By ByteBot
Anthropic is paying Elon Musk’s xAI $1.25 billion per month for compute — and if you use Claude Code or the Claude API, you’ve already felt the effect. On May 6, 2026, Anthropic doubled Claude Code’s 5-hour session limits, removed peak-hour throttling entirely, and raised Claude Opus API input token limits by 1,500%. The full financial terms only became public on May 20, when SpaceX’s S-1 IPO filing with the SEC listed Anthropic as an xAI compute customer. That figure: $1.25 billion a month, through May 2029, for exclusive access to xAI’s Colossus 1 data center.
What Changed for Claude Users This Month
The rate limit improvements are real and already applied. Claude Code’s 5-hour session cap is doubled across Pro, Max, Team, and Enterprise seat plans. Peak-hour throttling — which frustrated developers running long sessions in the afternoon — is gone. On the API side, Claude Opus input tokens per minute jumped from 30,000 to 500,000 (a 1,500% increase), and output tokens per minute went from 8,000 to 80,000. According to Anthropic’s official announcement, the Colossus 1 compute came online incrementally in the four weeks following May 6.
Free plan users weren’t upgraded, but every paid tier got meaningful headroom. If you’ve been hitting the “5-hour wall” with Claude Code or running into throughput ceilings on Opus, the situation improved materially this month — not because Anthropic suddenly found efficiency gains, but because 220,000 more NVIDIA GPUs came online.
Related: Claude Cache Diagnostics: Stop Paying for Cache Misses You Can’t See
The $40 Billion Infrastructure Arrangement
The deal gives Anthropic exclusive access to xAI’s Colossus 1 facility in Memphis, Tennessee: 300+ megawatts of power and more than 220,000 NVIDIA GPUs. At $1.25 billion per month through May 2029, the contract is worth over $40 billion if held to term — though either party can exit with 90 days’ notice, and the first two months carry a discounted rate during ramp-up. IDC noted that Anthropic passed on established neocloud providers like CoreWeave, Lambda, and Nebius, suggesting frontier labs at this scale prefer consolidated, purpose-built infrastructure over distributed options.
The specific dollar figure only surfaced because SpaceX needed to disclose its revenue streams in its SEC IPO filing. xAI operates under the SpaceX corporate umbrella for this purpose, so the Anthropic deal appeared in the S-1 as a major customer contract. The AI industry got a rare public window into what inference capacity at this scale costs — and the number is larger than most analysts expected. As TechCrunch reported, the deal “allows us to monetize unused compute capacity in our infrastructure,” per xAI’s own framing.
Why xAI Needed to Sell
xAI built Colossus 1 to train and run Grok. Grok’s user base dropped significantly as competitors improved, and xAI had already migrated its training workloads to the larger Colossus 2. Colossus 1 sat underused — reportedly running at around 11% model flops utilization, compared to roughly 40% for rival AI infrastructure operators. At that utilization rate, Colossus 1 is not generating competitive value; it’s a $1.25-billion-per-month liability turned revenue stream.
By renting Colossus 1 to Anthropic, xAI converts idle infrastructure into revenue ahead of SpaceX’s IPO. This “neocloud” model — building AI compute clusters, then monetizing excess capacity by renting to other AI companies — is becoming an industry pattern. As TechCrunch asked directly: “Is xAI a neocloud now?” The short answer is yes.
Related: Grok Build vs Claude Code vs Codex CLI: 2026 Verdict
The Strings Attached
Two details about Colossus 1 are worth knowing. First, the facility has a disputed environmental record: its gas turbines operated without Clean Air Act permits by claiming “temporary” status, with credible reports linking it to elevated hospital admissions from air quality issues. Developer and Datasette creator Simon Willison was blunt in his assessment: “a really bad look,” even accounting for Anthropic’s genuine compute constraints.
Second, the contract contains an unusual clause allowing Elon Musk to reclaim the compute if Anthropic’s AI “engages in actions that harm humanity.” That’s a vague, subjectively defined trigger in an infrastructure agreement. The 90-day exit clause cuts both ways — Anthropic can leave, but so can xAI. Whether these terms represent real operational risk or standard legal boilerplate is an open question worth watching as the deal progresses toward its 2029 expiration.
Key Takeaways
- Claude Code rate limits are doubled for all paid plans — the infrastructure backing it is Colossus 1, now live
- Claude Opus API input token limits jumped 1,500% (30K → 500K/minute); output limits jumped 900%
- Anthropic pays xAI $1.25B/month through May 2029 — $40B+ total — for 220,000+ NVIDIA GPUs and 300+ megawatts
- xAI was selling because Grok usage dropped and Colossus 1 was running at ~11% utilization; renting to Anthropic turns a liability into revenue ahead of SpaceX’s IPO
- The contract includes a Musk “harm to humanity” override clause and an environmental record worth monitoring
Compute availability — not model architecture — is the primary constraint on frontier AI in 2026. The fact that a company needs to rent infrastructure from a direct competitor to meet demand tells you exactly where the bottleneck is.
