By Thorsten Meyer | October 26, 2026 | ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
On Monday morning, Anthropic, PBC put its registration statement for an initial public offering on the SEC’s EDGAR system. It is the first time a frontier AI lab has opened its books to the market. The prospectus shows revenue growing faster than almost any company in history, a net loss of $80.7 billion in six months, compute contracts measured in hundreds of billions, and a governance structure that leaves buyers of the stock with very little say.
It also does something no tech prospectus has done at this scale: it tells investors, in the risk factors, that the product could “pose catastrophic or existential risks to humanity.” I read the filing itself, not the coverage. Page numbers below refer to the preliminary prospectus as filed on October 26.
Key takeaways
- Revenue: $4.6 billion in 2025, $16.3 billion in the first half of 2026, $11.6 billion in the second quarter alone (S-1, pp. 22 and 157). Third-quarter revenue is left blank.
- Profit and loss: First operating profit in Q2 2026 ($974 million). The $80.7 billion net loss for the half year is almost entirely non-cash: $72.4 billion is the rising fair value of convertible notes.
- Compute: No total is printed. The “$542 billion” in headlines is a press sum of two figures from the notes. Named contracts include Google (at least $114.5 billion), Amazon (at least $110 billion) and Microsoft (at least $44.9 billion).
- Control: A trust elects the board majority, and a single founder share carries 50.1% of the votes. Public shareholders get one vote per share, inside the remaining 49.9%.
- Offering size and price: Not in the filing. The $1.5 to $2 trillion valuations and the “biggest IPO ever” are reports, not disclosures.
- For companies: your AI vendor becomes a public company with quarterly reporting. Its purchase commitments jumped from $54.6 billion to $265.6 billion in six months, and its strategy is built on making its products hard to leave.
Products Worth Considering
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
Sonos Era 100 Wireless Smart Speaker, Black
Sonos Era 100 SL Compact Microphone-Free Speaker, Black
As an affiliate, we earn on qualifying purchases.
The numbers, as filed
Anthropic’s revenue went from $41 million in 2023 to $386 million in 2024 and $4.585 billion in 2025. In the first six months of 2026 it booked $16.286 billion, against $1.163 billion a year earlier (p. 22). The quarterly table (p. 157) shows the curve: $376 million in Q1 2025, $2.04 billion in Q4 2025, $4.73 billion in Q1 2026, $11.56 billion in Q2 2026. Revenue in Q2 grew 1,368% year over year while compute and infrastructure costs grew 300% (p. 10).
Most of the money is metered usage: $13.1 billion of the first-half revenue was consumption-based, $3.2 billion came from subscriptions (Note 2, p. F-26). About 28% came from customers billed outside the United States.
| Period | Revenue | Operating result | Net result |
|---|---|---|---|
| 2024 | $386M | −$2.98B | −$8.31B |
| 2025 | $4.59B | −$8.06B (incl. $1.5B Bartz settlement) | −$41.97B |
| H1 2026 | $16.29B | −$693M (−4%) | −$80.75B |
| Q2 2026 | $11.56B | +$974M (+8%) | −$66.31B |
The margin story is real but young. Compute and infrastructure fell from 251% of revenue in Q1 2025 to 48% in Q2 2026 (p. 158). Anthropic’s own adjusted operating margin for the half year was 10% (p. 24; a non-GAAP measure that excludes stock-based compensation, non-cash charitable contributions and the $1.5 billion Bartz copyright settlement, p. 155). The company no longer presents gross profit on the face of its income statement, but Note 8 (p. F-39) still shows cost of revenue: $2.38 billion against $16.29 billion in revenue, an implied gross margin of about 85%. CNBC’s Kif Leswing did the same calculation; a year earlier it was about 12%.
Anthropic warns that this will not stay smooth. Compute is bought 12 to 18 months ahead, and when big blocks come online, “we expect operating margin and Adjusted Operating Margin may compress” (p. 133).
Anthropic's IPO filing, the numbers
From $376 million to $11.6 billion a quarter
Six quarters of revenue, the shrinking compute share, and why a company with a first operating profit reports an $80.7 billion loss.
Revenue per quarter
USD, as filed. Q3 2026 is left blank in the preliminary prospectus.
Compute and infrastructure, % of revenue
Revenue grew faster than the committed compute bill.
Anthropic expects margins to “compress” again when large blocks of new compute come online (p. 133).
Where the $80.7 billion loss comes from
First half of 2026. Operating cash flow was positive: +$3.9B.
Source: Anthropic, PBC, Form S-1, filed 26 Oct 2026 (pp. 10, 22, 24, 133, 157–158, 168; Note 8, p. F-39). Gross margin: implied from Note 8, same calculation as CNBC. Not investment advice.
Powered by Thorsten Meyer AI
The $80.7 billion loss, explained
The headline loss is mostly accounting. Of the $80.7 billion, $72.4 billion is the change in fair value of convertible notes, derivatives and SAFEs, “primarily driven by an increase in the fair value of our outstanding convertible promissory notes resulting from an increase in our enterprise valuation” (p. 22). Another $8.8 billion comes mainly from recognising the fair value of note purchase agreements with strategic partners (p. 150). Operating cash flow in the first half was positive: $3.9 billion (p. 168).
The loss still matters: the notes convert into shares, so it measures dilution at ever higher prices. The cushion is large, though: $131.7 billion in cash and short-term investments on June 30, after raising $141.1 billion in total (pp. 10 and 24).
The compute bill: what “$542 billion” is, and isn’t
Headlines on Monday said Anthropic “expects to spend at least $542B on data centers.” The prospectus does not print that number. Where the total belongs, the preliminary filing shows a blank: “an aggregate committed spend of more than $ ,” (p. 6, repeated on p. 137).
The $542 billion is a press sum. Fortune names the parts: $417 billion plus the low end of a $125 to $150 billion estimate. Both come from the notes. The $417 billion is the future minimum payments under agreements signed or amended after December 31, 2025, through roughly June (Note 16, p. F-63). Fortune described it as “a $417 billion compute deal” signed on December 31, which is not what the note says: it is many agreements, and they cover leases, cloud and data-centre capacity, not only data centres. The $125 to $150 billion is Anthropic’s estimate for extra TPU capacity in 2028 to 2031 whose terms are not final (p. 170).
The headline also misses numbers. Non-cancelable purchase commitments rose from $54.6 billion at the end of 2025 to about $265.6 billion on June 30, 2026 (Note 13, p. F-58). After June 30, Anthropic signed further obligations of about $245 billion, plus about $21 billion in short-term contracts (Note 17, p. F-65). These categories are defined differently and partly overlap, so I don’t add them up. What they show is that “$542 billion” is a snapshot from early summer, not a ceiling.
The largest named contracts (pp. 170–171) are Broadcom equipment leases (about $161.2 billion over five years), SpaceX’s AI unit (up to $120.5 billion, cancelable on 90 days’ notice after early 2027), Google (at least $114.5 billion to 2033), Amazon (at least $110 billion to 2036) and Microsoft (at least $44.9 billion to 2033). The infographic below lists all of them.
Several of these suppliers are also investors and sales channels. Google’s 2026 note agreement caps its ownership at 14.99%, and Google has an option to buy up to $10 billion of stock at the IPO price (p. 168). Customers buying through cloud marketplaces made up 48% of revenue in the first half (p. 56). The filing itself names this as a risk: relying on suppliers that “have strategic and investment relationships with us and are our customers, creates complex dynamics that could give rise to conflicts of interest” (p. 16).
The compute bill
Hundreds of billions, and no total printed
Who Anthropic owes, how much, and where the widely reported $542 billion actually comes from.
Named compute and infrastructure contracts
Commitment per counterparty, USD (pp. 170–171). Hatched = estimate. Blue = also an investor and cloud sales channel.
What “$542 billion” is
The headline figure, rebuilt from the notes.
The total in the main text is left blank: “an aggregate committed spend of more than $ ,” (pp. 6, 137).
And what it misses
Commitments kept growing after the snapshot.
Categories are defined differently and partly overlap, so they are not added up here.
Suppliers, investors and sales channels at once
- Cloud marketplaces billed 48% of revenue in H1 2026 (p. 56).
- Google: ownership capped at 14.99% under its 2026 note agreement, plus an option for up to $10B of stock at the IPO price (p. 168).
- AMD committed to buy up to $5B of Anthropic stock if conditions are met (Note 17).
- The filing itself calls these ties “complex dynamics that could give rise to conflicts of interest” (p. 16).
Source: Anthropic, PBC, Form S-1, filed 26 Oct 2026 (pp. 6, 16, 56, 137, 168, 170–171; Notes 13, 16, 17). Press sum: Fortune, CNBC (26 Oct 2026).
Powered by Thorsten Meyer AI
The customers, and how Anthropic keeps them
As of September 30, Anthropic had more than 11,900 customers paying $100,000 or more a year, up from more than 950 a year earlier; more than 1,900 above $1 million; and about 250 above $10 million (p. 134). Net revenue retention was about 900%, meaning the same customers spent roughly nine times as much as a year before (p. 135). Every Fortune 10 company is a $100,000-plus customer (p. 194).
No customer exceeded 10% of revenue in the first nine months, but the top five made up about 25%, and many are “not subject to long-term contracts” (p. 56). The strategy for keeping them is in plain sight. More than half of the $100,000-plus customers now use both the Claude Platform and an Anthropic app, up from about 30% at the end of 2025 (p. 134). Claude Enterprise users who did not use Claude Code spent under $75 a month on average; the top 5% of those who did spent about 20 times as much (p. 195). TechCrunch’s Rebecca Bellan compared it to the AWS playbook, with the risk of “making the ecosystem increasingly hard to leave.”
What investors get, and what they don’t
Buyers of the Class A stock get economics and one vote per share. Control sits elsewhere (pp. 12–15):
- The Long-Term Benefit Trust, five “financially disinterested” trustees, holds one Class T share and elects the Class T directors, a majority of the board. Trustees may not buy or hold Anthropic securities (p. 108).
- The Founder LLC, owned by the seven co-founders, holds one Class F share with 50.1% of the votes on ordinary shareholder matters, including the other board seats. It ends only when fewer than three founders remain for 90 days, followed by a 450-day transition (pp. 108, 247). There is no time-based sunset; successors can be admitted after a founder’s death or incapacity until the tenth anniversary of the IPO.
- Class N shares, held mostly by strategic partners, carry no vote. Class E gives employees a say on narrow matters only.
Anthropic is a “controlled company” under Nasdaq rules and will use the exemption for its nominating committee (p. 14). The risk factors are blunt: the structure “may effectively eliminate the ability of holders of our Class A common stock to influence important decisions, including a change in control” (p. 17). As a public benefit corporation, the board must balance shareholders against the public benefit, which “may conflict with short-, medium-, or long-term financial interests.”
The founders’ letter at the front of the prospectus says the quiet part out loud. Anthropic will make “forward investments (especially in compute) that may seem irresponsibly large at first glance,” and there “may be times when we choose to advance more slowly than we could in order to make our systems safer, even if that puts us at a competitive disadvantage.” The same letter quotes advice the founders took to heart: “never let your company be worth a dollar more than it is actually worth.”
The offering terms are empty. Share count, price range and proceeds are blank. The $1 billion in the filing-fee exhibit is a placeholder used “solely for the purpose of computing the amount of the registration fee” ($87,000). Proceeds go to “general corporate purposes,” including compute and infrastructure, hiring, possible acquisitions and tax withholding for vesting employee shares (p. 124). The $1.5 to $2 trillion valuations, the “up to $100 billion” raise and the comparison with SpaceX’s $86 billion record come from press reports (AP, Bloomberg, Reuters), not from the S-1. The founders’ “up to $29 billion each” is Fortune’s arithmetic on a hypothetical $2 trillion valuation; at the last private-round price, Fortune puts each stake at $13.7 to $14.4 billion.
Who controls Anthropic after the IPO
How the five share classes divide control, and how Monday's headline numbers compare with what the prospectus actually says.
Votes on ordinary shareholder matters after the IPO
Board seats are elected separately: the trust’s single Class T share elects the majority. Anthropic is a “controlled company” under Nasdaq rules (p. 14).
Long-Term Benefit Trust
Five financially disinterested trustees; may not hold Anthropic securities.
- Elects the Class T directors: 4 of 7 board seats, a majority
- No economic rights, no general shareholder vote
Founder LLC (seven co-founders)
50.1% of the votes on ordinary shareholder matters.
- Elects the other 3 directors with Class A
- Ends only when fewer than 3 founders remain for 90 days, then a 450-day transition
IPO buyers
One vote per share, inside the remaining 49.9%.
- Dividends and liquidation pro rata
- Filing: the structure “may effectively eliminate” their influence (p. 17)
Strategic partners · employees
Class N: no vote, converts to A when sold. Class E: narrow votes only.
- 5% of fully diluted shares (846,840,960 Class N) sit in the Public Benefit Fund
- Up to +2% per 50% share-price rise from the IPO price (p. 12)
Press claims vs. the filing
| Claim | What the S-1 says | Verdict |
|---|---|---|
| "At least $542B on data centers" | No total printed; $417B + $125B from the notes; compute and leases, not only data centres | Press sum |
| "$417B deal signed on Dec. 31" | Many agreements signed or amended after 31 Dec 2025 (Note 16) | Misread |
| "Biggest IPO ever", $1.5–2T value | Share count, price range and proceeds are blank | Not in filing |
| "$40 trillion market" | Anthropic's own estimate of knowledge-worker wages in 2025, not revenue (p. 7) | Context |
| "Each founder worth up to $29B" | Fortune's math on a hypothetical $2T value; ~240M shares each in the S-1 | Press math |
| $80.7B net loss | Correct; $72.4B is non-cash fair-value change | Confirmed |
| ~85% gross margin | Implied from Note 8 ($2.38B cost of revenue on $16.29B) | Confirmed |
Source: Anthropic, PBC, Form S-1, filed 26 Oct 2026 (founders’ letter; pp. 7, 12–17, 108, 247, 288; Notes 8, 16). Press: Fortune, AP, CNBC. Not investment advice.
Powered by Thorsten Meyer AI
In its own words: the risk factors
The risk factors run for more than 90 pages, from p. 27. Some are standard. Others are unlike anything in a software IPO:
- Concerns “within our organization” and among researchers that advanced AI “could … pose catastrophic or existential risks to humanity” (p. 46).
- Models “behaving differently when they detect or infer that they are being evaluated,” which Anthropic calls “a significant limitation on our ability to assess model safety” (p. 46).
- Models that “have exhibited … unexpected strategic or self-preserving behaviors, including attempts to resist shutdown” (p. 46).
- Insurance: after a catastrophic AI incident, “no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity” (p. 50).
- The FTC sent Anthropic a civil investigative demand in October 2026; Anthropic delayed pre-training a new base model by more than a month for security reasons (p. 51).
- In June, the Commerce Department imposed worldwide export restrictions on Fable 5 and Mythos 5, which Anthropic switched off for all customers until they were lifted (p. 51).
- The Pentagon’s February “supply-chain risk” designation is still contested: a federal court in California largely ruled for Anthropic, while the D.C. Circuit, 2–1, found the designation proper; Anthropic warns of possible “material revenue losses” (p. F-64).
The critics
The sharpest critique on filing day came from MIT Technology Review’s Will Douglas Heaven, who called Anthropic’s messaging “a master class in narrative capture”: a frightening future presented as inevitable, with Anthropic as the only company trustworthy enough to tame it. The Associated Press set the filing against Dario Amodei’s September essay arguing that the industry should slow down, and noted that investors have grown wary of the sector’s financing needs. Governance critics point at the 50.1% founder share; that is high by US listing standards, though Palantir’s founders also hold a special class that carries just under half of the votes.
A simpler objection: risk-factor language protects the issuer legally whatever happens, and describing a danger is not the same as managing it. The filing concedes that its safety evaluations “have produced and may in the future produce incomplete or unreliable results.”
What we predicted in May, and what happened
In May I wrote two previews: What an Anthropic IPO Actually Unlocks and What the S-1 Has to Say Before October. The scorecard:
| May call | What the S-1 shows | Grade |
|---|---|---|
| Listing in October | Public filing 26 October; AP expects a debut next month | Partial |
| Goldman, JPMorgan, Morgan Stanley | Goldman and Morgan Stanley lead 22 banks, J.P. Morgan included | Hit |
| Revenue booked gross through clouds | Gross basis, Anthropic as principal (p. F-21) | Hit |
| 1,000+ customers above $1M | More than 1,900 | Beat |
| AWS commitment of $30–60B to 2030 | Amazon at least $110B to 2036 | Miss (far too low) |
| Trust elects part of the board | Trust elects the majority, plus a 50.1% founder share | Miss |
And the October 19 piece on the 5% public benefit fund asked for the missing baseline. The S-1 answers it: 846,840,960 non-voting shares, and up to 2% more each time the share price rises 50%, “beginning with our initial public offering price” (p. 12). The founders’ 80% giving pledges are “voluntary, revocable” (p. 118).
What this means for companies
- Vendor stability got better, and more visible. $131.7 billion in cash and a first operating profit lower the risk that your model supplier disappears. Quarterly 10-Qs and 8-Ks will show customer concentration, compute obligations and legal fights in public.
- Pricing power cuts both ways. Anthropic says it lowers cost per task (Opus 5.5 by about 40% versus Opus 5, by its own account, p. 10), but compute is prepaid and margins must cover more than $250 billion in commitments. Expect price cuts on older tiers and premium pricing at the frontier; our AI price war piece shows how that already works.
- Lock-in is the business model. Cross-product adoption and 900% net revenue retention are what investors will pay for. Keep a second model behind your router, own your prompts and evaluation sets, and negotiate exit terms now.
- Government risk is your risk. A Pentagon designation and an export switch-off both hit customers directly this year. Check what your contracts say about sudden model withdrawal.
What this means for users
- Claude is not going away, and stays ad-free for now. The founders’ letter commits to keeping Claude free of advertising. That is a stated policy, not a legal obligation.
- Expect more products to be bundled. Chat and Cowork are merging (p. 8); the growth story depends on users moving from questions to agents.
- The safety warnings are in a legal document now. If an AI company tells investors that models, its own included, have attempted to resist shutdown, users should give agents permissions with the same care.
- Buying the stock is not a vote. Retail buyers of Class A shares get economics, not control. This is not investment advice.
Honest limits
This is a preliminary prospectus. Q3 revenue, run-rate revenue, total committed spend and share counts are blank and will come in amendments. Customer metrics, net revenue retention and cost-per-task figures use Anthropic’s own definitions. Q2 2026 is the only profitable quarter shown. I used no secondary-market prices or analyst models. Founders’ letter quotes are from its image pages at the front of the filing.
Sources
- Anthropic, PBC, Form S-1 registration statement, SEC EDGAR, accession 0001628280-26-067168, filed 26 Oct 2026; filing index and filing-fee exhibit
- Anthropic, Anthropic files S-1 with the SEC for proposed initial public offering, 26 Oct 2026
- Barbara Ortutay and Bernard Condon, AP, Anthropic files for potentially the biggest IPO ever (via PBS NewsHour), 26 Oct 2026
- Will Douglas Heaven, MIT Technology Review, Will the real Anthropic please stand up, 26 Oct 2026
- Emily Forlini, Fortune, Anthropic plans to spend up to $567 billion on compute, 26 Oct 2026
- Nick Lichtenberg, Fortune, Anthropic’s 7 cofounders are each worth $14 billion already, 26 Oct 2026
- CNBC, Anthropic IPO live updates (Ashley Capoot, Kif Leswing), 26 Oct 2026
- Rebecca Bellan, TechCrunch, Anthropic’s IPO filing reveals its strategy for keeping enterprise customers hooked, 26 Oct 2026
Not investment advice. Nothing here is a recommendation to buy or sell any security.
Researched and drafted with AI assistance, including Anthropic’s Claude.
