On April 4, Anthropic changed the terms. Starting at noon Pacific time, Claude subscriptions no longer covered usage through third-party tools. OpenClaw, the agent orchestration framework that had grown to more than 135,000 running instances, went dark for every user who relied on a Pro or Max subscription to power it.
The ban was immediate. OpenClaw's founder, Alex Steinberger, and board member Dave Morin had negotiated with Anthropic for weeks, The Verge reported, and managed to delay enforcement by exactly one week. That was the concession.
What filled the vacuum tells you something about who actually runs agent infrastructure at scale: China. The country now has roughly double the OpenClaw activity of the United States, according to SecurityScorecard analysis cited by TNW. That is a striking data point given that Beijing restricted the platform on March 11, 2026 — eleven days before Anthropic's subscription ban. Regulatory restriction on paper, heavy adoption in practice.
The scramble to replace Anthropic's offering began immediately. Tencent launched ClawPro, an OpenClaw-compatible enterprise product, on March 22. During its internal beta, more than 200 organisations across finance, government, and manufacturing had adopted it, TNW reported. ByteDance and Baidu are also building or positioning OpenClaw-adjacent products. The Chinese AI startup MiniMax, which runs on OpenClaw infrastructure and reported $79 million in revenue in 2025 with 159% year-over-year growth, put the case bluntly. "Limiting AI subs to first-party products kills these ideas before they are ever born," a MiniMax representative told the South China Morning Post.
The cost shock is where this becomes concrete for any engineer or procurement lead who thought they had a budget. API pricing for Claude Sonnet 4.6 runs $3 per million input tokens and $15 per million output tokens. Heavy OpenClaw users are now looking at cost increases of 10 to 50 times their previous monthly outlay, according to multiple user reports. That is not a pricing adjustment. That is a category change.
Anthropic's public position, from spokesperson Boris Cherny: subscriptions "weren't built for the usage patterns of these third-party tools," and the company is "prioritising customers using our products and API." That is coherent from a business perspective. Capacity is finite. Subscriptions were priced for individuals, not agent fleets. But it reads differently when you see which fleets were most dependent on the old model.
There is an irony worth naming. Steinberger, who left OpenClaw to join OpenAI in February 2026 and subsequently transferred the project to an open-source foundation, put it this way: "First they copy some popular features into their closed harness, then they lock out open source." The observation predates this specific situation, but it lands differently in the context of April 4.
Tencent's positioning is the most concrete near-term story. The company spent 18 billion yuan on AI products in 2025 and plans to double that in 2026, TNW reported. ClawPro is not a research project — it is a commercial product with a pricing model and an enterprise sales cycle. Whether it can absorb the API cost structure that replaces the subscription model is a live question.
The unresolved question — who pays the LLM API bills at scale — is the part that matters most for anyone building on top of this stack. OpenClaw is open-source infrastructure. Anthropic is a commercial AI company. The gap between those two things just became a billing problem for every enterprise that built an agent workflow assuming subscription-priced access. Chinese platforms may have just solved the distribution problem. The cost structure problem is still open.
This is also a story about platform economics and the recurring pattern of dependency. OpenClaw is not the first infrastructure layer to get pulled out from under a user base. Twitter's API shutdown in 2023 was the same story with different characters. What varies is who steps in to fill the gap — and this time, it is Chinese platforms with capital and commercial incentives that did not exist three years ago.
The Anthropic decision may ultimately consolidate agent infrastructure in China rather than displace it. That outcome would be worth watching regardless of where anyone stands on the original policy question.
