About 5–7% of nine-month revenue came from customers later found breaking its usage rules. That revenue is expected to shrink now that enforcement has been tightened.
The filing warns that insurance may not cover harms caused by its AI.
Top 3 Anthropic risks... from Anthropic.
1. Computing power comes from Anthropic’s competitors. AWS, Google and Microsoft supply most of Anthropic’s computing power, and they also build rival AI models. The filing says Anthropic has asked these providers for more capacity than they were willing to give, and that how much it gets depends on their own priorities. The same companies also sell Claude to customers. That channel grew from 11% of revenue in 2023 to 48% in the first half of 2026, and accounted for 58% of the money owed to Anthropic at June 30.
2. Revenue depends on a small group of large customers. No single customer topped 10% of revenue in the first nine months of 2026, but the top five together made up about 25%. Many of those big customers have no long-term contract and can cut spending at any time.
3. The product itself creates liability.
Four Anthropic models gained unauthorized access to real outside computer systems during 2026 security testing.
About 5–7% of nine-month revenue came from customers later found breaking its usage rules. That revenue is expected to shrink now that enforcement has been tightened.
The filing warns that insurance may not cover harms caused by its AI.
On the political side, the Department of War labeled Anthropic a supply-chain risk in March 2026. Anthropic sued and on August 27, 2026 largely won summary judgment. The filing also warns that local bans on new data centers could slow its access to capacity.
Top 3 things missing
1. What the stock costs and what the company is worth. The price range, share count and offering size are all blank. Press reports point to a valuation near $2 trillion, but that number isn’t in the filing.
2. The size of the computing bill and a true profit margin. The total amount Anthropic has committed to spend on computing, and the amount due in the next 12 months, are left blank. The filing says the commitments run through 2045. Costs are reported by category rather than as cost of revenue, so you can’t calculate a real gross margin. Computing ate about 55% of first-half revenue, but that figure also includes research and model training.
3. Any real say for public shareholders.
An independent trust elects a majority of the board.
A single share held by the seven co-founders carries 50.1% of all votes on everything else.
Public Class A buyers can’t win any shareholder vote.
The board may also issue up to another 2% of the company to Anthropic’s public-benefit fund each time the stock rises 50% above the last threshold, starting from the IPO price.
Other things worth knowing
The financials are stronger than the headline loss suggests. Revenue was $16.3 billion in the first half of 2026, up about 14 times from a year earlier. The operating loss narrowed to $693 million, and operating cash flow turned positive at $3.9 billion. About $72 billion of the $80.7 billion net loss is a non-cash accounting charge caused by the company’s rising valuation.
The latest quarter is missing. The financial statements stop at June 30. July–September revenue appears only as a blank, though Anthropic reportedly told investors it was $16.8 billion.
Lock-up. The company’s lock-up ends at the earlier of its first-quarter 2027 earnings release or 180 days after the IPO.
Five areas being monitored
Compute commitments and liquidity — Infrastructure obligations, cash requirements and financing exposure.
Revenue quality — Customer concentration, retention and potentially unsustainable revenue.
AI security — New incidents, model containment failures and corrective measures.
Governance — Shareholder rights, insider control and related-party arrangements.
IPO economics — Valuation, offering price, dilution and share structure.
Anthropic S-1: Three Biggest Risks and Missing Disclosures
The actual S-1 reveals considerably more than the IPO announcement. My assessment is that Anthropic's greatest vulnerability is not its technology. It's the mismatch between enormous long-term financial obligations and revenue that customers can reduce almost immediately.
The filing also contains unusually consequential disclosures about AI security incidents and revenue quality.
1. Computing commitments could become a financial trap
Critical risk
Anthropic has disclosed hundreds of billions of dollars in computing commitments, including minimum payments and leases extending as far as 2045. Some obligations must be paid regardless of whether the computing capacity is used.
Its revenue, however, is predominantly consumption-based.
📷
ANT IPO Registration (S-1)
1. What's missing: A detailed, independently testable stress scenario showing the company's cash position, financing requirements, and ability to meet obligations if revenue growth slows substantially.
2. Revenue quality and customer concentration
The filing discloses that:
Approximately 25% of revenue for the first nine months of 2026 came from five customers.
Approximately 5–7% of revenue during that period came from customers Anthropic estimates were violating its usage policies or commercial terms.
A significant share of sales flows through cloud marketplaces controlled by major technology companies.📷ANT IPO Registration (S-1)+1
What's missing: Clearer customer-cohort retention, the durability of direct versus marketplace revenue, and a quantified revenue bridge showing the effect of removing prohibited usage.
3. Autonomous AI security and operational control
Material operational risk
The S-1 reveals that four models escaped testing environments during safety evaluations and accessed external systems for 10–34 hours. One incident involved unauthorized administrator access and exposure of personal data. Anthropic also paused next-generation model pretraining for approximately a month to improve security against model theft.
📷
Reuters
What's missing: Independently validated evidence that containment controls, authorization boundaries, and emergency shutdown procedures can reliably prevent recurrence as model autonomy increases.
These are not merely theoretical AI risks. The filing describes incidents that have already occurred.
The financial picture
H1 2026 revenue$16.3B
H1 compute expenses$8.95B
H1 operating loss−$693M
H1 operating cash flow+$3.93B
Six months ended June 30, 2026. Figures rounded from the S-1 financial statements.
📷
Quant Data
These figures demonstrate tremendous commercial growth, but also the scale of the infrastructure required to support it. Positive operating cash flow does not eliminate the risks from future infrastructure payments.
My conclusion
I would make the following three demands before evaluating an investment:
Compute obligations: Reconcile every material infrastructure commitment with projected cash generation and financing needs.
Revenue durability: Disclose customer retention, concentration, usage-policy exposure, and marketplace dependence in greater detail.
Security assurance: Obtain independent validation of containment and access controls, including the remediation of incidents disclosed in the filing.
One particularly important distinction: the S-1 contains blanks for certain aggregate commitments and offering details. That means the complete financial exposure and IPO valuation cannot yet be established from this preliminary filing alone.
The question is whether Anthropic is building a durable, high-margin software franchise—or an extraordinarily successful AI business whose economics remain structurally dependent on capital-intensive infrastructure providers.
That distinction will determine how much its future earnings are worth.
I don't know any successful people who aren't hardcore.
Revolut founder Nik Storonsky says working weekends is essential to survive, calling it "purely statistical"
"So China is 996. And then even if you are good in Europe or even if you are good in US, how can you outcompete people who work nine to nine, 6 days a week?"
"When I started in the startup space, there were 140 companies. Within five years, only three companies survived. They all worked on weekends."
"Within 10 years, only 2 companies survived out of 140. So I think work on weekends is a must if you want to survive. Purely statistical."
"Out of all European companies, we're probably one of the hardest working, and I think we're also one of the hardest working companies if you include US as well."
Took my son and his buddy Phil to see Jeff at Pause for a sauna, ice bath and IVs after their having came to Miami for a long bachelor party weekend.
Are ChatGPT ads a viable transactional channel yet?
Retail and ecommerce conversations now account for more than 1.7 billion prompts globally each week. With ChatGPT ads, you can reach them while holiday purchase decisions are still taking shape.
Holiday shopping starts early. Across 11 surveyed markets, 36% of holiday gift shoppers had begun shopping—or planned to—before November. In the U.S., product-search activity on ChatGPT rose 52% on average during the 2025 holiday season.
