The reaction to Anthropic's recent rate limit changes has followed a familiar pattern: users see restrictions tighten, assume it signals greed or mismanagement, and the discourse escalates from there. An AI engineer who holds professional relationships at both Anthropic and OpenAI - and subscribes to both services personally - argues the conversation is missing the structural reality driving these decisions.
The core point: Anthropic is in a fundamentally harder position than its major competitors, and the rate decisions reflect genuine business pressure rather than arbitrary policy.
The Infrastructure Math Is Different
OpenAI benefits from Microsoft's multi-billion dollar compute commitment and embedded distribution through Azure. Google's model teams operate inside a company that owns its own chip manufacturing pipeline and global data center footprint. Anthropic is raising capital independently to fund both frontier research and the compute required to run those models at scale.
Running a large language model (a type of AI that generates text by predicting likely next words, trained on massive datasets) at Anthropic's usage volume isn't comparable to hosting a web application. Each query runs on specialized chips - GPUs and TPUs - that cost hundreds of thousands of dollars per rack to acquire and significant ongoing power to operate. Claude 3.7 Sonnet, the company's current flagship, is a large and expensive model to serve.
The Subscription Price Was Set to Win Users, Not Cover Costs
Claude's Pro plan is $20/month. Heavy users - developers running multi-hour coding sessions, researchers querying long documents repeatedly, content teams working through dozens of tasks daily - cost Anthropic significantly more than that to serve. The $100/month Max plan closes the gap for power users, but most subscribers aren't on Max.
This creates a structural problem: the users most likely to hit rate limits and complain publicly are exactly the users who are most expensive to serve at current price points. Rate limits function, in part, as a cost management mechanism when raising base prices is competitively difficult.
The engineer's framing, offered without defending any specific decision Anthropic made: the public discourse tends to treat Anthropic as a company with abundant resources choosing to be restrictive. The actual situation is a company balancing real infrastructure costs against subscription pricing that was set aggressively during a user acquisition phase, now working toward unit economics that can sustain the business before its next raise.
That doesn't make every rate limit call the right one. But the financial pressure behind them is not manufactured - and understanding it changes what a realistic ask from users looks like.