By ByteBot
On May 6, Anthropic announced they’re taking the full capacity of SpaceX’s Colossus 1 data center in Memphis, Tennessee—over 300 megawatts of compute power and 220,000 Nvidia GPUs. The kicker: Elon Musk, who owns SpaceX, called Anthropic “evil” just three months ago. Now he’s their data landlord, collecting an estimated $4 billion to host their infrastructure while simultaneously suing OpenAI for commercialization. The irony is thick.
Claude Code Gets Immediate Upgrades
If you noticed Claude Code running smoother this week, this is why. Anthropic doubled the 5-hour rate limits for Pro, Max, Team, and Enterprise plans, removed peak-hour throttling for paid subscribers, and substantially increased API quotas. Tier 1 API developers saw a 1500% boost in input tokens per minute and 900% in output tokens. The free plan got nothing, but if you’re paying, you just got a serious upgrade.
Anthropic needed this. In Q1 2026, they experienced 80-fold growth—eight times higher than the 10x they’d planned for. Revenue run rate surged from $87 million in January 2024 to $30 billion by April 2026. CEO Dario Amodei admitted the growth was “just crazy” and “too hard to handle.” The result: infrastructure strain, reliability issues during peak hours, and users hitting rate limits mid-session. This deal is an emergency response, not a strategic partnership.
300 Megawatts of Nvidia GPUs
The hardware Anthropic is getting is substantial. Colossus 1 in Memphis houses over 220,000 Nvidia GPUs—a mix of H100s, H200s, and GB200s. The H200 alone packs 141GB of memory and 4.8 TB/s bandwidth, delivering 2x faster inference than the H100. Power consumption: 300+ megawatts, enough to run 200,000 homes. SpaceX built the facility in 122 days, then migrated their xAI training workloads to the larger Colossus 2 (1.5 gigawatts), freeing Colossus 1 for rent. Anthropic gets all of it within the month.
Musk’s $4 Billion Contradiction
Now about that Musk contradiction. In February 2026, he posted that Anthropic “hates Western civilization.” Three months before this deal, he called them “evil” and “misanthropic.” He’s currently in court suing OpenAI for allegedly betraying its nonprofit mission—testimony from former employees revealed OpenAI shifted from research-focused to product-focused, disbanded its AGI readiness team, and shut down Super Alignment. Musk’s argument: OpenAI sold out. Yet here he is, renting 220,000 GPUs to Anthropic, a for-profit AI lab scaling as fast as possible. On X, he wrote: “No one set off my evil detector.” Apparently, evil is negotiable at $4 billion.
This isn’t hypocrisy—it’s business. SpaceX had spare capacity after migrating to Colossus 2. Leaving it idle makes no sense. Renting it to Anthropic generates revenue to fund xAI’s competition with Claude. Musk keeps the bigger facility (Colossus 2) for his own models, profits from Anthropic’s growth, and maintains leverage. If Anthropic becomes too dependent, Musk controls their infrastructure. Smart play, but it exposes Anthropic to risk. What happens if Musk changes terms or priorities shift? Vendor lock-in to your critic’s data center is a precarious position.
Space Compute: Hype vs Reality
Anthropic and SpaceX also “expressed interest” in developing gigawatts of space-based compute. Sounds futuristic, but it’s vaporware for now. Generating 1 gigawatt in orbit requires roughly 1 square kilometer of solar panels. Experts estimate a 1GW orbital data center would cost over $50 billion—three times the cost of an equivalent Earth-based facility. Challenges include thermal management (no air cooling in space), radiation that can kill electronics during solar storms, and replacing chips every 5-6 years via costly launches. Launch costs need to drop below $200 per kilogram by 2035 for viability, and even then, the economics are questionable. Starcloud launched a 60kg satellite with an H100 as proof-of-concept, but gigawatt-scale is decades away. Anthropic needs GPUs now, not science fiction.
Infrastructure Is the New AI Bottleneck
This deal signals where AI infrastructure is headed. Big Tech is spending $725 billion on AI capex in 2026, up 75% year-over-year, almost entirely on data centers, GPUs, and power. Infrastructure access is the new competitive moat. Anthropic’s 80x growth hit a wall because they couldn’t secure enough compute. The solution: rent an entire data center. Expect more AI labs to pursue full-capacity deals instead of incremental scaling. Power and access will separate winners from losers, not algorithms. Smaller labs without deep pockets or strategic partnerships will struggle to compete.
Anthropic gets immediate relief. Peak-hour throttling is gone, rate limits doubled, and 300 megawatts of compute buys them time to plan for the next growth surge. But the long-term strategy is unclear. Relying on Musk’s infrastructure is a gamble. Other AI labs will face the same bottleneck and pursue similar deals. Infrastructure, not innovation, determines who scales in 2026. If you’re building on Claude Code, you just got a capacity boost—but your app’s reliability still depends on Anthropic’s ability to navigate GPU politics.
