Anthropic, the developer of the Claude neural network, has revealed a paradoxical financial detail in its preparations for a public offering on the Nasdaq exchange. According to an S-1 registration statement filed with the U.S. Securities and Exchange Commission (SEC), between 5% and 7% of the startup's total revenue for the first nine months of the current year came from users directly violating the Terms of Service. In monetary terms, this "gray" revenue is estimated between $1.66 billion and $2.33 billion. The public version of the registration statement was accepted by the regulator on October 26, 2026, while the confidential draft was submitted by the company on June 1, 2026.
Anatomy of the Gray Market: Where the Billions Came From
Following the introduction of geographical restrictions on access to advanced language models, a vast shadow infrastructure emerged worldwide to bypass blocks and resell computing power. According to the prospectus, the financial flow to Anthropic was generated by three key segments. First, transit hubs: in jurisdictions where Claude is officially unavailable (including Russia, China, and Iran), end-user requests were masked through proxy pools, VPN tunnels, and so-called "transfer stations"—server aggregation nodes through which payments were routed via foreign channels. Second, the exploitation of consumer subscriptions instead of API: developers extracted OAuth authorization tokens from fixed-rate plans (including Claude Max at $200 per month) and connected them to third-party agent systems and automation scripts, which proved significantly cheaper than official per-minute pricing. Third, reselling and account pooling, where a single premium account was shared among dozens of users, aggregating total billing in favor of the company itself.
Why Blocks Were Not Implemented Immediately
In its reporting materials, the company explicitly points out that identifying sophisticated evasion schemes—mobile and residential proxy networks, multi-layer tunnels—requires significant time and deep telemetry analysis. While monitoring was underway, subscription fees and model call charges continued to be debited from linked payment instruments. Experts also add a purely commercial argument: the higher the gross revenue, the more impressive the startup's capitalization looks in the eyes of future investors, creating a strong incentive to turn a blind eye to violations until preparations for the listing began.
The Purgatory Before Listing: Why the Screws Were Tightened Now
The period of leniency ended in late summer and autumn of 2026, when, ahead of its Nasdaq debut, the company came under close scrutiny by U.S. regulators. Earning money from users in countries under U.S. sanctions and export restrictions carries direct legal risks, prompting Anthropic to conduct massive purges: cutting off authorization via third-party CLI handlers, prohibiting the use of user tokens outside the official interface, and migrating developers to a commercial API, alongside rolling out ban waves against accounts caught using popular VPN/proxy networks and virtual payment cards.
Conclusion: Billions Before the Customer Base Cleanup
Ultimately, Anthropic managed to extract significant revenue from the global hype surrounding its models and then demonstrate compliance with strict exchange standards by cleaning up its customer base. The offering is being managed by Goldman Sachs and Morgan Stanley, with shares planned for listing under the ticker ANT, though the number of securities and the price range have not yet been established in the filed document. For the market, this is a signal: even the largest players in the AI sector are forced to disclose the share of "gray" traffic before gaining access to public capital.
