is no longer “AI startup ARR.” It is whether the Q3 slowdown is a pause in a supply-constrained ramp or the start of a lower slope. The filing itself says revenue should not be expected to keep growing at these rates. Net revenue retention of about 900% as of September 30 is measured on a three-month-versus-year-ago-three-month basis for qualified customers, not the trailing-twelve-month SaaS definition, so it describes a base that was tiny a year earlier, not a steady-state expansion rate. [6] [10]

For a competitor, the economic fact is the gross booking, not the fee. Nearly half of first-half revenue was billed by Amazon, Google, or Microsoft, who are also investors, compute suppliers, and model competitors. Losing marketplace placement would hit both distribution and the receivables balance — marketplace sales were 58% of net receivables at June 30. [25]

For valuation math, do not put a revenue multiple on the July $65 billion run-rate and call it 2026 sales. Recognized 2025 revenue was $4.6 billion. Recognized first-half 2026 revenue was $16.3 billion. Nine-month recognized revenue, using the investor-day third quarter, is about $33 billion. A full-year 2026 number still depends on a fourth quarter that has not been reported, and on whether October’s model releases re-accelerated the 28-day metric the S-1 has not yet filled in. The next amended filing is the document that replaces the blanks for third-quarter revenue and October 31 run-rate. Until then, $16.8 billion and “above $65 billion” are management disclosures to investors, and ~$70–74 billion is other people’s annualization.

---

Recent Findings Supplement (October 2026)

Anthropic filed its S-1 on October 26, 2026 (Nasdaq: ANT planned), providing the first detailed public financials. This filing, combined with an October 14, 2026 investor presentation and contemporaneous reporting, supplies the core recent data (all post-April 28, 2026). Key themes include explosive but decelerating growth (still triple-digit YoY in Q2 2026 but ~45% QoQ in Q3), improving efficiency (revenue per compute dollar rising sharply), massive compute commitments, and a shift toward diversified customers.[1][2]

Reported quarterly and half-year revenue (S-1 primary source) shows H1 2026 already exceeding full-year 2025 by >3.5x. Q3 2026 revenue comes from the investor presentation (not dollarized in the initial S-1).[3][4]

Run-rate/annualized figures are estimates or derived; the S-1 defines a specific "Run-Rate Revenue" metric but left the October 31, 2026 value blank in the initial filing. Earlier run-rates (e.g., $65B end-July) come from investor updates and analysis.[5][6]

Cloud marketplace sales (AWS, Google Cloud, Azure) accounted for ~48% of H1 2026 revenue (~$7.8B). Anthropic recognizes these on a gross basis (as principal, with direct customer contracts and pricing control) and records CSP fees in sales/marketing expenses (not COGS). This inflates reported revenue relative to a net-of-fees treatment; fees were previously estimated in the hundreds of millions annually.[7][8]

Most revenue (~80% in H1 2026) is consumption-based (per-token usage); ~20% subscription-based. Net revenue retention reached ~900% as of September 30, 2026 (S-1; excludes Pro/Max plans). Top-5 customers ~25% of first-nine-months 2026 revenue; no single customer >10%.[2][9]

Dated Revenue Table (2024–Q3 2026)

All figures in USD billions unless noted. "Reported" = directly from S-1 or company statements; "Estimated/Publicly estimated" = derived from multiples, investor presentations, or analyst synthesis. Sources primarily October 2026 S-1 summaries and contemporaneous reporting.

• FY 2024: $0.386B (reported, S-1).[1]
• FY 2025: $4.585B–$4.59B (reported, S-1); grew ~11.9x from 2024.[1][10]
• H1 2025: $1.163B (reported, S-1).[2]
• Q1 2026: ~$4.731B (reported, S-1/MD&A).[11]
• Q2 2026: $11.555B–$11.6B (reported, S-1); ~144% QoQ, 1,368% YoY.[1][12]
• H1 2026: $16.286B–$16.3B (reported, S-1); ~14x YoY.[2][9]
• Q3 2026 (Jul–Sep): $16.8B (reported to investors; S-1 states ~86x Q4 2024’s $0.197B ≈ $16.9B range). ~45–46% QoQ from Q2.[3][13]

Quarterly growth slowed markedly in Q3 2026 after prior quarters of 100%+ sequential gains.[3]

Annualized Run-Rate / ARR Figures

• End-2025: ~$9B (publicly estimated from prior updates).[6]
• Late July 2026: >$65B (publicly estimated/run-rate from investor updates).[5][14]
• October 2026 estimates: ~$70B–$74B (publicly estimated; e.g., SemiAnalysis ~$74B based on Q3 data).[3][15]
• End-2026 projections (various): Below or around $100B (publicly estimated; growth slowing).[3][15]

S-1 Run-Rate Revenue definition (distinct from traditional ARR due to usage-based model): Trailing 28-day consumption fees × 13 + monthly subscriptions × 12. Value as of Oct 31, 2026 left blank in initial filing.[2][16]

Key Context and Definitional Notes

• Recognized revenue vs. run-rate: S-1 figures are recognized revenue (primarily usage-based, recognized as consumed). Run-rate annualizes recent activity for forward-looking scale but is not contracted recurring revenue.
• Gross vs. net cloud-reseller treatment: ~48% of H1 2026 revenue via marketplaces is booked gross by Anthropic (principal accounting); CSP fees (~16% effective rate in one prior-year estimate) hit operating expenses. This boosts top-line vs. net recognition peers.[17]
• Other notes: 80% consumption-based in H1 2026; improving unit economics (revenue per $1 compute rose from ~$0.40 in Q1 2025 to $2.08 in Q2 2026). Massive long-term compute commitments ($600B+ range cited across reports).[12][18]

These October 2026 disclosures represent the most material new public data; earlier 2026 estimates have been superseded or aligned with S-1 numbers. Growth remains exceptional but has moderated, with efficiency gains offsetting some cost pressures. For competition or valuation analysis, the gross marketplace accounting and usage-based concentration are critical differentiators versus traditional software metrics.

Anthropic’s October 26, 2026 S-1 does not split revenue the way outside observers usually do. It books one operating segment and two pricing lines—token consumption versus subscriptions—then overlays a cloud-marketplace channel. API (direct, Bedrock, Vertex, Azure/Foundry), enterprise minimums, Claude Code, and Pro/Max/Team all sit inside those two lines and are not disclosed as products.

What the S-1 actually reports

The filing’s economic unit is the token, not the product. Consumption is “metered in tokens and priced according to the model selected.” Subscriptions are prepaid monthly or annually, usage-capped, and recognized ratably. Enterprise customers often sign annual or multi-year minimums, but revenue is still recognized when tokens are used, so a “contract” is not a separate GAAP bucket. Cloud service providers (CSPs) bill and collect, then remit net of fees, while Anthropic keeps the end-customer contract and books the sale gross.

For the six months ended June 30, 2026, revenue was $16.286 billion, up from $1.163 billion a year earlier. Consumption was $13.064 billion (80%) and subscriptions $3.222 billion (20%). Full-year 2025 was $4.585 billion, of which consumption was about $3.80 billion (83%) and subscriptions $789 million (17%). [1] [2] [3]

S-1 line	FY2024	FY2025	H1 2025	H1 2026	Share of H1 2026	Confidence
Total revenue	$386M	$4,585M	$1,163M	$16,286M	100%	High — filing, multiple outlets
Consumption-based	$288M (75%)	~$3.80B (83%)	$983M	$13,064M	80%	High
Subscription-based	$98M (25%)	$789M (17%)	$180M	$3,222M	20%	High
Through CSP marketplaces	~32%	~47% ($2.16B via Amazon + Google)	~44%	~48% (~$7.8B)	48%	High on share; medium on the $7.8B (48% × $16.286B)
Direct / non-marketplace (residual)	—	—	—	~52% (~$8.5B)	52%	Medium — arithmetic, not a labeled line
US billing address	—	—	$820M	$11,763M	72%	High
International billing address	—	—	$343M	$4,523M	28%	High

Sources: Form S-1 filed October 26, 2026, as reported by FourWeekMBA, Kepler, Reuters (draft), Blockspace, and TokenPost. [1] [4] [5] [6] [7]

Quarterly recognized revenue, compiled from the same filing coverage: Q1 2025 $376 million, Q2 $787 million, Q3 $1.382 billion, Q4 $2.040 billion, Q1 2026 $4.731 billion, Q2 2026 $11.555 billion. [8] [4] Q3 2026 was disclosed to investors at about $16.8 billion (Bloomberg, Financial Times, Reuters), up roughly 45% sequentially. The S-1 instead states Q3 was 86 times Q4 2024; one reconstruction using a $197 million Q4 2024 base gets about $16.9 billion. Treat $16.8–16.9 billion as the same quarter reported two ways, not two different numbers. [9] [10]

What this means if you are mapping channels: Bedrock, Vertex, and Microsoft Foundry/Azure are a distribution overlay on consumption, not a product segment. Direct API, Claude Code token burn, and enterprise overages are also consumption. Pro, Max, Team, and at least some Enterprise seat fees are the subscription line. You cannot add “API + enterprise + Claude Code + consumer” and expect 100% without double-counting.

Where the requested categories diverge from the S-1

The prospectus has one segment, coordinated by the co-founders as CODM. It does not report API versus applications, direct versus Bedrock versus Vertex versus Azure, or Pro versus Max versus Team versus Enterprise. [3] [11]

The mismatches that matter:

• API (direct and via clouds) is not a line. Marketplace sales were about 11% of revenue in 2023, 32% in 2024, 47% in 2025, and 48% in H1 2026. The 2025 $2.16 billion figure Reuters saw in the confidential draft was Amazon and Google only. The public H1 2026 48% includes Amazon, Google, and Microsoft, with no provider split. [1] [6] [7] Direct access is the residual ~52%, but that residual mixes the first-party API, claude.ai, Claude Code, Cowork, and direct enterprise—not “direct API” alone.
• Enterprise contracts overlap consumption. Minimum-spend deals convert usage into a commitment, then revenue is still recognized on tokens. The closest disclosed proxy is customer size, not contract type: customers with $100,000 or more of trailing-twelve-month revenue produced $12.0 billion, about 74% of H1 2026 revenue (more than 6,000 such customers at June 30; more than 11,900 by September 30). [2] [12]
• Consumer subscriptions are inside the 20%, not equal to it. Subscription revenue includes Claude Pro and Max plus Team and Enterprise packages. The filing does not isolate individuals. Net revenue retention of about 900% as of September 30, 2026 explicitly excludes Pro and Max, because fixed fees do not expand the way usage does. Consumer is therefore less than 20% of total revenue; how much less is not disclosed. Secondary write-ups disagree on whether “much” of the $3.2 billion is Pro/Max or whether individuals are a small slice of that 20%. [1] [13] [14]
• Claude Code is a usage multiplier, not a segment. The S-1 calls it the product that “most concretely defined what autonomous AI work looks like,” ties ~900% net revenue retention to agentic products like Claude Code, and gives a spend gap rather than a dollar total: in the quarter ended September 30, 2026, Claude Enterprise users who did not use Claude Code spent less than $75 a month on average, while the top 5% of Claude Enterprise users who did use it spent about 20 times that. Over half of $100,000-plus customers used Claude Platform plus at least one first-party app (Claude Code or Cowork) in Q3 2026, up from about 30% in Q4 2025. [1] [2] [15]

Gross accounting also inflates the top line relative to a net reseller model. Platforms bill the end customer and remit net of “platform and other fees.” Reuters’ analysis of the 2025 draft put distribution fees above $350 million on $2.16 billion of Amazon and Google marketplace sales, roughly 16 cents per dollar—well above published marketplace listing rates of 1.5–3%, because the fee bucket is broader than a store commission. [6] [5]

Estimated split into the categories people actually want

These shares are reconstructions. Anthropic has not published them. They do not add cleanly to the S-1 lines, because Claude Code and enterprise usage are mostly consumption, and Bedrock/Vertex/Azure are a channel through that consumption.

Category asked for	Best available figure	Period	What it actually measures	Confidence
API + enterprise usage (combined)	~80% of revenue; $13.1B consumption in H1 2026	H1 2026	S-1 consumption line. Includes direct API, cloud-marketplace API, Claude Code tokens, and enterprise usage. Not “API only.”	High as a pricing split; low as a product split
Cloud marketplaces (Bedrock + Vertex + Azure/Foundry, combined)	~48%, ~$7.8B	H1 2026	Channel, mostly consumption. No AWS vs Google vs Microsoft split.	High on combined share
Amazon + Google marketplaces only	$2.16B, 47%	FY2025	Confidential draft, Reuters. Microsoft not in this pair.	High for 2025; not comparable 1:1 to the 2026 three-cloud figure
Direct / non-marketplace	~52%, ~$8.5B	H1 2026	Residual. Includes direct API and first-party apps.	Medium
Subscriptions (Pro, Max, Team, and enterprise seats)	$3.222B, 20%	H1 2026	Upper bound on consumer + Team. Enterprise seats are inside this number.	High on the total; low on the consumer slice
Consumer Pro/Max alone	Not disclosed. Must be <20% of total. One synthesized estimate puts Pro/Max at ~8% and Team at ~2%.	2026	Value Add VC, explicitly not an audited segment. An earlier chart from the same outlet had consumer nearer 22%—do not use that.	Low
Enterprise as large accounts, not “contracts”	$12.0B, ~74%	H1 2026	Customers with ≥$100k TTM revenue. Overlaps API, Claude Code, and seats.	High as a size cut; not a contract cut
Claude Code	Company: >$0.5B run-rate (Sept 2025), $1B (Dec 2025), >$2.5B (Feb 12, 2026), enterprise > half of Claude Code revenue	Company disclosures	Run-rate, not GAAP revenue. Feb 2026 figure was ~18% of the company’s then ~$14B run rate.	High on company dates through Feb 2026
Claude Code	~$8B annualized	May 2026	Sacra / FutureSearch, alongside company ARR of $47B (~17%). Not in the S-1.	Medium
Claude Code	$15.12B tracked ARR; ~21.9% of tracked company ARR	Week ending Aug 10, 2026	TickerTrends alt-data. Anthropic has not confirmed. Growth had slowed to ~5% over four weeks.	Low–medium
Synthesized channel mix	API 45%, enterprise contracts 35%, Claude Code 10%, Pro/Max 8%, Team 2%; clouds “included in API”	2026 estimate, updated Oct 26	Value Add VC. Conflicts with the $8–15B Claude Code run-rate estimates above, which imply a higher Code share of ARR than 10% of recognized revenue.	Low

Company Claude Code milestones: general availability in May 2025; over $500 million run-rate by the September 2, 2025 Series F announcement, with usage up more than 10x in three months; $1 billion by December 3, 2025; over $2.5 billion on February 12, 2026, more than doubled since January 1, with weekly active users doubled and business subscriptions up 4x, and enterprise more than half of Claude Code revenue. [16] [17] May and August figures are third-party. [18] [19] [20]

The mechanism is straightforward and is why a seat count understates the business. A chat user on a capped plan produces a fixed fee. An agent that reads a repo, edits files, runs tests, and retries burns orders of magnitude more output tokens—the expensive side of Anthropic’s price card. That is why the S-1 can show Enterprise users who skip Claude Code at under $75 a month while the heaviest Code users inside Enterprise spend about 20 times that, and why net revenue retention, excluding Pro and Max, was about 900%. [1] [21]

Company-level run rate, for scale: about $9 billion at the end of 2025, $14 billion in February 2026, over $47 billion at the late-May Series H, $65 billion at the end of July (figures shared with investors; CNBC confirmed the July figure), with SemiAnalysis around $74 billion by mid-October and Sacra around $70 billion in October after Q3 growth slowed. Recognized H1 revenue of $16.3 billion is not the same object as a 28-day consumption run rate. [18] [17] [10]

Customer concentration: Cursor, Copilot, and the top five

Concentration was real in 2025 and has diluted in share even as the largest accounts kept setting revenue records.

Disclosed, unnamed:

• Top two customers were an aggregate of about 24% of 2025 revenue. Reuters, which saw the confidential draft, reported each at about 12%. On $4.585 billion, that is roughly $550 million each, about $1.1 billion combined. [6] [22]
• For the nine months ended September 30, 2026, no single customer was more than 10%, and the top five were about 25%. Outside the top five: about 64% of Q1 2026 revenue and about 79% of Q3, so the top five fell from about 36% of the quarter to about 21%, even though revenue from each of the top five hit a record in Q3. [23] [9]
• A draft circulated before the public filing, reported by the Wall Street Journal, put about one-fifth of H1 2026 revenue in two customers. The public nine-month language is the one to use for the later period. [24] [25]
• Many large customers are not on long-term locks and can cut usage immediately. That is the risk factor the filing emphasizes, not a named logo. [23]

Named only by inference, not by the S-1:

• In August 2025, VentureBeat reported that Cursor and GitHub Copilot drove about $1.2 billion of a roughly $4–5 billion run rate, close to a quarter. That lines up in magnitude with the later 24% / two-customer 2025 disclosure, but a mid-year run rate is not full-year GAAP, and Anthropic has never confirmed the names. [26] [27]
• Motley Fool and others have speculated that the H1 2026 pair is SpaceX (which acquired Cursor for about $60 billion in August 2026) and Microsoft via GitHub Copilot. That is industry inference, not a filing fact. A claim that the official prospectus identifies the two customers as Amazon and Google confuses the marketplace channel (47–48% of sales, with Anthropic still contracting the end customer) with the 10% customer test. Reuters treated those as separate disclosures. Do not treat Amazon and Google as the named 12% customers. [28] [6] [11]
• Sacra’s post-filing note says top customers including Meta and Microsoft cut internal Claude Code usage in Q3 (about 50% and 35%, respectively) as part of a broader slowdown. That is Sacra’s reporting, not an S-1 table. [18]

Other concentration the filing does quantify: government contracts are about 1% of revenue. About 5–7% of revenue in the first nine months of 2026 came from usage that violated policy, including model distillation, and Anthropic warned that cracking down could reduce revenue. [29] [30] [31]

The base under the top five has widened fast. $100,000-plus customers went from about 20 at the end of 2023 (about 90% of revenue) to about 200 in 2024, about 1,500 in 2025 (about 76% of revenue, $3.5 billion), more than 6,000 at June 30, 2026 ($12.0 billion, 74% of H1), and more than 11,900 at September 30, including more than 1,900 at $1 million-plus and about 250 at $10 million-plus. The long tail is about 40 million customers each under $100,000. The $100,000-plus group includes every Fortune 10 company and more than half the Fortune 500, but Fortune 500 spend in that tier averaged less than 0.1% of those companies’ total expenses in Q3—penetration, not saturation. [32] [2] [14]

Claude Code’s growth, and why it is not a clean revenue line

Claude Code is the clearest product-level ramp Anthropic has disclosed, and then stopped disclosing in dollars.

Date	Claude Code figure	Source	Confidence
May 2025	General availability	Anthropic	High
Sept 2, 2025	>$500M run-rate; usage >10x in three months	Anthropic	High
Dec 3, 2025	$1B run-rate, ~6 months after GA	Anthropic	High
Feb 12, 2026	>$2.5B run-rate; WAUs doubled since Jan 1; business subs 4x; enterprise >50% of Code revenue	Anthropic	High
May 2026	~$8B annualized	Sacra, FutureSearch	Medium
Aug 10, 2026	$15.12B tracked ARR; ~5% growth vs four weeks earlier; ~21.9% of tracked company ARR	TickerTrends	Low–medium
May 2027	FutureSearch p50 $18.5B (p10 $9.8B, p90 $33B)	Forecast, not a disclosure	Low

The strategic point is that Code competes with Anthropic’s own largest historical customers. Cursor and Copilot made Claude the default model inside someone else’s product and, on the 2025 evidence, a quarter of the company. Claude Code takes the same workload direct: the developer (or the enterprise that standardizes on it) pays Anthropic for the agent, not only for tokens resold through a third-party IDE. That is why Code can both add revenue and pressure the Cursor/Copilot share at the same time. SpaceX’s August 2026 acquisition of Cursor, and Anthropic’s public commitment to keep serving Claude inside Cursor after OpenAI moved to cut Cursor off, shows the indirect channel is still strategically important even if its share of a much larger base has fallen below the 10% naming threshold. [33] [28]

Code is also the wedge into the rest of the suite. Cowork, the knowledge-work agent previewed in January 2026, had by September 30 outpaced Claude Code’s adoption at the same point after Code’s February 2025 preview, and Anthropic began merging Cowork into Claude.ai in September 2026. Multi-product attach among $100,000-plus customers rose from about 30% to over 50% in three quarters. The S-1’s growth story is not “more chatbot seats.” It is “a coding agent creates a heavy user, who then pulls platform spend and a second app.” [1] [15]

How to use this if you are competing or underwriting

Do not build a model that treats API, enterprise, and Claude Code as additive S-1 segments. Start with 80/20 consumption versus subscription, haircut marketplace revenue for fees if you care about net dollars (2025 implied about 16% of Amazon/Google marketplace sales, and that rate is not disclosed for 2026), and treat Claude Code as a driver inside consumption whose dollar share is somewhere between the company’s last official print (>$2.5 billion run-rate in February, then about 18% of company run rate) and alt-data around 20% of tracked ARR by August—not a precise 10% or 17%.

The concentration risk has changed shape. In 2025, two customers were 24%, and the leading external hypothesis is still Cursor and GitHub Copilot. By the first nine months of 2026, no one customer was above 10% and the top five were 25%, with their quarterly share falling from about 36% to about 21%. That is diversification by explosion of the denominator, not proof those logos got smaller in dollars—the filing says each of the top five set a Q3 revenue record. The binding risk is usage-based contracts with no lock-in against multi-year compute commitments, not a single IDE vendor at 12%.

For a competitor, the opening is not the consumer plan. Pro and Max are excluded from the retention metric precisely because they do not expand, and they are a subset of a 20% subscription line. The opening is either (1) being the default model inside someone else’s agent, which is how Cursor and Copilot became outsized in 2025, or (2) owning the agent surface so token burn accrues directly, which is what Claude Code did from a standing start in May 2025 to a multi-billion run rate by early 2026. Cloud marketplaces are a distribution tax and a distribution advantage at once: nearly half of H1 2026 revenue, no disclosed split among AWS, Google, and Microsoft, and the same three companies are also compute suppliers.

---

Recent Findings Supplement (October 2026)

Anthropic filed its S-1 on October 26, 2026, providing the first detailed public financials (all data below from or directly derived from this filing and contemporaneous analyses unless noted). H1 2026 revenue reached $16.286 billion (up ~14x from $1.163 billion in H1 2025), with Q2 alone at $11.555 billion. The filing segments primarily by consumption-based (usage/API tokens, 80% or $13.064 billion) versus subscription-based (20% or $3.222 billion), plus channel (cloud marketplaces ~48%). It does not break out the exact requested categories (direct API vs. Bedrock/Vertex, specific consumer plans, or Claude Code standalone).[1][2]

Claude Code (Anthropic’s agentic coding product) is a major growth and expansion driver within consumption revenue. Enterprise users of Claude Code spend ~20x more on average than non-users (<$75/month vs. ~$1,500/month for top 5% of Claude Enterprise users). This usage fuels the ~900% net revenue retention (NRR) as of September 30, 2026 (excludes flat-fee Pro/Max subscribers). Third-party estimates place Claude Code run-rate revenue at >$2.5 billion by February 2026 and ~$8 billion annualized by May 2026.[1][3]

Customer concentration is notable but diversifying, with heavy reliance on large usage-based accounts including coding platforms. More than 11,900 customers generated ≥$100k TTM revenue as of September 30, 2026 (up from >950 a year earlier); these accounted for ~74% of H1 2026 revenue ($12 billion). Top 5 customers represented ~25% of revenue over the nine months to September 30, 2026 (no single customer >10%). Reports indicate two customers (widely speculated as Cursor and GitHub Copilot) accounted for ~20% ($3.26 billion) of H1 2026 revenue. Cursor reached ~$4 billion annualized revenue by June 2026; GitHub Copilot had ~4.7 million paid users. Over 50% of $100k+ customers now use both the Claude Platform and at least one first-party app (up from ~30% in Q4 2025).[4][5]

Cloud marketplaces (AWS Bedrock, Google Vertex AI, Microsoft Azure) handled ~48% of H1 2026 revenue (~$7.8 billion), up from ~11% in 2023 and ~47% in 2025. Anthropic recognizes this gross (as principal, with direct customer contracts) while CSPs handle billing/collection and take fees. Direct sales are the remaining ~52%. International billing was 28% of H1 revenue.[6][1]

S-1 segments differ materially from requested categories. The filing uses only consumption (primarily API/usage, including via Claude Code and other agents) versus subscription, plus the marketplace channel. It provides no product-level splits (e.g., exact Claude Code contribution, Pro/Max/Team/Enterprise within subscriptions, or direct API vs. Bedrock/Vertex within consumption or marketplaces). Consumer subscriptions form an unspecified portion of the 20% subscription bucket.[7]

Table of estimates (H1 2026 or latest available; all post-April 2026 data prioritized):

• API (direct + via AWS Bedrock/Google Vertex/Microsoft Azure): ~$13.1B consumption-based (80% of total); of which marketplaces ~48% of total revenue (~$7.8B). Direct ~52%. Source: S-1; confidence high (explicit aggregates).[1]
• Enterprise contracts: Mix of consumption (majority) and subscriptions; >11,900 $100k+ customers drove ~74% of H1 revenue. Top 5 ~25%. Source: S-1; confidence high.[4]
• Claude consumer subscriptions (Pro, Max, Team): Unspecified portion of $3.2B subscription-based (20%). Pro/Max excluded from NRR metric. Source: S-1 inferences; confidence medium-low (no split).[1]
• Coding products (Claude Code): Key driver of consumption/expansion; users spend ~20x more; third-party est. ~$8B ARR (May 2026). Source: S-1 context + Sacra analysis; confidence medium (growth/impact explicit, dollar est. third-party).[3]
• Customer concentration (e.g., Cursor/GitHub Copilot reliance): Top 2 plausibly ~20% ($3.26B) of H1; Cursor ~$4B ARR (June 2026 est.); Copilot ~4.7M paid users. Top 5 ~25%. Source: S-1 + reports (Stahel/Motley Fool); confidence medium (speculation on exact names, but concentration confirmed).[5][8]

Overall run-rate context (third-party, post-S-1 analyses): Annualized revenue reached ~$65B by July 2026, slowing to ~$70B by October amid Q3 growth moderation (top customers reportedly reduced Claude Code usage). Claude Code and agents drove much of the prior hyper-growth and efficiency gains (compute fell to 48% of revenue; first GAAP operating profit in Q2).[3]

These figures reflect explosive but usage-dependent growth with channel and customer risks (e.g., marketplace dependence, variable consumption). No finer public breakdowns exist beyond the S-1 aggregates as of late October 2026.

Anthropic’s public S-1 (filed with the SEC on October 26, 2026) is the first audited view of a frontier lab’s books. The operating business is close to break-even and already cash-generative. The GAAP net loss is almost entirely a mark-to-market on financing instruments whose value rose with Anthropic’s own valuation. The binding constraint is not current cash burn; it is multi-year take-or-pay compute that will hit the income statement only as capacity is delivered.

1. Gross margin: an 85% functional margin that is not an inference margin

The S-1 does not publish a “gross margin.” What it does publish, in the functional note, is cost of revenue of $2,380 million on H1 2026 revenue of $16,286 million — about 85% — versus 12% in H1 2025 ($1,019 million of cost of revenue on $1,163 million of revenue) and about 40% for full-year 2025 ($2,773 million on $4,585 million). Full-year 2024 was negative, about −95%.

That 85% is an accounting allocation, not a pure inference margin. The filing says “substantially all” of cost of revenue is compute and infrastructure for customer-serving workloads, plus some personnel and overhead. Training, research, internal Claude use, and free-tier serving sit elsewhere. Total compute and infrastructure in the same half was $8,945 million, so at least about $6.6 billion of the compute bill — roughly three-quarters — is outside cost of revenue. Research and development was $9,741 million in the half and “primarily” consists of compute for model development, research, and internal usage. Sales and marketing of $3,808 million includes compute for free access.

Two further distortions matter for anyone comparing labs:

• Marketplace revenue is booked gross. About 48% of H1 2026 revenue came through AWS, Google Cloud, and Azure. Channel-partner fees sit in sales and marketing, not as a contra-revenue or cost-of-revenue item. Secondary readings of the filing put those fees at roughly 17% of H1 revenue (on the order of $2.7 billion of the S&M increase). A “gross margin” that ignores a 25–30% reseller cut on nearly half of sales overstates unit economics relative to a direct-sold software company.
• Consumption vs. subscription mix is filed: $13.1 billion (80%) usage-based, $3.2 billion (20%) subscriptions in H1 2026. Subscriptions are the less efficient slice of the inference pool.

Filed vs. estimated. The 85% / 40% / 12% figures are arithmetic on filed cost of revenue, not a company metric. SemiAnalysis estimates, published before and around the filing, are a different construct: blended inference gross margin in the mid-60s by mid-2026 (from −94% in 2024 and about 38% in 2025), API above 80% and potentially as high as 88%, with subscriptions about 10% of revenue but more than 40% of inference compute and still roughly 50% compute margin under realistic utilization. Those are outside estimates, not S-1 lines. Post-filing, SemiAnalysis has also described the filed customer-serving margin as about 85% and incremental economics as roughly 90% gross / 40% EBIT — commentary on the filing, not a second set of audited numbers.

What this means for a competitor. You cannot underwrite Anthropic’s token economics off the 85% alone. The margin expands when the shared compute pool is pointed at paying inference and compresses when the same pool is pointed at the next training run. The filing itself warns that when large new capacity comes online, operating margin “may compress… even if the underlying economics of the business continue to improve.”

2. Inference and training compute spend: one pool, two expense lines

Anthropic runs AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs as one fungible pool and directs it among research, model development, customers, and internal use. Expense is recognized when capacity becomes available, not when the contract is signed, and most of it is contracted 12–18 months ahead. The company does not disclose a dollar split of compute between training and inference beyond the functional allocation above.

Filed compute and infrastructure (nature presentation):

Period	Revenue	Compute & infrastructure	Compute / revenue
FY2023	$41M	$433M	1,056%
FY2024	$386M	$2,522M	653%
FY2025	$4,585M	$7,326M	160%
H1 2025	$1,163M	$2,334M	201%
H1 2026	$16,286M	$8,945M	55%
Q1 2025	$376M	$944M	251%
Q2 2025	$787M	$1,390M	177%
Q3 2025	$1,382M	$2,047M	148%
Q4 2025	$2,040M	$2,945M	144%
Q1 2026	$4,731M	$3,387M	72%
Q2 2026	$11,555M	$5,558M	48%

Revenue per compute dollar went from about $0.40 in Q1 2025 to about $2.08 in Q2 2026. H1 2026 as a whole was about $1.82 of revenue per compute dollar, versus about $0.50 a year earlier and about $0.63 for all of 2025. Q2 2026 revenue grew 1,368% year over year while compute grew about 300%.

The best filed proxy for “inference compute for paying customers” is the cost-of-revenue ceiling of $2.4 billion in H1 2026. The best filed proxy for “training plus research plus internal plus free tier” is the residual of at least $6.6 billion, most of it inside the $9.7 billion R&D line. There is no filed number that isolates pre-training or post-training spend from serving research traffic.

Capex is small because the model is rented, not owned: $482 million of property and equipment purchases in H1 2026, against $8.9 billion of compute expense. Owned PP&E, net, was about $1.2 billion at June 30, 2026.

3. Cloud and chip commitments: floors, ceilings, and estimates that do not add cleanly

These are contractual obligations, mostly not yet in the income statement. The S-1 uses several non-comparable buckets. Adding them double-counts.

Balance-sheet-date aggregates (filed):

• Fixed or minimum guaranteed commitments under compute purchase agreements: about $265.6 billion at June 30, 2026, terms of 1–20 years, up from $54.6 billion at December 31, 2025 (the year-end figure counted only commitments with more than a year left, so the jump is not fully like-for-like).
• Leases signed but not yet commenced, primarily compute and infrastructure: about $321.4 billion of future undiscounted obligations at June 30, 2026, up from about $50.2 billion at December 31, 2025. Some are subject to contractual conditions. Commencement 2026–2028; terms from under a year to 20 years.
• Since year-end 2025, new and amended lease and commercial agreements with aggregate future minimum payment obligations of about $417 billion, commencing 2026–2027.
• After June 30, 2026, further agreements of about $245 billion in minimum payments, some conditional.
• Additional TPU capacity for delivery 2028–2031, commercial terms not finalized: company estimate of $125–150 billion, excluded from the commitments table.
• Short-term non-cancelable minimums of about $21 billion are also disclosed in footnotes. One reading of the footnotes (Wired) puts the all-in “if we follow through” figure near $808 billion. That is a sum of overlapping and estimated categories, not a single contractual total.

Named counterparties (filed deal terms, as extracted by multiple readings of the same S-1):

Counterparty	What	Amount	Firmness	Term
Broadcom	Google TPU hardware, 5-year master equipment leases (finance leases); Anthropic Direct	~$161.2B rent, excluding interest	Non-cancellable except default	5 years
Google Cloud	TPUs / cloud, true-up if spend falls short	At least ~$114.5B	Non-cancellable except Google breach	Apr 2026–Jul 2033
Amazon / AWS	Trainium and other compute, true-up; up to 5 GW	At least $110B	Non-cancellable except Amazon breach	May 2026–Apr 2036
SpaceXAI (SpaceX; formerly xAI)	NVIDIA GPU capacity, ~$3.1B/month at full delivery	Up to $120.5B	Cancellable on 90 days’ notice after Jan/Mar 2027	Into 2029
Microsoft Azure	NVIDIA GPUs	At least $44.9B	Non-cancellable except breach or undelivered tranches	Nov 2026–May 2033
Nscale	NVIDIA GPUs, paid as delivered	Up to ~$44.6B	Limited cancellation	Into 2033
Lambda	NVIDIA GPUs; a large tranche depends on a facility lease	~$41.3B	Limited	Into 2034
AMD	Instinct MI450 / Helios; AMD to buy up to $5B of Anthropic stock as chips deliver	Expected to exceed $20B	Non-cancellable except AMD breach	First tranche 2027
Additional TPU (Google and others)	Capacity 2028–2031	Est. $125–150B	Not in the commitments table	Estimate

Microsoft’s floor was amended upward. The November 2025 announcement was about $30 billion; a confidential draft still showed about $31.4 billion; the public S-1 shows at least $44.9 billion after an October 2026 amendment. September 2026 press accounts of a confidential draft’s “$518 billion” total (Google ~$111 billion, Amazon $110 billion, Microsoft ~$31 billion, Broadcom $161 billion, plus other items, ~80% non-cancelable) are draft figures, superseded in part by the public filing. Do not treat $518 billion, $750 billion, or $808 billion as a single filed obligation.

Supplier financing is part of the structure. Broadcom agreed to lend up to $42 billion in convertible notes (9% paid-in-kind, convertible at a 10% discount) solely to fund Anthropic’s rent. Amazon holds the only outstanding convertible notes ($6.2 billion principal). Google, Amazon, and Microsoft are simultaneously investors, compute vendors, resellers, and (for Google and Amazon) model competitors.

Implication. The income statement will not show these dollars until machines are delivered. Once delivered, most of the Google, Amazon, Microsoft, and Broadcom paper is take-or-pay: Anthropic pays the shortfall if usage undershoots. SpaceXAI is the main pressure valve. A demand air-pocket does not shrink the Broadcom/Google/AWS bill; it turns unused capacity into a fixed operating cost on top of whatever training run is in flight.

4. Operating loss versus net loss: the charge bridge

All figures below are from the S-1 (FY audited; H1 unaudited), in millions of USD.

	FY2023	FY2024	FY2025	H1 2025	H1 2026
Revenue	41	386	4,585	1,163	16,286
Total operating expenses	592	3,366	12,645	3,298	16,979
Operating loss	(551)	(2,980)	(8,060)	(2,135)	(693)
Interest income	69	106	431	115	1,111
Change in FV of convertible notes, derivative liabilities, and SAFEs	(1,045)	(5,434)	(34,308)	(10,942)	(72,388)
Other income (expense), net	(2)	—	(8)	—	(8,757)
Net loss	(1,531)	(8,309)	(41,965)	(12,969)	(80,748)

H1 2026 walk from operations to net loss (filed):

• Operating loss: $(693)
• Interest income: +$1,111
• Fair-value change on convertible notes, derivative liabilities, and SAFEs: $(72,388) — “primarily driven by an increase in the fair value of our outstanding convertible promissory notes resulting from an increase in our enterprise valuation.” Amazon rollover notes only: $6,203 million principal ($2,186 million under the 2023 NPA, $4,017 million under the 2024 NPA), carried at $108,566 million at June 30, 2026, up from $41,677 million at December 31, 2025. Conversion prices $2.0928 and $5.0478; about 1.84 billion Class N shares issuable, subject to a 12% ownership cap unless Amazon waives it. Fair-value option under the convertible accounting election; the charge stops when the notes convert at the IPO.
• Other expense, net $(8,757), itemized in the notes as: recognition and change in fair value of 2026 note purchase agreements (Google, March 2026; Amazon, April 2026) $(9,405); holdback liability $(39); change in fair value of warrant rights +$724; other $(37).
• Tax provision: $(21)
• Net loss: $(80,748)

Quarterly operating income turned positive in Q2 2026 at +$974 million, after a Q1 2026 operating loss of $(1,667) million. Operating cash flow was +$3,929 million in H1 2026 (versus $(1,513) million in H1 2025 and $(2,637) million in FY2025). Cash and short-term investments were $131.7 billion at June 30, 2026 ($112.4 billion cash, $19.3 billion short-term investments).

What sits inside operating loss (cash and non-cash)

These are operating items. They are already in the $(693) million and the $(8,060) million. They are not additional charges below the operating line.

Item	FY2025	H1 2025	H1 2026	Character
Stock-based compensation	$1,165M (footnote rounds the year to $1.2B; cash-flow add-back $1,158M)	$286M (cash-flow add-back $279M)	$2,185M	Non-cash; inside personnel. Personnel was $3,152M in H1 2026, so cash compensation was roughly $1B. Unrecognized SBC on unvested RSUs was about $26.6B at June 30, 2026 — a future operating charge, not a current one.
Non-cash charitable contributions	$541M	—	$126M	Non-cash; inside opex.
Bartz copyright settlement	$1,500M	—	—	Cash settlement, recorded entirely in FY2025 operating expenses. Not a fair-value or impairment charge.

Bartz. Bartz et al. v. Anthropic settled for $1.5 billion, the largest known U.S. copyright class settlement. Final approval was July 20, 2026. The S-1 says $300 million was paid in 2025 and $1.2 billion remained in current liabilities at June 30, 2026. Installments run through September 2027. The company excludes it from adjusted operating income as not indicative of ongoing operations. Other copyright matters (music publishers, Canadian actions, Reddit, Bartz opt-outs) are disclosed with no loss estimate.

Impairments. The income-statement bridge and the non-GAAP reconciliation do not identify a material impairment charge in FY2025 or H1 2026. No filed impairment number belongs in this loss walk.

Company-defined adjustments (non-GAAP, filed reconciliation)

Adjusted operating income excludes SBC, non-cash charitable contributions, and the Bartz settlement. Adjusted net income also excludes the two fair-value lines.

Add-back	FY2025	H1 2025	H1 2026
Net loss	(41,965)	(12,969)	(80,748)
SBC	1,165	286	2,185
Non-cash charitable contributions	541	—	126
Bartz settlement	1,500	—	—
FV of notes, derivatives, SAFEs	34,308	10,942	72,388
FV of certain mark-to-market instruments	3	—	8,720
Adjusted net income (loss)	(4,448)	(1,741)	2,671
Adjusted operating income (loss)	(4,854)	(1,849)	1,618
Adjusted operating margin	(106)%	(159)%	10%

H1 2026 GAAP operating loss of $693 million plus $2,185 million of SBC plus $126 million of charitable contributions equals the $1,618 million adjusted operating profit. Interest income of $1,111 million then more than covers the GAAP operating loss before any fair-value add-back. The $2,671 million adjusted net profit is the company’s measure, not cash earnings and not GAAP.

5. Filed versus estimated — a clean split

Filed (S-1, October 26, 2026), or arithmetic strictly on filed lines:

• Revenue, compute and infrastructure, operating loss, net loss, interest income, and the fair-value lines in the table above.
• Cost of revenue and the functional split (R&D $9,741M, S&M $3,808M, G&A $1,050M in H1 2026); the ~85% “gross margin” is derived, and the S-1 does not label it as such.
• SBC, Bartz ($1.5B in FY2025 opex; $1.2B liability at June 30), non-cash charitable contributions.
• Other-expense components ($9,405M note-purchase marks, $724M warrant gain, $39M holdback).
• Amazon note principal $6.2B vs. carrying value $108.6B.
• Operating cash flow $3,929M; capex $482M; cash and short-term investments $131.7B.
• Purchase-commitment floor $265.6B and uncommenced-lease obligations $321.4B at June 30; the $417B / $245B / $125–150B subsequent and estimated buckets as the company states them.
• Named deal floors and caps (Broadcom ~$161.2B, Google at least ~$114.5B, Amazon at least $110B, Microsoft at least $44.9B, SpaceXAI up to $120.5B, Nscale up to ~$44.6B, Lambda ~$41.3B, AMD expected to exceed $20B).
• Q2 2026 operating income of $974M. Consumption/subscription mix and 48% marketplace share.

Estimated or third-party, not a filed line:

• SemiAnalysis inference margins (mid-60s blended, API >80%, up to 88%, subscription compute share >40%).
• Any single “total commitments” number — $518B (September draft, via Reuters), $542–567B, $637B (sum of seven named deals, Tunguz arithmetic), $657B, $750B, or ~$808B. The filing leaves the all-in total structured as separate buckets, some estimated, some conditional, some overlapping with the June 30 notes.
• The exact dollar of compute inside cost of revenue (at most $2.4B) and the exact training-versus-inference split (not disclosed).
• Q3 2026 revenue. The S-1 leaves the quarter blank and says revenue was 86 times the $197 million of Q4 2024, which implies on the order of $17 billion. Press reports of $16.8 billion for Q3, and of ~$33 billion for the first nine months, are outside the H1 financial statements.

6. How to read this if you are underwriting the business

The cash engine and the GAAP loss are different objects. In H1 2026 the company lost $693 million on operations, earned $1.1 billion of interest on a $132 billion cash pile, and generated $3.9 billion of operating cash flow, while booking $81 billion of non-cash valuation marks that more than explain the $80.7 billion net loss. Those marks are real dilution — Amazon’s $6.2 billion of principal is a claim on equity carried at $109 billion — but they are not compute spend and they end at conversion.

The economic risk is the delivery schedule. Compute was 48% of revenue in Q2 only because revenue outran capacity that had already been ordered. The next tranches (Microsoft from November 2026, SpaceXAI ramping toward ~$3.1 billion a month, Broadcom lease commencements, AWS and Google true-ups through the mid-2030s) are largely fixed once delivered. Margin will oscillate with when that capacity lands relative to token demand, not with whether the fair-value charge prints again.

A competitor or a public-market investor should underwrite three numbers separately: customer-serving compute (the ≤$2.4 billion H1 cost-of-revenue pool, before reseller fees), the training residual (most of the other ~$6.6 billion of H1 compute, inside R&D), and the undiscounted obligation stack ($265.6 billion of minimum purchase commitments plus $321.4 billion of uncommenced leases at June 30, plus what was signed after). Conflating the 85% functional margin with company-level profitability, or the $80.7 billion net loss with cash burn, is the misread the filing is built to invite.

---

Recent Findings Supplement (October 2026)

Anthropic’s October 26, 2026 S-1 filing (its first detailed public financial disclosure ahead of a planned Nasdaq IPO under ticker ANT) provides the primary new data on H1 2026 results (ended June 30, 2026), separating operational performance from large non-cash accounting items tied to rising valuation and prior financing instruments.[1][2]

This filing updates prior estimates with audited/unaudited figures, showing revenue growth outpacing compute costs enough for a Q2 operating profit while the GAAP net loss is dominated by fair-value adjustments.[3][4]

Revenue Growth and Operating Results

Anthropic reported $16.286 billion in revenue for H1 2026 (more than 14x the $1.163 billion in H1 2025 and exceeding all of 2025’s $4.585 billion), driven primarily by usage-based consumption (80%+ of revenue).[5][6]

• Operating loss narrowed sharply to $693 million (from $2.135 billion in H1 2025); Q2 alone flipped to $974 million operating income as revenue more than doubled QoQ to $11.555 billion while compute/infrastructure costs rose more modestly to $5.558 billion.[3][7]
• Adjusted operating income reached $1.618 billion (10% margin) after excluding stock-based compensation and other items; operating cash flow was positive $3.929 billion.[4][8]

For competitors or entrants: The data shows a viable path to operating leverage once scale is achieved, but highlights heavy reliance on usage-based revenue that can fluctuate seasonally or with customer budgets.[9]

Net Loss Breakdown and Non-Cash Charges

The GAAP net loss totaled $80.748 billion for H1 2026 (vs. $12.969 billion in H1 2025), driven overwhelmingly by non-operating, non-cash fair-value adjustments rather than cash outflows.[1][10]

• $72.388 billion from changes in fair value of convertible notes, derivative liabilities, and SAFEs (primarily reflecting higher enterprise valuation).[4]
• $8.720 billion (or ~$8.7 billion) from changes in fair value of other mark-to-market financial instruments.[4]
• Stock-based compensation of $2.185 billion (included in operating expenses; up significantly YoY).[4]
• Non-cash charitable contributions of $126 million.[4]
• The $1.5 billion Bartz et al. copyright settlement (literary works used in training) was recorded as a charge in 2025 ($300 million paid that year; $1.2 billion remaining current liability at June 30, 2026); no incremental charge appears in H1 2026 results.[11][12]

Company-adjusted net income (excluding the above valuation changes, SBC, etc.) was a $2.671 billion profit.[4] No impairments are detailed in the reporting.[4]

Implication: GAAP net loss figures are not indicative of cash burn or operational health; adjusted metrics and cash flow provide a clearer operational view, though valuation-driven charges will recur with further equity raises or valuation shifts.[2]

Compute and Infrastructure Spend

Compute and infrastructure expenses reached $8.945 billion in H1 2026 (nearly 4x the $2.334 billion in H1 2025), covering both training and inference.[5][10]

• Only $2.38 billion was classified in cost of revenue (“substantially all” of which relates to customer-serving workloads), implying a proxy gross margin (revenue less CoR) of ~85% (vs. ~12% in H1 2025); the balance (~$6.565 billion) appears in R&D and sales/marketing (e.g., training, research, free-tier access).[13][13]
• Overall compute spend fell to 48% of Q2 revenue (from much higher earlier percentages).[3]

For market entrants: High fixed/variable compute costs remain a barrier, but the accounting split and improving efficiency (revenue growth outpacing spend) suggest gross-margin-like metrics can appear strong when most training/free usage is expensed elsewhere.[13]

Cloud and Compute Commitments

The S-1 details substantial multi-year minimum commitments (largely non-cancellable except in limited cases), totaling hundreds of billions across providers; these support a multi-chip strategy (AWS Trainium, Google TPUs, Nvidia GPUs via partners, etc.).[14][15]

• AWS: At least $110 billion (through April 2036).[16]
• Google Cloud/TPUs: At least $114.5 billion (through July 2033); additional estimated $125–150 billion for further TPU capacity (2028–2031).[14]
• Microsoft Azure: At least $44.9 billion (through May 2033; recently amended upward from prior ~$30–31 billion figures).[14]
• Broadcom (TPU-related leases): ~$161.2 billion.[14]
• Other Nvidia-powered: Nscale ~$44.6 billion; Lambda ~$41.3 billion; SpaceXAI up to ~$120.5 billion (largely cancellable on 90 days’ notice after early 2027).[14]

Minimum non-cancellable commitments are commonly cited in the $265–430+ billion range across major providers, with broader totals (including estimates and softer deals) reaching $542–567 billion or higher.[17][18]

Implication: These locked-in obligations create significant supplier leverage (and revenue share, with ~48% of H1 revenue flowing through Google/Amazon/Microsoft channels) while representing a major portion of cloud providers’ backlogs; new entrants face similar scale requirements or must negotiate differently.[19]

Cash Position and Context

Cash and short-term investments stood at ~$131.7 billion as of June 30, 2026 (supported by $141.1 billion total capital raised historically).[3]

All figures above derive from the October 26, 2026 S-1 (public estimates prior to filing have been superseded by these filed numbers). No additional post-filing updates or separate research/policy items appear in recent coverage.[20]

These disclosures mark a shift from private estimates to standardized, comparable public data, enabling better benchmarking of AI lab economics.

Anthropic has turned a capital-intensive model business into something closer to a high-margin software company on a cash basis, while still signing multi-hundred-billion-dollar compute contracts it mostly cannot cancel. Its October 26, 2026 S-1 is the first audited window into a frontier lab. OpenAI remains private, larger on a consumer footprint, and still burning cash against a plan that does not turn free-cash-flow positive until around 2030.

Anthropic’s funding path: from a $124 million seed-stage lab to a $965 billion Series H

Anthropic’s valuation did not compound smoothly. It jumped in steps that tracked the cost of the next Claude training run and the willingness of cloud vendors to fund a customer that would also buy their chips.

Round	Date	Amount	Post-money valuation	Lead / notable investors
Series A	May 2021	$124 million	~$550 million	Jaan Tallinn; Dustin Moskovitz, Eric Schmidt
Series B	Apr 2022	$580 million	~$4.1 billion	Sam Bankman-Fried / FTX-Alameda
Series C	May 2023	$450 million	~$4.1 billion	Spark Capital; Google, Salesforce Ventures
Series D	Early 2024	$750 million	$18.4 billion	Menlo Ventures
Series E	Mar 3, 2025	$3.5 billion	$61.5 billion	Lightspeed
Series F	Sep 2, 2025	$13 billion	$183 billion	ICONIQ; Fidelity, Lightspeed co-leads
Series G	Feb 12, 2026	$30 billion	$380 billion	GIC and Coatue
Series H	May 28, 2026	$65 billion	$965 billion	Altimeter, Dragoneer, Greenoaks, Sequoia

Sources for the lettered rounds include company announcements as compiled by secondary trackers and contemporaneous reporting. [1] [2] [3] [4]

Strategic capital sits beside those rounds and is easy to double-count:

• Amazon has invested at least $13 billion and in April 2026 agreed to invest up to another $20 billion. In return Anthropic committed to spend more than $100 billion on Amazon cloud services over 10 years. [5] [6]
• Google has committed up to $40 billion, including an initial $10 billion announced in April 2026, tied to cloud capacity and Google’s TPU chips. A separate analysis of the S-1 puts Google’s 2026 equity purchases at about $10 billion of Series G-2 and $10 billion of Series H. [5] [7]
• Microsoft and Nvidia announced in November 2025 commitments of up to $5 billion and $10 billion respectively. [3] [8]

Other recurring names on the later rounds include Fidelity, Blackstone, T. Rowe Price, Temasek, QIA, MGX, Founders Fund, General Catalyst, Jane Street, Capital Group, D1, and Baillie Gifford. Spark’s Yasmin Razavi has been described as the only traditional VC on the board. [5] [9]

What the S-1 settles. As of June 30, 2026, Anthropic said it had raised an aggregate of $141.1 billion. Roughly $110 billion of that arrived in the first half of 2026, with the Series H the largest single piece (about $65.2 billion in the filing’s accounting). [10] [11] Aggregator totals that land anywhere from about $68 billion to more than $160 billion are not reliable; they mix equity, contingent vendor commitments, and debt, or stop before the S-1.

Cash, burn, and whether the business already funds itself

The cash number is not a rumor. The S-1 shows $112.4 billion of cash and cash equivalents and $19.3 billion of short-term investments — $131.7 billion — at June 30, 2026. CNBC’s comparison of non-bank U.S. companies put that pile behind only Berkshire Hathaway and Alphabet. [10] [12]

The burn story is the opposite of the headline loss:

• First-half 2026 revenue was $16.29 billion, versus $1.16 billion a year earlier. Full-year figures: $41 million (2023), $386 million (2024), $4.59 billion (2025). [13] [14]
• Operating loss narrowed to $693 million in the first half, from $2.14 billion a year earlier. The June quarter itself was a $974 million GAAP operating profit — the first operating profit the filing reports. [11] [15]
• The $80.7 billion first-half net loss is mostly non-cash: about $72.4 billion is a fair-value mark-up on convertible notes, derivative liabilities, and SAFEs as the private valuation rose. Adjusted operating income was $1.62 billion (about a 10% margin). Adjusted net income was about $2.7 billion. [11] [14] [16]
• Operating cash flow was +$3.93 billion in the first half. Capex was only about $482 million because Anthropic rents most compute. The Information reported $3.5 billion of free cash flow on that $16.3 billion of revenue, reversing a $1.6 billion cash burn a year earlier. [17] [18] [19]

SemiAnalysis and Yahoo Finance both read the filing as roughly $9–10 billion of cumulative cash consumed out of $141.1 billion raised — less than $10 billion of lifetime burn to a business that later told investors it did $16.8 billion in a single quarter. [11] [15]

Breakeven status, with the caveats. On an operating-cash and adjusted-profit basis, Anthropic has already crossed. Management told investors in mid-October that the third quarter was cash-flow positive, adjusted-profitable for a second straight quarter, with an 11% adjusted operating margin and $1.70 of revenue per $1 of compute. Third-quarter revenue was $16.8 billion, up about 45% from roughly $11.6 billion in the second quarter and about $4.7 billion in the first. [20] [21] [22] GAAP net income is still deeply negative because of those fair-value charges; the notes are expected to convert around the IPO, subject to an ownership cap on Amazon. The prospectus also says cash, sales, and the credit facility should cover at least the next 12 months — and then warns the company may need more financing sooner than it expects, precisely because the compute contracts are fixed and revenue is not. [12]

The balance-sheet “convertible notes payable” of $108.6 billion is a fair-value liability, not cash principal. Analysts reading the filing put actual outstanding Amazon note principal near $6.2 billion. [12] [19]

How the business actually makes money

Anthropic is an enterprise consumption business that happens to have a consumer chatbot, not the reverse.

In the first half of 2026, consumption-based (token) fees were $13.1 billion, about 80% of revenue; subscriptions were $3.2 billion, about 20%. [23] [24] More than 11,900 customers spending at least $100,000 a year generated about $12 billion, or 74% of first-half revenue. That cohort includes every Fortune 10 company and more than half the Fortune 500. More than 1,900 customers spend over $1 million a year and about 250 spend over $10 million. Net revenue retention, excluding Claude Pro and Max, was about 900% as of September 30, 2026. [11] [23]

Distribution is the structural quirk. Sales through Amazon, Google, and Microsoft cloud marketplaces were 48% of first-half revenue (about $7.8 billion), up from 11% in 2023, 32% in 2024, and 47% in 2025. Distribution fees to those resellers were about 17% of first-half revenue in one reading of the filing. [25] [26] [27] The same three companies are investors, compute suppliers, sales channels, and, in Google and Microsoft’s case, model competitors.

Gross margin needs a definition. Cost of revenue implies an 85% gross margin in the first half, up from 12% a year earlier. [23] [28] That is not the full compute bill. “Compute and infrastructure,” largely below gross profit in the operating statement, was $8.95 billion in the first half (55% of revenue) and fell from 251% of revenue in Q1 2025 to 48% in Q2 2026. Revenue per dollar of that compute line went from $0.40 to $2.08 over the same span. [29] [30] Training and new capacity still hit in discrete steps, so the ratio can reverse when a large block of contracted machines comes online.

Run-rate figures that circulated before the S-1 — over $47 billion in May 2026, over $65 billion by late July — are company or source-cited annualizations, not GAAP. Growth then slowed: Q3’s 45% sequential gain was a sharp deceleration from Q2’s more-than-doubling, and Sacra reported annualized revenue only around $70 billion by October. [2] [31] [20]

The compute overhang that the cash pile does not retire

Positive cash flow does not retire the contracts. Analysts summing the seven dollar figures in the S-1 get about $637 billion:

• Broadcom equipment leases (TPU capacity): about $161.2 billion of rent, largely non-cancellable
• SpaceXAI (Nvidia GPUs): up to $120.5 billion, about $3.1 billion a month once fully delivered; cancellable on 90 days’ notice after early 2027
• Google Cloud: at least about $114.5 billion, April 2026–July 2033, payable regardless of usage
• Amazon: at least $110 billion, May 2026–April 2036, payable regardless of usage
• Microsoft: at least $44.9 billion
• Nscale and Lambda: up to roughly $44.6 billion and $41.3 billion

[24] [32] [30]

A September draft prospectus had put commitments at at least $518 billion, about 80% non-cancellable — the gap is mostly later amendments (Microsoft rose from about $31 billion) plus the SpaceXAI and Broadcom lines being fully specified. [33] [34] Additional TPU capacity under discussion was estimated at $125–150 billion, and a full AMD buy would exceed $20 billion; those sit outside the $637 billion sum. [24] The filing’s own risk factor states the mismatch: revenue is mostly consumption-based and not locked in, while compute commitments run for years.

For someone entering this market, that is the competitive fact. Anthropic’s near-term P&L can look like software. Its decade-long cost base looks like a utility with take-or-pay contracts written by its own shareholders.

Side-by-side with OpenAI

OpenAI’s last primary mark is its own: on March 31, 2026 it closed $122 billion of committed capital at an $852 billion post-money valuation. The round was anchored by Amazon (up to $50 billion, of which about $35 billion is contingent on an IPO or an AGI milestone), Nvidia ($30 billion), and SoftBank ($30 billion), with Microsoft continuing and a broader book including a16z, D. E. Shaw, MGX, TPG, and T. Rowe Price. About $3 billion came from individual investors through bank channels. The credit facility was expanded to about $4.7 billion. [35] [36] [37] Prior steps: a $6.6 billion round at $157 billion in October 2024, a $40 billion SoftBank-led round at $300 billion in March 2025, and a secondary around $500 billion in October 2025. [38] Lifetime totals in commercial databases cluster around $176–187 billion, but they disagree on whether contingent tranches count; Statista had $176.4 billion as of February 27, 2026, before the round fully closed. [39] [40] Treat “total raised” for OpenAI as less clean than Anthropic’s S-1 figure.

Metric	Anthropic	OpenAI
Latest primary valuation	$965 billion (May 28, 2026 Series H)	$852 billion post-money (March 31, 2026)
Capital raised	$141.1 billion aggregate through June 30, 2026 (S-1)	$122 billion committed in the March 2026 round alone; lifetime estimates ~$176–187 billion, partly contingent
Cash and equivalents	$131.7 billion cash + short-term investments at June 30, 2026	About $40 billion at end-2025; more than $73 billion cash and securities at end of Q1 2026, after the round began funding
Booked revenue	2025: $4.59 billion. H1 2026: $16.3 billion. Q3 2026: $16.8 billion (investor meeting, not yet in the S-1)	2025: $13.07 billion. Q1 2026: $5.7 billion. Q2 2026: $6.7 billion. H1 2026 about $12.4 billion
Run rate	Over $47 billion (May); over $65 billion (late July). October annualized figures slowed toward ~$70 billion in some estimates	About $50 billion at end of September 2026, confirmed by the company. A ~$70 billion figure counted gross partner revenue and was walked back
Gross margin	~85% in H1 2026 on cost of revenue; compute-and-infrastructure still 55% of H1 revenue and 48% in Q2	About 33% in 2025; one secondary compilation put Q1 2026 near 39%. Not from an S-1
Operating result	H1 2026 operating loss $693 million; Q2 operating profit $974 million; adjusted operating income $1.62 billion	2025 operating loss $20.92 billion. Q1 2026 cash burn $3.7 billion. Q2 operating loss including stock compensation reported at $12.3 billion
Cash flow	H1 operating cash flow +$3.9 billion; H1 free cash flow about +$3.5 billion. Q3 described as cash-flow positive	Internal July 2026 plan: $278 billion of negative free cash flow from 2026 through 2030. March raise expected to be exhausted by 2028, with about $116 billion still to raise
Compute commitments	~$637 billion itemized in the S-1 across seven counterparties; cloud minimums with Amazon, Google, and Microsoft are largely payable regardless of use	Supplier filings: incremental $250 billion to Microsoft Azure, $138 billion to Amazon (a $38 billion deal expanded by $100 billion), up to about $18–22 billion to CoreWeave. Internal plan: about $856 billion of compute and infrastructure through 2030. Altman has also cited on the order of $1.4 trillion of data-center commitments
Business mix	~80% token consumption, ~20% subscriptions; 48% of revenue via cloud marketplaces; enterprise-heavy, ~74% of H1 from $100k+ accounts	Entered 2026 about 60/40 consumer/enterprise. By August 2026, CFO Sarah Friar told investors enterprise had overtaken consumer, two quarters ahead of the year-end parity target. ChatGPT scale (900 million-plus weekly actives; later ~1 billion monthly actives) is the consumer moat Anthropic does not have

Anthropic column: S-1 and the October investor meeting. [10] [11] [14] [23] [20] OpenAI column: company announcement, FT-verified 2025 financials, The Information, FT, and CNBC. [41] [42] [43] [44] [45] [35] [46] [47]

Do not compare the revenue lines as if they use the same ruler. Anthropic books the gross value of sales through cloud partners and expenses the partner’s cut. OpenAI books only its own share of certain partner arrangements. OpenAI itself said reports of a $70 billion run rate were wrong and confirmed about $50 billion. On a gross basis the gap to Anthropic would be smaller; on OpenAI’s net basis, Anthropic’s quarterly booked revenue ($16.8 billion in Q3) is already in the same neighborhood as a $50 billion annualized rate. [42] [48] Third-quarter run-rate growth at OpenAI was still fast — about 77% for the company and 107% for enterprise, per what it told investors. [47]

Capital efficiency is the real split. Anthropic reached a first-half free-cash-flow margin that one cloud-comps benchmark put near 21% (11% on a trailing-twelve-month basis) after consuming on the order of $10 billion of cash in its entire history. [11] OpenAI’s audited 2025 result was an operating loss larger than revenue ($20.9 billion on $13.1 billion), R&D alone of $19.2 billion, and a plan that still has expenses outrunning cash until the end of the decade even after cutting the five-year burn forecast from about $305 billion to $278 billion. [41] [43] Its Q1 2026 cash burn of $3.7 billion was more than half of that quarter’s $5.7 billion of revenue. [45]

The mechanism is mix plus cost timing. Anthropic’s revenue is concentrated in high-intensity enterprise and coding workloads (Claude Code was cited at an $8 billion annualized rate as of May, and agents multiplied billable tokens). That produces high net revenue retention and a compute bill that, for a few quarters, grew much more slowly than sales. OpenAI still carries the cost of a free and lightly monetized consumer base at ChatGPT scale, plus heavier training spend, even after enterprise crossed 50% of revenue. Gross margin in the low-to-mid 30s cannot fund an $856 billion compute plan. Anthropic’s 85% gross margin can fund operations today and still cannot, by itself, fund $110 billion minimums to Amazon and Google that are owed whether customers show up.

Both labs are financed in a circle. Amazon, Google, Microsoft, and Nvidia are simultaneously investors, suppliers, and in several cases distributors. Of the capital OpenAI has raised since the start of 2025, a large share came from companies it buys compute from — Amazon and Nvidia alone account for $80 billion of commitments in that window, and Amazon had paid $28.7 billion of its $50 billion by the end of June. [33] Anthropic’s Series H explicitly folded in previously committed hyperscaler money. Equity in these rounds is partly a prepayment for future cloud invoices.

What this means if you are competing, investing, or selling into either company

Anthropic is the one with public books, a November listing targeted under the ticker ANT (reports have pointed to the week of November 16, a raise that could exceed $100 billion, and a valuation discussed in a $1.5–2 trillion range — down from $2 trillion talk after Q3 growth slowed). Goldman Sachs and Morgan Stanley lead the syndicate. The last primary price is $965 billion; a secondary indication the day before the filing’s market reaction was already lower. [1] [49] [16] OpenAI has talked about a listing, then pushed it; by late summer Friar was describing an IPO as a 2027 milestone, and October reports had it discussing another private round near $30 billion at a valuation above $1 trillion. Those later OpenAI round figures are press reports, not a closed deal. [50] [40]

Three implications follow from the numbers rather than from the narrative:

1. Near-term cash is not the constraint at either lab. Contracted compute is. Anthropic can fund operations from sales and still has to meet usage-independent minimums that dwarf 2025 revenue. OpenAI’s own plan says the March round runs out in 2028. A competitor that cannot sign take-or-pay deals with Amazon, Google, or Microsoft is not in the same supply market, regardless of model quality.

2. Enterprise API economics are currently more efficient than consumer scale. Anthropic got to operating profit without ChatGPT’s distribution. OpenAI’s enterprise crossover is the attempt to import that efficiency onto a consumer cost base. If coding and agent workloads keep concentrating on Claude, Anthropic’s 900% net retention is a harder moat than weekly active users. If those workloads commoditize — the S-1 risk factors and Sacra’s October note both flag customers cutting Claude Code usage and cheaper open models — the fixed compute bill lands on a shrinking numerator.

3. Reported revenue and reported losses are both easy to misuse. Anthropic’s $80.7 billion loss is mostly a mark-to-market on its own rising valuation. OpenAI’s $50 billion versus $70 billion run-rate dispute is an accounting-policy dispute. Any comparison that ignores gross-versus-net marketplace revenue, or that treats fair-value charges as cash burn, will get the capital-efficiency ranking backward.

---

Recent Findings Supplement (October 2026)

Anthropic closed its largest-ever private round in May 2026, raising $65 billion at a $965 billion post-money valuation—surpassing OpenAI’s prior $852 billion mark—and filed its S-1 on October 26, 2026, for a Nasdaq IPO (“ANT”) potentially valuing it at $1.8–2 trillion.[1][2]

This round incorporated $15 billion of prior hyperscaler commitments (including $5 billion from Amazon) alongside new capital from Altimeter Capital, Dragoneer, Greenoaks, Sequoia, Coatue, GIC, Capital Group, and strategic chip partners (Samsung, SK Hynix, Micron). Run-rate revenue had reached $47 billion earlier in May and climbed to approximately $65 billion by July (gross basis, including partner sales).[3][4]

• Total capital raised reached $141.1 billion by June 30, 2026 (roughly $110–111 billion in 2026 alone), with earlier rounds including $30 billion Series G in February at $380 billion post-money.[5]
• Google committed up to $40 billion in April 2026 ($10 billion immediate at a $350 billion valuation, plus $30 billion milestones). Amazon’s cumulative commitments now stand at up to $33 billion potential.[6]
• The IPO filing positions Anthropic for a potential $100 billion+ raise (existing investors also selling shares), with expectations of a blockbuster debut testing broader investor appetite beyond Silicon Valley.[5]

This capital influx and IPO path give Anthropic a war chest and liquidity event ahead of OpenAI, pressuring rivals to match scale while highlighting how hyperscaler partnerships convert into both funding and locked-in compute demand.

Anthropic’s S-1 reveals explosive H1 2026 revenue growth to $16.3 billion (14x YoY from $1.16 billion), narrowing operating losses, positive operating cash flow, and a massive $131.7 billion cash and short-term investments balance as of June 30.[2][7]

H1 compute/infrastructure spend was $8.95 billion against that revenue, supporting ~85% gross margins on existing models and an 85%+ net revenue retention rate. Q2 operating income turned positive at $974 million (8.4% GAAP margin, 19% adjusted), with full H1 operating loss at just $693 million. Operating cash flow reached +$3.9 billion for the half (free cash flow +$3.5 billion reported), reversing prior burns.[8][9]

• 2025 full-year revenue was $4.59 billion; H1 2026 alone exceeded all of 2025. Q2 revenue hit ~$11.6 billion (up sharply QoQ).[7]
• Cash pile ($112.4 billion equivalents + $19.3 billion short-term investments) exceeds most major U.S. non-bank companies and will grow further with the IPO.[10]
• Net loss of $80.7 billion was driven almost entirely by $72.4 billion non-cash fair-value adjustments on convertible notes; underlying operations show rapid improvement.[2]
• Over $518 billion in compute commitments (largely non-cancellable) over ~10 years, plus hundreds of billions in future lease/equipment obligations, underscore the scale of infrastructure bets.[11]

Anthropic has transitioned from heavy burn to cash-flow positive operations faster than expected, using its data and enterprise focus to achieve high margins on deployed capacity—creating a capital-efficiency moat that forces competitors to either match pricing power or absorb deeper losses.

OpenAI’s annualized revenue run rate reached approximately $50 billion (net basis) by end-September 2026—up from ~$40 billion in August and ~$20 billion at end-2025—while the company explores a $30+ billion pre-IPO round at a $1.4 trillion pre-money valuation and delays its own IPO to 2027+.[12][13]

Enterprise revenue has overtaken consumer, with business customers exceeding two million; the company projects $70+ billion run rate by year-end 2026. Its March 2026 $122 billion round (at $852 billion post-money) brought cash in hand to roughly $162 billion, but internal plans project a $278 billion cumulative cash burn through 2030, with current resources potentially exhausted by 2028 and ~$116 billion more needed.[14][15]

• Revenue accounting differs from Anthropic’s gross method (OpenAI reports only its share of partner sales, e.g., via Microsoft).[16]
• Compute and infrastructure commitments remain enormous (~$856 billion planned through 2030 in one projection).[15]
• Recent growth accelerated post-GPT-5.6 (July), with enterprise doubling in Q3, though some investor estimates of higher run rates were revised downward.[17]

OpenAI maintains stronger consumer scale and partnership leverage but faces steeper projected cash needs and slower reported profitability trajectory, making its next funding round a critical test of whether enterprise momentum can close the efficiency gap with Anthropic.

Side-by-side (most recent public or reported figures, post-April 2026):

• Revenue/Run-rate: Anthropic ~$65 billion (July gross); OpenAI ~$50 billion (Sep net, targeting $70+ billion YE). Anthropic showed faster recent scaling in H1 S-1 data.[3][12]
• Margins/Cash Flow: Anthropic H1 2026 operating margin improving to positive in Q2, +$3.9 billion operating cash flow, high gross margins on deployed models; OpenAI still projecting heavy multi-year burns ($278 billion through 2030).[8][15]
• Cash on Hand: Anthropic $131.7 billion (June 30); OpenAI ~$162 billion (post-March round).[2][15]
• Capital Raised/Valuation: Anthropic $141.1 billion total (~$65 billion Series H at $965 billion May); OpenAI $122 billion March round at $852 billion, seeking $30+ billion at $1.4 trillion pre-money.[1][13]
• Commitments: Both face hundreds of billions in compute obligations; Anthropic’s are more explicitly quantified in S-1 with high non-cancellable portions.[11]

Anthropic leads on near-term capital efficiency and cash generation while OpenAI retains advantages in absolute consumer reach and certain partnership structures—differences that will be tested in their respective IPO timelines and ability to fund non-cancellable infrastructure without eroding margins.

Business mix and capital efficiency diverge sharply: Anthropic’s enterprise/API focus (over 1,000 customers >$1M/year, strong in regulated sectors via safety/interpretability positioning) pairs with higher reported gross margins and quicker path to positive cash flow on existing capacity.[3] OpenAI balances a massive consumer base (ChatGPT) with fast-growing enterprise but reports net revenue and faces deeper projected losses tied to aggressive scaling.[14]

Anthropic’s model treats safety research and enterprise governance as core differentiators that unlock sticky, high-value contracts, enabling it to underwrite compute spend more efficiently. OpenAI’s consumer flywheel drives volume but requires heavier discounting or subsidies in some enterprise deals, pressuring margins.[18]

New entrants or incumbents must choose: replicate Anthropic’s enterprise/safety moat for better unit economics or OpenAI’s consumer scale for volume—while both paths demand unprecedented capital discipline amid non-cancellable compute bets that dwarf current revenues.

For competitors, Anthropic’s S-1 and positive cash-flow inflection set a new benchmark for capital efficiency in frontier AI, while OpenAI’s larger consumer base and upcoming raise highlight the ongoing premium on distribution and partnerships; the race now hinges on who sustains revenue growth without exhausting cash reserves or triggering valuation resets ahead of public markets.

Anthropic’s public S-1 (filed October 26, 2026) makes the audited half-year numbers solid and leaves the numbers investors have been trading—year-end run rate, third-quarter dollars, and the full compute stack—either blank or defined in a way that does not match how headlines use them. The filing is the first frontier-lab prospectus, ticker ANT, and the gaps are mostly mechanical, not evidence of a fabricated document. They are still large enough to misprice the equity if run rate, gross marketplace sales, or “$600–800 billion of commitments” are treated as interchangeable with GAAP revenue.

Run rate is a 28-day snapshot, and the S-1 leaves it blank

The number that has driven private valuations is not recognized revenue. Anthropic defines Run-Rate Revenue in the prospectus as trailing 28-day consumption fees multiplied by 13, plus monthly subscription fees multiplied by 12. That is a four-week annualization of metered usage, not contracted ARR, and the sentence that would give the figure as of October 31, 2026 is empty. The cover-style claim that Claude went from zero at commercial launch to a dollar amount for the quarter ended September 30, 2026 is also blank.

What is filled in is GAAP. First-half 2026 revenue was $16.286 billion, versus $1.163 billion a year earlier. The March quarter was $4.731 billion and the June quarter $11.555 billion—larger than all of 2025 ($4.585 billion). Annualizing the June quarter produces about $46 billion, which lines up with the roughly $47 billion run rate the company cited around its May Series H, not with the roughly $65 billion figure circulated for the end of July. A business that more than doubled from Q1 to Q2 will print a run rate far above trailing recognized revenue; the reverse is also true if a hot four weeks cools. The filing itself warns that revenue is mostly consumption-based and much of it is not under long-term contract, and that growth is not expected to continue at comparable rates.

• Q3 is stated only as a multiple: 86 times fourth-quarter 2024 revenue of $197 million, which is about $16.9 billion. The Financial Times and Bloomberg, citing an October investor meeting, reported $16.8 billion—roughly 45% sequential growth after 144% in Q2.
• Remaining performance obligations were only $10.8 billion at June 30. That is the contracted backlog. It is not the run rate.
• Spend is seasonal: customer spend fell about 20% in the two weeks ended January 2, 2026, versus the prior two weeks. A late-year 28-day window can overstate a full year.

For anyone underwriting a multiple on “$65–70 billion ARR,” the primary check is the amended S-1’s filled-in Run-Rate Revenue and the audited (or reviewed) September quarter, not a private slide from July.

Gross marketplace accounting inflates the top line relative to cash kept—and relative to OpenAI

About 48% of first-half 2026 revenue came through Amazon, Google, and Microsoft cloud marketplaces, up from about 11% in 2023, 32% in 2024, and 47% in 2025. Anthropic says it is the principal: it has the customer contract, sets price, and delivers the model, so it books the full amount and records platform fees in sales, marketing, and partnerships. The platforms bill the customer and remit cash net of those fees. Marketplace-originated balances were 58% of net receivables at June 30.

The public S-1 does not break out the fee rate for 2026. On the confidential draft, Reuters calculated that 2025 marketplace sales were about $2.16 billion and distribution fees about $351 million—roughly 16 cents per marketplace dollar. Applied to a 48% channel mix, that is several billion dollars of H1 “revenue” that never arrives as cash. OpenAI, by contrast, has told investors it records only its share of certain Microsoft-routed sales. The Financial Times reported OpenAI’s September annualized revenue near $50 billion on that net basis, about $20 billion below a roughly $70 billion figure that investors had built by grossing OpenAI up to look like Anthropic. Both treatments can be GAAP-compliant. They are not the same ruler. Comparing the two labs’ headline run rates without a fee bridge overstates Anthropic’s lead and, earlier this month, briefly overstated OpenAI’s.

• Amazon and Google are simultaneously investors, compute suppliers, resellers, and model competitors. A dispute that slows remittance hits cash even if the revenue has already been recognized.
• The S-1 does not quantify H1 platform fees, so a net-revenue restatement is not possible from the public document alone.

Concentration looks better than 2025, but the revenue is still optional and partly non-compliant

No single customer exceeded 10% of revenue in the nine months to September 30, and the top five were about 25%. Customers outside the top five were about 79% of third-quarter revenue, up from 64% in the first quarter. That is an improvement on the confidential-draft picture, in which two unnamed customers each supplied about 12% of 2025 revenue. It is not diversification in the SaaS sense. Eighty percent of H1 revenue was token consumption ($13.1 billion); subscriptions were 20% ($3.2 billion). Large accounts can cut usage without breaking a contract. Net revenue retention of about 900% as of September 30 measures what a qualified cohort spent in the latest three months versus the same three months a year earlier. At this growth rate that mostly says last year’s base was tiny and Claude Code expanded it, not that the book is locked. The metric excludes the fixed Pro and Max plans, whose fees do not expand the same way. Claude Enterprise users who also use Claude Code spend on the order of 20 times users who do not.

A separate quality issue sits inside the growth. Anthropic estimates that 5% to 7% of revenue for the nine months ended September 30 came from customers later found to be violating usage policy or commercial terms—largely distillation, in which other labs train on Claude outputs. That revenue is recognized as tokens are served, before accounts are cut. Detection was tightened in August and September. The company expects the share to fall and says the hit to future revenue and growth is uncertain in both timing and size. On a nine-month base near $33 billion, 5–7% is roughly $1.6–2.3 billion that was real cash and is not a durable customer. Pre-filing investor meetings used the same 5–7% range to explain part of the Q3 deceleration. The S-1 confirms the estimate; it does not name the accounts or split distillation from other violations.

Price cuts and reasoning workloads pull in opposite directions, and the 85% “gross margin” is not the compute bill

Unit prices are falling on purpose. The filing says Opus 5.5 cut cost per task by about 40% versus Opus 5, and that cheaper tasks drove more usage. Fable 5.1 cut typical workloads by about 25% and complex coding and agentic tasks by up to 45%. Models released in the prior six months accounted for more than 80% of third-quarter revenue, so Anthropic cannot sit on an old price list: the frontier release is the revenue. Open-weight models are named as pricing pressure on workloads that care more about cost than intelligence. Forrester’s read of the filing is that this is now a volume business: fixed compute has been bought, and customers have only $10.8 billion of remaining performance obligations against a much larger supplier stack, so buyers have leverage to demand lower task prices.

Reasoning and agents are the volume offset and the margin risk at once. Agentic loops reread context, call tools, and emit “thinking” tokens. That multiplies billable consumption—the mechanism behind Claude Code’s outsized spend—but it also multiplies inference compute per completed task. If price per task falls 25–45% per generation and reasoning tokens rise, serving margin depends on whether cache hits, cheaper chips, and higher utilization outrun the token inflation. The S-1 says they have so far, and that margins held through the recent price cuts. It also says the company expects to buy more short-term compute, at a higher unit cost, when demand exceeds long-term capacity, and that this will compress operating margin in the near term. New capacity is expensed when it comes online, often in large steps, before the revenue it supports. Operating margin is therefore expected to swing “at times meaningfully” even if underlying unit economics improve.

The 85% figure that circulated as a cloud-like gross margin is a functional allocation, not the full compute cost. Cost of revenue was $2.38 billion in the first half—substantially all customer-serving compute—on $16.286 billion of revenue, or about 85%, up from about 12% a year earlier. Total compute and infrastructure expense was $8.945 billion, or 55% of revenue (48% in the June quarter, down from 251% in Q1 2025). Roughly $6.6 billion of compute sits in research and development (training, research, internal Claude use) and in sales (free-tier serving), not in cost of revenue. H1 GAAP operating loss was $693 million; adjusted operating income was positive; the June quarter was the first GAAP operating profit at $974 million. An 85% gross margin does not mean 85 cents of every revenue dollar is available after the chips are paid for.

“The compute commitment” is several different numbers, and the company’s own total is blank

The prospectus states that, as of and after September 30, 2026, purchase commitments and leases—including short-term deals and leases not yet commenced—have an aggregate committed spend of “more than $____,” substantially all compute, with most of the remainder expected between 2031 and 2045. The dollar is not filled in. Everything else in the press is a reconstruction.

Named deals that analysts have added up come to roughly $637 billion: Broadcom equipment leases about $161.2 billion (largely non-cancellable), SpaceXAI up to $120.5 billion (cancelable on 90 days’ notice after early 2027), Google at least $114.5 billion, Amazon at least $110 billion, Microsoft at least $44.9 billion, Nscale up to about $44.6 billion, Lambda about $41.3 billion. Of that, on the order of $431 billion is minimum spend or lease rent owed largely regardless of usage; about $206 billion is “up to” or billed as capacity is accepted. A further $125–150 billion of TPU capacity for 2028–2031 is estimated and not finalized, and AMD hardware is expected to exceed $20 billion if fully taken. Other published totals measure different things: about $265.6 billion of fixed or minimum-guaranteed commitments at June 30; about $417 billion of future minimum obligations on agreements entered or amended since year-end 2025; Reuters’ pre-filing $518 billion; Wall Street Journal draft ranges around $450 billion of cloud deals and more than $600 billion overall; Fortune’s $657 billion with about $498 billion called non-cancellable; some secondary write-ups at $750–810 billion once estimates are stacked on top of caps.

The economic point does not depend on which total wins. Revenue is optional and monthly; a large core of Google, Amazon, and Microsoft spend is take-or-pay. Cash and short-term investments were $131.7 billion at June 30 ($112.4 billion cash plus $19.3 billion of investments), against obligations that, on any of the larger readings, run for most of a decade. H1 operating cash flow was positive at $3.9 billion. That does not fund a $400 billion-plus minimum stack. The $80.7 billion first-half net loss is mostly not cash either: about $72.4 billion is the mark-to-market on convertible notes, derivatives, and SAFEs as the private valuation rose. Treating that loss as burn, or treating every “up to” cap as a debt-like obligation, are both wrong.

What outlets got wrong, and what was never a company claim

Discrepancies are real; a fabricated S-1 is not what the record shows. The document is on EDGAR (accession 0001628280-26-067168, file 333-299633). Blanks for price range, share count, September revenue, October run rate, and the aggregate commitment total are ordinary in an initial S-1 filed weeks after quarter-end. An accounting professor told WIRED there are practical reasons the September close is not in the first public draft. Gary Marcus’s claim that blanking Q3 means “demand has dropped” and that the company is skirting the truth overstates the omission: the filing does give the 86-times multiple, and investor-meeting reporting of $16.8 billion is directionally the same number.

The unreliable layer is secondary arithmetic and pre-filing leaks that keep circulating as if they were the prospectus:

• Draft Microsoft commitment figures near $31 billion became at least $44.9 billion in the public filing. Draft “two customers are a fifth of revenue” does not match the nine-month concentration in the S-1.
• Chinese-language recaps have multiplied dollar figures by ten (H1 revenue rendered as $162.86 billion, the net loss as $807 billion). Those are unit errors, not new disclosures.
• OpenAI’s “$70 billion” was an investor gross-up using Anthropic’s method, not an OpenAI print. The company’s own figure to investors was nearer $50 billion net.
• Sacra’s “~$70 billion by October, growth almost stalled” and SemiAnalysis’s “~$74 billion annualized from the Q3 print” are outside estimates. The S-1 does not contain an October run rate.
• Social posts that treat the $80.7 billion loss as cash burn, or the 85% margin as profit after all compute, are misreading lines the filing separates.

Claims least reliable, and the evidence that would settle them

1. Any specific October or year-end run rate ($65 billion, $70 billion, $74 billion, $100–110 billion). Least reliable because the defined metric is blank and prior prints are point-in-time, gross, and uncontracted. Resolve with the amended S-1’s Run-Rate Revenue as of the stated measurement date, plus the monthly consumption bridge, and the reviewed Q3 and Q4 income statements.

2. A single compute-commitment total ($417 billion, $518 billion, $637 billion, $750–810 billion). These add different populations (minimums vs caps vs unsigned estimates vs post-balance-sheet deals). Resolve with the filled-in “aggregate committed spend of more than $____” and the contractual-obligations table, split into non-cancellable minimums, usage-based, and cancelable (especially SpaceXAI after the 2027 notice windows).

3. “85% gross margin” as the economic margin on serving. It excludes most of the compute bill by allocation. Resolve with a quarterly split of compute and infrastructure among cost of revenue, R&D, and sales, and with revenue per dollar of inference compute versus training compute—neither of which the first S-1 fully gives.

4. Apples-to-apples lead over OpenAI on run rate. Anthropic is gross on a channel that was 48% of H1 revenue; OpenAI has described itself as net on Microsoft-routed sales. Resolve with a quantified marketplace-fee line (H1 2026 platform fees, not the 2025 Reuters estimate) and, eventually, OpenAI’s own S-1 principal-versus-agent footnote.

5. 900% net revenue retention as proof of locked, high-quality expansion. The cohort math on a tiny year-ago base, the exclusion of fixed consumer plans, and the 5–7% policy-violator slice all inflate the quality read. Resolve with NRR ex-distillation and ex-top-five, the dollar revenue subsequently reversed or foregone after the August–September enforcement, and the share of consumption under minimum-commit contracts versus pure pay-as-you-go.

6. Q3 as either a collapse or a clean $16.8 billion print. The multiple and the investor-meeting figure agree within rounding; neither is a reviewed statement, and adjusted profit for that quarter was described to investors without a published reconciliation. Resolve with the amended financial statements for the quarter ended September 30, including the GAAP-to-adjusted bridge and whether marketplace fees and violator revenue are called out.

7. Claims that the S-1 was fabricated, secretly rewrote Q3, or hid a demand crash by leaving blanks. The filing is public, the Q3 multiple is in the business section, and blank forward items are standard. Resolve by comparing the October 26 S-1 to the next amendment on EDGAR, not by treating an empty field as a negative disclosure.

---

Recent Findings Supplement (October 2026)

Anthropic filed its S-1 on or around October 26, 2026, disclosing H1 2026 (ended June 30) results that clarify several headline distortions while introducing new specifics on accounting treatment, commitments, and risks.[1][2]

The filing shows $16.286 billion in revenue for the period (up ~1,300% YoY from $1.163 billion), with Q2 alone at ~$11.555 billion; full-year 2025 revenue was $4.585 billion. A net loss of $80.7 billion included a $72.4 billion non-cash charge tied to fair-value adjustments on convertible notes, derivatives, and SAFEs as the company’s valuation rose. Operating results improved markedly, with Q2 delivering the first reported operating profit of $974 million (H1 operating loss narrowed to $693 million).[3][4]

Marketplace sales (primarily AWS Bedrock, Google Vertex AI, and Azure) accounted for 48% of H1 2026 revenue (up from 47% in 2025, 32% in 2024, and 11% in 2023), with 58% of receivables also flowing through these channels.[1][5]

Anthropic acts as the principal in these arrangements, maintaining direct customer contracts, control over the service, and pricing authority. It therefore recognizes the full customer payment as gross revenue; cloud partners bill, collect, and remit net of their fees (treated as a sales/marketing expense). Channel partner fees drove most of the rise in sales, marketing, and partnerships expense to $3.168 billion in H1 2026 (including a $2.7 billion increase).[6][1]

• This gross recognition contrasts with reports that OpenAI books marketplace sales net of partner cuts, making direct topline comparisons misleading.[7]
• Consumption-based (token-metered) revenue comprised ~80% of H1 2026 totals ($13.1 billion), with subscriptions at ~20%.[2]
• Net revenue retention reached ~900% as of September 30, 2026 (enterprise customers spending ~9x year-over-year on average).[3]

This structure inflates reported revenue relative to net-of-fees peers and highlights dependency on a small number of cloud partners that simultaneously supply critical (and rationed) compute capacity, creating potential conflicts.[8]

For competitors or new entrants, the gross-book treatment rewards scale through marketplaces but exposes margins to rising channel fees (already pressuring sales expenses to 19% of revenue). Success requires either bypassing these channels or negotiating better terms while matching Anthropic’s consumption-based model.

Customer concentration has moderated but remains a noted risk, with top-five customers accounting for ~25% of revenue in the first nine months of 2026 (no single customer >10%); many large accounts lack long-term contracts.[1][2]

The customer base expanded rapidly: >11,900 customers with ≥$100k trailing-twelve-month (TTM) revenue as of September 30, 2026 (vs. >950 a year earlier); >1,900 with ≥$1 million; and ~250 with ≥$10 million. U.S. billing addresses accounted for ~72% of H1 revenue.[3][2]

• Revenue from customers outside the top five rose to ~79% in Q3 2026 (from 64% in Q1).[3]
• The filing flags that top customers can reduce or discontinue spending at any time.

Broadening adoption reduces single-customer risk compared with earlier periods, but the high NRR and consumption model mean revenue remains sensitive to usage patterns among a still-concentrated set of large enterprise and partner-driven accounts.[2]

Entrants must demonstrate sticky, diversified usage to compete on valuation multiples; heavy reliance on a few hyperscalers or large logos invites the same concentration scrutiny.

Compute and infrastructure commitments have ballooned, with non-cancelable or minimum obligations reaching hundreds of billions—far outpacing current revenue and creating long-term margin and liquidity pressure.[9]

At June 30, 2026, the filing disclosed $265.6 billion in future purchase commitments and $321.4 billion in leases not yet commenced. Post-June 30 additions totaled ~$245 billion (some conditional); earlier aggregates referenced ~$417 billion in minimum obligations from late 2025 onward, with further estimates pushing totals toward $542–567 billion or higher when including Broadcom (~$161 billion), Google (~$114.5 billion+), AWS (~$110 billion+), Microsoft (~$44.9 billion+), and others. Many contracts require payment “regardless of actual usage.”[10][11]

Compute/infrastructure expense fell sharply as a percentage of revenue (from 201% in H1 2025 to 55% in H1 2026, or ~48% in Q2), enabling the Q2 operating profit despite absolute spend rising to ~$8.9–9 billion for the half.[2][9]

• The S-1 leaves some aggregate post-September 30 figures blank or estimated, and sources vary in how they sum floors vs. caps or post-period amendments.[11]
• Partners have rationed capacity below requested levels in some cases.[8]

These fixed or minimum commitments lock in massive future cash outflows (potentially >$800 billion across disclosed and estimated deals) while revenue depends on filling that capacity with usage.[12]

New entrants face a high barrier: either secure comparable long-term deals (risking overcommitment) or rely on spot/short-term capacity at higher unit costs, compressing margins until scale is achieved.

Recent model releases delivered substantial cost-per-task reductions (up to 45% for certain workloads with September 2026 updates) while the company claims margins were preserved, amid broader price competition from Chinese rivals and OpenAI discounts.[13]

Open-weight and lower-cost models (e.g., from Moonshot, DeepSeek) are cited as sources of pricing pressure; Anthropic notes it may need to adjust pricing or shift to task-based models. Token prices have declined ~50% year-over-year in some datasets due to efficiency gains, cache hits, and competitive cuts.[14]

• Gross margins (calculated) reached ~85% in H1 2026 vs. much lower prior periods.[15]
• Channel fees and rising absolute compute spend still create margin headwinds even as percentages improve.

Efficiency gains allow price cuts without immediate margin collapse, but sustained competition risks eroding the premium pricing that supports high gross margins and the ability to service compute commitments.[16]

Competitors can gain share by undercutting on cost-per-task or offering deeper discounts, forcing Anthropic (and peers) to accelerate efficiency or accept lower margins to retain volume.

Headline net losses are dominated by non-cash valuation adjustments rather than operations, and some pre-filing draft figures or extrapolations created minor discrepancies across outlets; no evidence of fabricated S-1 claims emerged.[4]

The $80.7 billion H1 net loss was driven overwhelmingly by the $72.4 billion fair-value charge. Earlier leaked or draft reports referenced different 2025 loss figures (~$42 billion net vs. ~$8 billion operating) or customer concentration shares. The public S-1 includes blanks for certain Q3 or aggregate figures and leaves some commitment reconciliations to footnotes or estimates.[17][18]

• Revenue figures (~$16.3 billion H1) have been consistent across WSJ, Reuters, Bloomberg, and direct analyses.[19]
• Run-rate extrapolations (e.g., from Q2 or NRR) appear in commentary but are caveated by the consumption-based, usage-dependent nature of revenue.

The least reliable claims (based on available post-April 2026 reporting) are:

• Precise aggregate compute commitment totals (e.g., “$500B+” or “~$800B”) — vary by inclusion of estimates, post-period adds, and conditional vs. firm obligations; primary evidence needed is the full detailed schedule and any amended S-1 reconciliations.
• Direct revenue or margin comparisons to OpenAI (or peers) without adjusting for gross-vs-net marketplace treatment — distorts scale; primary evidence is side-by-side audited revenue recognition policies.
• Run-rate or annualized extrapolations implying sustained hyper-growth without usage caveats — sensitive to customer concentration and competition; primary evidence is full-year 2026 audited results and Q3/Q4 usage trends.
• Pre-filing draft specifics on customer concentration or losses — superseded by the public S-1; primary evidence is the final filed document and any SEC comments/amendments.

These elements underscore that while operational metrics show rapid scaling and improving efficiency, structural dependencies on cloud partners, massive fixed commitments, and competitive pricing dynamics introduce material risks not fully captured in headline revenue or growth rates. Updated filings or earnings will be key to resolving remaining gaps.