<h2>What happened</h2>
<p>Anthropic has signed a compute deal with SpaceX, gaining access to the full capacity of the Colossus 1 data center in Memphis, Tennessee: 220,000+ NVIDIA GPUs drawing over 300 megawatts of power. The deal takes effect immediately and has already translated into higher usage limits for Claude users.</p>
<p>The timing isn’t subtle. Anthropic had been quietly rationing compute for months — reducing peak-hour limits, ending promotional quotas, and drawing sustained complaints from Claude Code users who reported burning through their monthly allowance in days. The company eventually acknowledged that users were hitting limits “way faster than expected,” citing 80x annualized demand growth against a planned 10x. That’s not a rounding error; that’s a structural capacity problem.</p>
<h2>What changes for users</h2>
<p>Effective immediately:</p>
<ul>
<li>Claude Code’s five-hour rate limits are doubled across Pro, Max, Team, and seat-based Enterprise plans</li>
<li>Peak-hour throttling is removed for Pro and Max accounts</li>
<li>API rate limits for Claude Opus models are significantly higher (Tier 1 input tokens per minute, for example, jumps from 30,000 to 500,000)</li>
<li>Weekly limits remain unchanged</li>
</ul>
<h2>The strange bedfellows angle</h2>
<p>This deal is awkward in ways worth acknowledging. Elon Musk called Anthropic “evil” roughly three months ago. Anthropic had previously cut off xAI’s API access, apparently to prevent model distillation. Now xAI is renting Anthropic roughly a quarter of its data center capacity.</p>
<p>The explanation is straightforward: Colossus 1 was sitting largely idle after xAI migrated its own workloads to the newer, larger Colossus 2 cluster. Grok has minimal inference demand compared to Claude. Anthropic needed capacity in weeks, not the two-plus years it takes to build new data centers. Both parties had clear financial incentives, and mutual dislike of OpenAI apparently smoothed things over.</p>
<p>There’s also a contractual clause worth noting: SpaceX retains the right to reclaim compute if Anthropic’s AI “harms humanity.” That’s a novel form of supply chain risk — your infrastructure provider gets to decide whether your product crosses an ethical line.</p>
<h2>What it signals about xAI</h2>
<p>The deal raises real questions about xAI’s direction. Companies that are confident in their AI roadmap tend to hoard compute rather than rent it out — Google and Meta being the obvious examples. Renting to a competitor suggests either that Colossus 1 genuinely has excess capacity, or that xAI is pivoting toward a neocloud model (building and renting infrastructure) rather than competing on frontier models.</p>
<p>With Grok struggling to gain enterprise traction, co-founders departing, and SpaceX reportedly planning to absorb xAI ahead of an IPO, monetizing idle GPUs makes short-term financial sense. Whether it’s a sign of strategic retreat from the AI race is a harder question.</p>
<h2>Environmental footnote</h2>
<p>Colossus 1 has a troubled environmental record. Its gas turbines initially operated without Clean Air Act permits, and the facility has been linked to increased hospital admissions from air quality degradation in the surrounding Memphis neighborhood. Anthropic hasn’t addressed this publicly, which is a notable omission for a company that emphasizes responsible AI development.</p>
<h2>Broader compute picture</h2>
<p>This deal joins a growing list of infrastructure agreements Anthropic has signed:</p>
<ul>
<li>Amazon: up to 5 GW (Trainium/Graviton)</li>
<li>Google and Broadcom: 5 GW</li>
<li>Microsoft and NVIDIA: $30B in Azure capacity</li>
<li>Fluidstack: $50B infrastructure investment</li>
<li>SpaceX: 300 MW at Colossus 1</li>
</ul>
<p>Anthropic has also mentioned interest in orbital compute with SpaceX, though nothing is signed there yet.</p>
<p>The pattern is clear: compute capacity has become the primary constraint on what these companies can build and sell. Anthropic’s annualized revenue reportedly grew from $9B to over $30B in a matter of months, driven largely by Claude Code adoption. The bottleneck was never the model — it was the hardware to run it.</p>


Collections

Anthropic has secured access to SpaceX's Colossus 1 data center in Memphis, gaining 220,000+ NVIDIA GPUs to address a severe compute shortage. The deal immediately doubles Claude Code rate limits, removes peak-hour throttling for Pro and Max users, and dramatically increases API rate limits for Claude Opus. The arrangement is notable given Elon Musk's public hostility toward Anthropic and xAI's prior conflicts with the company. xAI migrated to the larger Colossus 2 cluster, leaving Colossus 1 largely idle, making the rental financially attractive for both sides. The deal includes an unusual clause allowing SpaceX to reclaim compute if Anthropic's AI 'harms humanity.' The move reflects a broader pattern of Anthropic signing massive infrastructure deals as Claude Code adoption drove 80x annualized demand growth against a planned 10x, making compute the primary bottleneck to growth.

Anthropic partners with SpaceX to double Claude Code rate limits

<p>That “harm humanity” part would allow for an immediate rugpull from xAI if they ever needed the computing capacity, because honestly one could argue that Anthropic is already harming humanity in multiple ways, as AI has pros and cons attached to it and most cons could be interpreted as “harming humanity.”</p>
