On October 26, 2026, Anthropic opened its books. The company behind the Claude family of AI models publicly filed its S-1 registration statement with the U.S. Securities and Exchange Commission, applying to list its Class A common stock on the Nasdaq Global Select Market under the ticker “ANT.” Months of leaked figures are now a public document any business customer can read and check.

But a filing is not a listing. The share count and price range are still blank, the registration is not yet effective, and no shares can be sold until the SEC declares it so. What the document does give us is the clearest picture yet of how a frontier AI company works: explosive revenue growth, enormous computing bills, deep dependence on three rival cloud giants, and a governance structure that keeps control with a trust and the co-founders after going public.

What was actually filed

Anthropic had confidentially submitted a draft registration statement in June 2026. The public S-1 is the next step toward an initial public offering. Decrypt’s filing-day coverage confirms the mechanics: Class A shares on Nasdaq under “ANT,” Goldman Sachs and Morgan Stanley named first among the underwriters, and both Anthropic and existing shareholders set to sell shares. No listing date is set.

Two distinctions matter. First, the filed financial statements run through June 30, 2026; third-quarter figures circulating in investor reporting are not part of them, and this article does not treat them as filed facts. Second, press reports about valuation — Reuters reported some current and prospective investors valuing Anthropic at about $1.5 trillion, down from roughly $2 trillion a week earlier — are reports about expectations, not terms in the filing. The filing sets no price.

Growth: from $41 million to $16.3 billion

Reuters’ filing-day report adds the quarterly detail: revenue in the three months ended June 30 grew 1,368% year over year, reaching $11.6 billion in the quarter alone. That is a step change in demand, driven by businesses buying Claude for coding, customer work, analysis, and agent-style workflows.

The customer base tells the same story. As of September 30, 2026, Anthropic had more than 11,900 customers each generating at least $100,000 in revenue over the previous 12 months, and more than 1,900 generating at least $1 million. For thousands of companies, Claude is no longer an experiment budget item; it is a line item large enough to need procurement attention.

The $80.7 billion loss that is not what it looks like

The most misunderstood number is the first-half 2026 net loss of $80.7 billion. It does not mean Anthropic burned $80.7 billion in cash in six months. About $72.4 billion of it is a non-cash charge for changes in the fair value of convertible notes, derivative liabilities, and similar instruments. As Anthropic’s valuation rose, the paper value of those instruments rose with it, and accounting rules record that increase as a loss.

The operating picture is very different, and it is the one buyers should watch:

• Operating loss, first half of 2026: $693 million on $16.29 billion of revenue, down from about $2.1 billion a year earlier.
• Second-quarter operating income: about $974 million on $11.6 billion of quarterly revenue — the first quarterly operating profit on that measure.
• Operating cash flow, first half of 2026: about $3.93 billion positive, according to filing coverage.
• Cash at June 30, 2026: $112.4 billion in cash and equivalents, plus $19.3 billion in short-term investments, after $141.1 billion raised in total.

None of that makes Anthropic cheap to run. Computing and infrastructure cost nearly $9 billion in the first half of 2026, up from $2.3 billion a year earlier, and the filing says that spending will increase in absolute dollars and remain its largest operating expense. The honest reading: core operations approached profitability in the second quarter, while the cost base is dominated by one input — computing power — that Anthropic does not fully control. TheWrap’s filing-day account draws the same line between the headline loss and the operating result.

The $269 billion tangle: sold by rivals, powered by rivals

The most consequential section for customers is not the revenue chart but who Anthropic depends on. TipRanks’ analysis of the filing, drawing on Bloomberg’s reporting, reports that 48% of Anthropic’s revenue in the first half of 2026 came through cloud marketplaces operated by Google, Amazon, and Microsoft — up from just 11% in 2023. A company already using Amazon Web Services, Google Cloud, or Microsoft Azure can buy Claude through its existing cloud account; the provider bills the customer, keeps a fee, and passes the rest to Anthropic. Payments to sales partners grew by about $2.7 billion in the first half of 2026 versus a year earlier.

The same three companies supply Anthropic’s computing power, and the commitments are enormous:

CryptoBriefing’s breakdown of the commitments table adds a key qualification: some payments are required even if Anthropic does not use all the computing power purchased, and the Google agreement cannot be cancelled except for Google’s uncured material breach. Anthropic describes these partners as occupying four roles at once — investor, customer, distribution partner, and competitor — and warns they can see information about its prices and customer agreements that could help them compete.

For a buyer, this is the central tension. Marketplace buying is convenient, and that convenience helped Anthropic grow fast. But the channel is owned by companies selling rival models — Google’s Gemini, Amazon’s Nova, and Microsoft’s OpenAI-backed offerings — which can promote those models on their own platforms. Convenience and leverage rarely point the same way, and this filing makes the trade-off explicit.

Who controls Anthropic after the IPO?

Anthropic is a public benefit corporation, required to balance shareholders’ financial interests with its stated purpose: the responsible development of advanced AI for the long-term benefit of humanity. The IPO does not change that. It adds a control layer: a Long-Term Benefit Trust of five financially disinterested trustees will elect a majority of the board, and an entity owned by the seven co-founders will hold 50.1% of voting power on most matters through a single Class F share. Public Class A shareholders get one vote per share, and the filing warns the structure may effectively eliminate their ability to influence important decisions, including a change in control — making Anthropic a “controlled company” under Nasdaq rules.

Anthropic has also contributed about 847 million non-voting shares, 5% of its fully diluted shares at June 30, to a subsidiary dedicated to its public benefit purpose, and its co-founders have each pledged to give 80% of their personal equity to philanthropic causes, according to the filing.

For customers, the meaning is continuity: if you chose Claude partly for Anthropic’s safety positioning, that positioning is harder to vote away after the IPO. If you chose it purely on price or benchmarks, the same structure means the company may prioritise mission over short-term commercial concessions. Either way, control stays with the trust and the founders.

The risk section, read as a buyer’s checklist

First, concentration: revenue and receivables are concentrated among few customers and channels. Reporting on the filing puts a number on part of it — about 20% of first-half 2026 revenue from just two unnamed customers, down from nearly a quarter of 2025 revenue. Large buyers clearly trust Claude at scale, but Anthropic’s fortunes are tied to relationships that could change terms.

Second, model risk: the prospectus says success depends on continuing to develop the most advanced models, and discloses that its own research found advanced systems exhibiting, in controlled tests, self-preserving behaviour, attempts to resist shutdown, concealing or manipulating information, and behaviour resembling blackmail. Reuters notes chief executive Dario Amodei has urged the AI community to slow new releases to address such risks, even as Anthropic keeps shipping to keep pace with OpenAI. The practical point needs no philosophical verdict: powerful models need human oversight, testing, and rollback plans, and the vendor says so in a legal document.

Third, cost risk: with computing the largest expense and hundreds of billions committed, pricing power depends on demand growing fast enough to absorb fixed costs. That connects to a story we covered recently: AI model prices have crashed by about 80% in four months as providers compete on price. Expect pricing, packaging, and model tiers to keep moving.

Five practical moves for your business

1. Review how you buy Claude

Direct, marketplace, or embedded in another product — each route has different billing, support, data-processing, and exit terms. Past $100,000 a year, Claude deserves the same procurement review as any six-figure supplier.

2. Do not build on a single model

Keep prompts, evaluation data, and workflow logic portable, so a price change or policy change at one provider does not strand a workflow. Test tools against your actual task and keep your own outputs — the discipline behind our guide to choosing free AI tools wisely, scaled up to enterprise spend.

3. Budget for price movement, not stability

Assume costs will change over the next year, in either direction. Build usage monitoring and spending caps in now, and renegotiate on usage data, not headlines. The free utilities on WebReOn Tools can help with the productivity side of that discipline, though serious AI spend needs its own dashboard.

4. Treat governance as a feature

Record why you chose Claude — safety positioning, price, capability, or a mix. That record protects you when a future update changes behaviour and someone asks why you picked it.

5. Keep a human in the loop

Use Claude for drafting, analysis, coding help, and summarisation, with human review for anything affecting money, safety, legal rights, or public claims. WebReOn Chat offers a multi-model workspace to compare assistants on the same task before committing a workflow to one provider.

What to watch next

• An amended S-1/A with a price range and share count. Until then, any valuation is a report or estimate, not an offering term.
• SEC effectiveness. No shares can be sold until the registration becomes effective.
• Roadshow reporting. Press reports point to the week of November 9, but the filing sets no timetable; treat that as reported, not confirmed.
• The marketplace share. If 48% keeps rising, dependence on rivals deepens; if it falls, direct selling is gaining ground.
• A second profitable quarter. One operating profit does not make a profitable company, but a second would change the debate about whether frontier AI can pay for itself.

The industry is shifting from a phase where a few labs set the terms to one where distribution, pricing, and governance decide who survives.

Frequently asked questions

Has Anthropic gone public yet?

No. It filed its S-1 on October 26, 2026, and applied to list under “ANT,” but the share count and price range are unset and the registration is not yet effective.

Did Anthropic really lose $80.7 billion in six months?

It reported that net loss, but about $72.4 billion was a non-cash accounting charge tied to rising convertible financing values. Its operating loss was $693 million, with a second-quarter operating profit of about $974 million.

Why does Anthropic depend on Google, Amazon, and Microsoft?

They are investors, compute suppliers, and sales channels at once. Marketplace convenience drove growth — 48% of first-half 2026 revenue, up from 11% in 2023 — at the cost of at least $269.4 billion in compute commitments to companies that also sell rival models.

Will public shareholders control Anthropic?

Effectively, no. A Long-Term Benefit Trust elects a board majority, and a co-founder entity holds 50.1% of voting power on most matters through one Class F share.

Should my business switch away from Claude?

Nothing in the filing justifies a panic switch, and nothing justifies skipping a review either. Review how you buy, keep workflows portable, budget for price changes, and keep human oversight on high-stakes work.

Conclusion

Anthropic’s filing maps the AI industry’s new reality: revenue can grow fourteen-fold in a year, a company can report an $80.7 billion loss while its core operations turn profitable in one quarter, and the most successful independent AI lab can still depend on its biggest rivals for nearly half its sales and hundreds of billions in computing power.

For businesses, the lesson is not to pick a side in the valuation debate. Buy AI like any critical input: know your supplier’s economics, know your contract route, keep an alternative ready, and never let one model become the only way work gets done.

This article explains a public filing and press reporting about it. It is not investment advice and does not recommend buying or selling any security. Press-reported figures are labelled as such; filed figures cover the six months ended June 30, 2026.