TNW: Anthropic's 'All-You-Can-Eat Buffet Just Closed'
>The Next Web delivers the sharpest summary of the economic disconnect driving Anthropic's policy change: a single OpenClaw instance running autonomously for a full day can consume the equivalent of $1,000–$5,000 in API costs, against a $200/month Max subscription. The buffet metaphor is perfect—users were consuming far more than they realized, and the restaurant has decided to change the rules.
Over 135,000 OpenClaw instances were running at the cutoff time, with analysts noting a price gap of more than 5x between what heavy users paid versus equivalent API rates. This isn't just a small adjustment—it's a fundamental rebalancing of who bears the true cost of autonomous AI workloads. When agents run 24/7, calling APIs, moving data, and executing workflows, they consume resources that were never priced into the original subscription model.
What makes TNW's take particularly sharp is the connection to Peter Steinberger's February move to OpenAI. Anthropic was effectively subsidizing the creation of a competitor by allowing third-party tools like OpenClaw to flourish under their subscription model. With Steinberger now at OpenAI and having helped build the very ecosystem that's now being cut off, this policy shift looks increasingly like an attempt to control their own destiny.
The corporate intrigue deepens when you consider Tencent's involvement. The Chinese tech giant had already built an enterprise platform on OpenClaw before this policy announcement, meaning Anthropic's decision affects not just individual developers but major corporate customers who had bet their AI strategy on the OpenClaw ecosystem.
This timing raises serious questions about whether Anthropic's move was driven by infrastructure costs or competitive concerns. With Peter Steinberger now at OpenAI and actively building out their agent strategy, Anthropic may have decided that controlling the full stack—from model to tool to usage—was more important than maintaining the open ecosystem that made Claude popular in the first place.
The economic reality is brutal for users who built workflows around fixed-price Claude subscriptions. When you can move to OpenAI's Codex program with explicit third-party support, or to alternative models like DeepSeek where the same workload costs 90% less, the decision becomes less about loyalty and more about basic economics.
For the broader AI industry, this signals a fundamental shift in how companies approach third-party integration. The "all-you-can-eat" buffet era of AI subscriptions may be ending, replaced by more granular, usage-based pricing that reflects the true cost of autonomous AI infrastructure. This change will accelerate consolidation around providers who can offer both powerful models and developer-friendly policies.