Anthropic recently imposed stricter rate limits on Claude Code subscribers during peak hours due to limited GPU resources and growing demand, prioritizing enterprise customers while doubling off-peak usage. Although the changes were driven by genuine capacity constraints, poor communication and a research-centric company culture led to significant user frustration and damaged trust.
Anthropic recently changed the rate limits for Claude Code subscribers, sparking significant user frustration. The key update restricts usage during peak hours (5:00 a.m. to 11:00 a.m. Pacific time) by making users consume their 5-hour session limits faster, while off-peak usage was doubled. Historically, Claude Code subscriptions were very generous, allowing users to consume compute worth thousands of dollars for a relatively low subscription fee. However, due to growing demand and limited GPU resources, Anthropic had to impose these new restrictions to manage capacity, especially to prioritize enterprise customers during peak times.
The core issue behind these changes is Anthropic’s struggle with GPU allocation. GPUs are essential not only for running Claude models but also for research and product development. Internally, there is tension between research teams, product teams, and users over limited GPU resources. Research requires GPUs for training new models, product teams need them to build and maintain services, and users consume them through subscriptions and enterprise usage. Despite rapid growth and increasing revenue, Anthropic has been slow to invest in expanding their GPU infrastructure, leading to a compute bottleneck that forces tough prioritization decisions.
Anthropic’s transition from a research-focused company to a product-driven one has been challenging. The company culture and leadership remain research-centric, which complicates balancing the needs of paying customers and internal teams. The communication around the rate limit changes was poorly handled, with announcements coming late and primarily through individual employees on Twitter rather than official channels. This lack of transparency and timely communication exacerbated user dissatisfaction, even though the changes themselves were driven by genuine resource constraints rather than greed or malice.
Comparisons with OpenAI highlight differences in approach and communication. OpenAI frequently resets usage limits and openly communicates changes, which has earned them more user goodwill. Anthropic’s approach, while arguably reasonable in trying to protect enterprise customers during peak hours, suffered from poor messaging and lack of clarity about the extent of the restrictions. Users reported significant impacts, with some experiencing a drastic reduction in allowed usage during peak times, fueling frustration and conspiracy theories about the company’s motives.
In summary, Anthropic’s rate limit changes reflect a broader compute crisis caused by limited GPU availability amid rapid growth. The company subsidized usage heavily in the past but now must ration resources more strictly. While the technical and economic reasons for the changes are understandable, Anthropic’s poor communication and internal cultural challenges have damaged trust with users. The situation underscores the difficulty of scaling AI services sustainably and the importance of transparent, user-focused communication in managing such transitions.