On March 13, 2026, Anthropic made significant changes to the usage limit mechanism for its Claude AI, introducing an explicit time‑based management approach: weekday peak hours are tightened so users exhaust the 5‑hour rolling quota faster, while limited‑time promotions substantially relax usage during off‑peak hours and weekends. The set of changes has been likened by users and media to “using like electricity time‑of‑use,” aimed at balancing surging demand against limited GPU capacity.
Specific adjustments
Claude now uses a dual mechanism of a rolling 5‑hour usage window plus a weekly cap to control access. From March 13 through March 28 (some reports say ending March 27), Anthropic launched a time‑limited promotion:
- Off‑Peak 2×: Outside weekday peak hours (Eastern Time 8 AM–2 PM, i.e., Pacific Time 5–11 AM, or Greenwich Time 12–6 PM) and during the entire weekend, the 5‑hour usage window is doubled. The additional off‑peak usage is not counted toward the weekly cap.
- Peak adjustments: Around March 26, Anthropic confirmed further tightening during peak hours (Pacific Time 5 AM–11 AM / Eastern Time 8 AM–2 PM), causing users to consume their 5‑hour quota more quickly (an effective “accelerated burn”), while the weekly total cap remains unchanged.
The mechanism covers Free, Pro, Max and Team plans (Enterprise excluded) and applies across web, desktop and mobile clients, as well as productivity integrations such as Claude Code, Cowork, and Excel/PowerPoint. Users do not need to opt in; the system applies automatically.
Peak times are defined in Pacific Time (PT), which affects global users differently. For example, Chinese users (Beijing Time) see peak periods correspond roughly to evening hours, while weekends receive the doubled allowance all day.
Reasons behind the change
The move is primarily about capacity management. As Claude — especially Claude Code and Cowork — surged in popularity within developer communities, demand spiked and GPU loads rose significantly during peak periods. Anthropic framed the change as a way “to manage growing demand.”
GPU compute cannot be scaled infinitely like traditional software: servers may be fully loaded at peak times while leaving substantial idle capacity overnight or on weekends. By incentivizing staggered usage, Anthropic encourages users to shift heavy tasks (such as long‑document analysis, complex code debugging, or large agent workflows) to off‑peak windows, smoothing infrastructure utilization and reducing the risk of outages or wide‑scale throttling at peak.
The approach resembles the utility industry’s time‑of‑use pricing: more expensive at peak, cheaper off‑peak, with the goal of flattening load. Cloud providers (for example, AWS) have long used time‑based pricing or throttling strategies. Anthropic’s move is seen by some as a sign that the AI industry is moving from “unlimited use” toward “explicit rationing.”
Competition also plays a role. Facing pressure from rivals such as OpenAI and given the token‑intensive nature of tools like Claude Code, Anthropic appears to be using off‑peak incentives to retain users while buying time to scale capacity.
Impact on users and practical advice
- Upside: Weekends and late‑night/early‑morning off‑peak periods offer higher usage allowances—suitable for long background jobs, batch processing, or heavy developer workloads. The extra off‑peak usage does not count against the weekly cap, effectively offering “more free use.”
- Challenges: Weekday peak hours (notably U.S. West Coast morning, which overlaps many developers’ work hours) are more likely to hit limits. Some Max users reported being constrained at critical times despite paying for top tiers, raising trust concerns. Heavy users will need to reschedule workflows.
Practical tips:
- Schedule complex, token‑intensive tasks (e.g., long‑chain reasoning in Claude Code or bulk document processing) during off‑peak hours or weekends.
- Use timezone‑conversion tools to monitor local peak/off‑peak windows.
- Watch the usage meter in the Claude interface and switch models or tasks in advance.
- Be aware the promotion may end or the mechanism may evolve into a permanent or more granular time‑segmentation policy.
Community reaction is mixed: some praise the “smart balancing of supply and demand,” preferring this to sudden blanket throttling; others criticize peak tightening for harming productivity, especially for developers dependent on Claude Code.
Industry implication: the era of AI usage rationing
Anthropic’s adjustment highlights a common challenge for generative AI: compute remains a scarce resource. Even as models improve rapidly, infrastructure expansion struggles to keep pace with explosive demand. More AI service providers may adopt similar “dynamic pricing + time segmentation” measures, potentially moving toward genuine time‑of‑use billing.
The change also nudges users toward more efficient habits: from “always‑on unlimited chat” to planned usage. For enterprise customers, it underscores the importance of combining RAG, caching and agent optimization to reduce per‑interaction costs.
Overall, Anthropic’s off‑peak strategy is a pragmatic compromise—easing immediate pressure while offering off‑peak benefits to maintain goodwill. It signals a shift from early‑stage “feasts” to more rationalized provisioning. For ordinary users, a bit of planning can yield more usable capacity.
The promotion is set to end in late March; whether Anthropic will make the mechanism permanent remains to be seen. If you use Claude, check the official support page for the latest rules and adjust your workflow to your timezone. If you need specific timezone conversions or task‑optimization suggestions, feel free to provide details.
▶️ Enjoyed this article? Creating content is hard!
Follow our YouTube channel:
AI Rosetta
Your likes, subscriptions and shares are our greatest support! 🙏
More AI news at: https://www.airosetta.com