What will the published weekly Sonnet-hour usage cap on the Claude Code Max 20x (highest-tier) consumer plan be as of the end of May 2027, in hours?

The core metric for the highest individual Claude Code tier (Max 20x, $200/mo) has historically been anchored by a published weekly Sonnet-hour range of 240–480, yielding a ~360-hour midpoint 2 sources. In May 2026, Anthropic implemented an explicitly temporary +50% weekly boost—scheduled to expire July 13, 2026 2 sources—pushing the effective midpoint to ~540 hours. The median of ~593 assumes this temporary boost becomes a permanent baseline as infrastructure catches up, followed by modest further expansion by mid-2027.

The primary upward pressure is a massive multi-GW compute buildout coming online through 2026 and 2027, including commitments from AWS (+5GW) anthropic.com, Google/Broadcom (5GW TPU) anthropic.com, and SpaceX Colossus 2 sources. Furthermore, competitive retention pressures against OpenAI's Codex encourage generosity on cheaper models like Sonnet.

However, intense demand and margin discipline will tightly constrain how much of this new capacity flows into fixed consumer allowances. With Anthropic's overall run-rate revenue at ~$47B and Claude Code specifically surging to ~$8B, demand continues to fiercely compete for compute. Reuters recently reported active peak-hour rationing reuters.com, demonstrating real system tightness despite the hardware expansion. Additionally, with a draft S-1 filed on June 1, 2026, pre-IPO margin protection incentivizes Anthropic to monetize overflow via API billing rather than offering unbounded flat-rate consumer usage anthropic.com. The recent global suspension of the Fable and Mythos line anthropic.com is treated as a second-order factor; while it frees up premium compute in the near term, it does not fundamentally alter the long-term hardware-to-demand trajectory for Sonnet.

The distribution balances these forces. The median (~593) reflects modest, controlled growth beyond the current boosted levels. The lower tail (p10: 320) captures the risk that the May 2026 +50% boost reverts to the ~360 baseline, or that explosive demand forces retrenchment. The fat right tail (p90: 1333) accounts for scenarios where scaling efficiency radically cheapens Sonnet inference, or where compute supply temporarily outstrips demand, enabling Anthropic to pivot to 'near-unlimited' marketing for its mainstream bucket.