Anthropic-SpaceX Compute Deal Doubles Claude Code Rate Limits (May 2026)

On May 6, 2026, Anthropic made two announcements simultaneously: a compute partnership with SpaceX giving access to more than 220,000 additional NVIDIA GPUs within a month, and — immediately more relevant for most developers — doubled rate limits across every paid Claude Code plan. If you’ve been hitting Claude Code’s five-hour usage ceiling during intensive sprints, or tolerating Sonnet fallback when you’d rather run Opus, this week’s changes are material. This guide covers exactly what changed by plan tier, why the Colossus 1 deal matters beyond the headline, and how to structure your workflow to extract full value from the new limits.

What Is Colossus 1?

Colossus 1 is a large-scale AI data center in Memphis, Tennessee, originally built as a joint operation between SpaceX and xAI to support Grok model training and inference. The facility runs on more than 300 megawatts of compute capacity — equivalent to roughly 220,000 NVIDIA H100-class GPUs — with the high-bandwidth interconnects and power infrastructure designed for months-long frontier model training runs.

Anthropic’s announcement specifies an agreement to use all of Colossus 1’s capacity within a month — a full-facility deal, not a partial allocation. xAI confirmed the partnership separately, describing it as part of a strategy to monetize Colossus infrastructure during the interval between major Grok model training cycles. The deal gives Anthropic rapid access to proven, operational GPU clusters rather than waiting for new hardware to be built and provisioned from scratch.

Anthropic’s appetite for this scale of compute is grounded in hard numbers. The company crossed $30 billion in annualized recurring revenue in April 2026, surpassing OpenAI at $24 billion for the first time. Claude Code is a primary driver of that growth, but growth at that pace runs into an infrastructure ceiling quickly: Claude Opus 4.7 is expensive to serve at scale, and Anthropic had been managing that constraint through usage tiering, dynamic throttling, and model fallback. Colossus 1 is the structural fix for that constraint.

The New Claude Code Limits: Plan-by-Plan Breakdown

The limit changes apply to all paid Claude Code tiers. The mechanics differ by plan:

• Claude Code Pro ($20/month): Five-hour rolling rate limits are doubled. Peak-hour usage caps — the dynamic throttling that previously reduced Opus access during high-traffic daytime periods — are removed entirely. Pro subscribers now get consistent Claude Opus 4.7 access throughout the five-hour window regardless of time of day.
• Claude Code Max ($100/month): Five-hour limits doubled from the previous Max ceiling, with peak-hour caps also removed. Max subscribers were already at 5× the original Pro limit; they’re now at the equivalent of 10× that baseline in a rolling five-hour window. This tier was designed for intensive professional use, and the new limits are genuinely difficult to exhaust in a full workday of continuous coding.
• Claude Code Team ($30/user/month): Same five-hour limit doubling, with usage pooled across the team. Organizations running multi-agent workflows — where several engineers or automated pipelines operate parallel Claude Code sessions simultaneously — will see the most immediate impact. Previously, team-level orchestration during intensive sprint periods would hit the shared rate ceiling regularly.
• Enterprise (seat-based): Seat-based Enterprise plans receive the same rate limit increases as Team, plus higher request volume for Claude Opus API calls. API-access Enterprise customers running Claude Code programmatically rather than via the CLI should contact their account team; per-account limit configuration is being rolled out over the next two weeks.

One technical detail worth noting: these limits operate on a five-hour rolling window, not a daily reset. Usage ages out on a rolling basis rather than resetting at a fixed time. For developers who use Claude Code continuously throughout a workday, the effective capacity increase is higher than a simple “double the daily cap” reading suggests, because early-session usage ages out while you’re still actively working.