Anthropic announced yesterday it’s leasing all compute capacity at SpaceX’s Colossus 1 data center in Memphis—300 megawatts and 220,000+ Nvidia GPUs. The deal, confirmed May 6, 2026, addresses Anthropic’s infrastructure crisis after 80-fold revenue growth in Q1 2026 broke their systems, forcing rate limits on Claude Pro and Max users during peak hours. Within hours, Anthropic doubled Claude Code rate limits and removed peak-hour throttling entirely.
The irony is striking. Colossus 1 was built by xAI for Grok AI development. After SpaceX acquired xAI in February 2026 for $1.25 trillion, Elon Musk—who called Anthropic “evil” just three months earlier—now powers their competitor’s infrastructure.
80-Fold Growth Broke Anthropic’s Infrastructure
Anthropic’s revenue and usage grew 80-fold in Q1 2026 on an annualized basis, hitting a $30-44 billion ARR run rate. CEO Dario Amodei admitted they “tried to plan for 10-fold growth” but demand quintupled their projections. The strain caused what Anthropic publicly acknowledged as “inevitable” reliability issues, forcing peak-hour usage caps on Pro and Max subscribers.
“The current level of growth is just crazy and too hard to handle,” Amodei said at the May 6 announcement. Anthropic couldn’t wait 3-7 years for traditional hyperscaler capacity. They needed 300MW now. SpaceX delivered.
Elon Musk: From “Evil” to “Impressed” in 90 Days
In February 2026, Elon Musk posted on X that Anthropic “hates Western civilization.” Three months later, after meeting with Anthropic’s senior team, Musk reversed completely: “Everyone I met was highly competent and cared a great deal about doing the right thing. No one set off my evil detector.”
The backdrop makes this reversal even more significant. SpaceX acquired xAI—Musk’s AI company and maker of Grok—for $1.25 trillion in February 2026, creating the world’s most valuable private company. Colossus 1 was originally built for xAI’s Grok model. Now it powers Claude, a direct competitor. Musk framed the arrangement similarly to SpaceX’s satellite launch model: selling compute capacity to customers while reserving rights to reclaim resources “if AI causes harm to humanity.”
That last part creates vendor risk. The “harm” criteria rest solely with Musk’s judgment.
What Claude Users Get Now
Effective May 6, 2026—the same day as the announcement—Anthropic doubled Claude Code’s five-hour rate limits for Pro, Max, Team, and Enterprise plans. Peak-hour throttling disappeared entirely for Pro and Max users. API rate limits for Claude Opus models increased significantly.
Before the deal, Claude Code imposed 5-hour usage limits that dropped further during peak hours (9am-5pm PT). After May 6, those limits doubled and peak-hour reductions vanished. Developers can now work uninterrupted during business hours without hitting “come back later” messages.
Related: OpenAI MRC Protocol: Six Tech Giants Fix GPU Network Failures
Why SpaceX? The GPU Shortage Answer
By 2026, hyperscalers like Microsoft, Google, and AWS have fully allocated GPU capacity through 2027. Lead times stretch 36-52 weeks. Grid interconnection queues in key markets hit 7 years. Combined hyperscaler capital expenditures total $660-690 billion, but they can’t build data centers fast enough to meet demand.
Colossus 1 represents a different model. xAI built the facility in just 122 days in 2024, converting a former Electrolux factory in Memphis’s Boxtown district. The 220,000+ Nvidia GPUs (H100, H200, and next-generation GB200 accelerators) deliver 300+ megawatts of compute power. SpaceX plans to expand toward 2 gigawatts total capacity.
This is the “neocloud” advantage: deployment measured in months rather than years. SpaceX, CoreWeave, and Lambda Labs fill the gap between urgent capacity needs and hyperscaler backlogs. Speed comes with trade-offs—Colossus 1 initially operated with 35 unpermitted gas turbines before environmental lawsuits forced compliance. But for Anthropic, speed mattered more than regulatory perfection.
Key Takeaways
- Anthropic’s 80x Q1 2026 growth forced urgent capacity needs that hyperscalers couldn’t meet on acceptable timelines
- SpaceX provides 300MW of compute (220,000+ GPUs) within a month versus 3-7 years for traditional data center procurement
- Claude users saw immediate benefits May 6: rate limits doubled, peak-hour throttling removed
- Elon Musk reversed from calling Anthropic “evil” to leasing them infrastructure in 90 days, though he retains unilateral power to reclaim capacity
- The neocloud model (SpaceX, CoreWeave, Lambda) emerges as the only viable option for AI companies facing explosive growth and GPU shortages
