Anthropic May Have Earned Up to $2.3 Billion From "Gray" Access to Claude — Up to 7% of Revenue
Vedomosti studied Anthropic's documents prepared for its stock market debut and calculated that from January through September 2026, "gray" access users of Claude brought the company between $1.66 and $2.33 billion — 5–7% of all revenue for the period. These are people who use the model in violation of the service's rules, yet still pay for it. The figures are the publication's estimate, not official company data.
AI-processed from RB.RU; edited by Hamidun News
Rule-breakers are usually discussed as a loss: we block, we ban, we lose money. The new story about Anthropic works the other way around. According to Vedomosti's calculations, based on company documents prepared for its stock market listing, "gray" access to Claude brought the developer between $1.66 and $2.33 billion over the first nine months of 2026.
What we know
The publication estimates that from January through September, users who accessed Claude in violation of the rules provided Anthropic with 5–7% of all year-to-date revenue. We have not seen the documents themselves, so all figures are given as relayed by Vedomosti. Anthropic has not publicly confirmed them.
There is also some simple arithmetic. If $1.66 billion is 5% and $2.33 billion is 7%, total revenue for nine months comes to about $33 billion. This is our calculation from the publication's percentages, not a figure stated by the company. For scale: by this estimate, the gray segment brings in on average $200–260 million per month.
What exactly is meant by "gray" access is not disclosed in the retelling. Usually the term refers to using the service from regions where it is officially unavailable, via intermediaries, other people's accounts, resold API keys, or payment with foreign cards. We do not know the exact makeup of this category in the documents, and it is worth keeping in mind when you read "5–7%."
Why it matters
There are two signals here for the market.
First: demand for top models is not limited by geographic restrictions. If people are willing to pay through workaround schemes, the official offering does not cover the need. Founders will recognize this picture: where a product is limited by policy rather than technology, a market of intermediaries always emerges.
The second signal is subtler. A company preparing for an IPO is obliged to describe its risks. Revenue from those who formally are not allowed to use the service is unstable income. It can vanish after stricter checks, a policy change, or regulatory pressure. Investors will look at how much 5–7% of "revenue with a caveat" could damage the forecast.
What practitioners should do
If you are building a product on top of Claude, there are three questions.
- Where your traffic comes from. If requests go through a proxy or reseller, you depend on someone else's account. It can be blocked at any moment, together with your service.
- What happens to the data. With an intermediary, prompts and responses pass through one more party. For customer data, code, and personal information, this is a serious risk.
- Whether there is a plan B. A direct contract with the vendor, a second provider, or an open-weight model as a fallback costs less than a product shutdown caused by a ban.
For those who sell access to models themselves, the conclusion is the opposite: a business built on the gap between the vendor's rules and real demand can be profitable, but it depends on the vendor's goodwill. This story shows precisely that the vendor sees such traffic in its reporting.
Limitations
First: this is journalists' estimate based on documents we do not have access to. The range of $1.66 to $2.33 billion is wide, and the calculation methodology is not disclosed in the retelling. Second: it is unknown how Anthropic classifies "gray" users and what it plans to do with them. Third: there is no data on how much of this share will remain after the IPO, when the company will have to answer more strictly for enforcing its own rules.
The bottom line is simple. The gray market around frontier models is not a marginal backwater but a noticeable revenue line. But building your own business on it is as risky as ever: revenue for the vendor and reliability for you are different things.
- Nov 12, 2026Gemini 4 Argon Was Left Without an Owner and Went for His Crypto Wallet With $1.57 in the Account
- Nov 12, 2026Anthropic Buys $3.4 Billion in Gas Turbines: 1.7 GW to Power AI
- Nov 12, 2026Claude Fable 5.2 Designed Three Electronic Devices From Scratch Using a Short Prompt
- Nov 12, 2026TSMC to invest $1 billion in OpenAI and Anthropic, closing the AI money loop once again
Key Questions Answered
What we know?
The publication estimates that from January through September, users who accessed Claude in violation of the rules provided Anthropic with 5–7% of all year-to-date revenue. We have not seen the documents themselves, so all figures are given as relayed by Vedomosti. Anthropic has not publicly confirmed them. There is also some simple arithmetic. If $1.66 billion is 5% and $2.33 billion is 7%, total revenue for nine months comes to about $33 billion.
Why it matters?
There are two signals here for the market. First: demand for top models is not limited by geographic restrictions. If people are willing to pay through workaround schemes, the official offering does not cover the need. Founders will recognize this picture: where a product is limited by policy rather than technology, a market of intermediaries always emerges. The second signal is subtler. A company preparing for an IPO is obliged to describe its risks.
What practitioners should do?
If you are building a product on top of Claude, there are three questions. Where your traffic comes from. If requests go through a proxy or reseller, you depend on someone else's account. It can be blocked at any moment, together with your service. What happens to the data. With an intermediary, prompts and responses pass through one more party. For customer data, code, and personal information, this is a serious risk. Whether there is a plan B.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.