I woke up to a tweet I genuinely thought would never happen. Elon Musk, of all people, just bailed out Anthropic with compute. The same Elon who’s been calling them “misanthropic” for months.
This is straight from Matthew Berman, who hopped on stream to untangle the whole thing. The original poster walks through the announcements, the bad blood, and the real reason this deal exists. I had to share his breakdown because the context changes everything.
What actually shipped
Anthropic signed a partnership with SpaceX. Effective today, every single GPU at the Colossus 1 data center in Memphis goes to Claude. That’s over 300 megawatts and more than 220,000 Nvidia GPUs. Online now, not in 18 months.
The twist? Just weeks ago, SpaceX signed a similar deal with Cursor. Now Elon clarified Colossus 2 is handling xAI training plus Cursor’s next coding model, while Colossus 1 is fully leased to Anthropic. The creator points out that if he were Cursor right now, he’d be nervous about that pending $60B acquisition option.
The immediate user wins
This is what Berman is most pumped about for everyday Claude users:
- Claude Code 5-hour rate limits doubled for Pro, Max, Team, and seat-based Enterprise
- Peak hours limit reduction removed for Pro and Max accounts
- Opus API rate limits jumped massively (Tier 1 went from 30K to 500K input tokens per minute, Tier 4 went from 2M to 10M)
The author has been hammering Anthropic for months over quota cuts and zero transparency. He made an entire “WTF Anthropic” video about it. Today’s announcement is the unlock he’s been waiting for.
Why Elon caved
Berman lays out the calculus cleanly. Elon had hundreds of megawatts of GPUs sitting idle, burning cash by the second. xAI moved its own training to Colossus 2 because, in his words, “xAI was not built right the first time around” and is being rebuilt from scratch (same playbook as Tesla’s autopilot rewrite to end-to-end neural nets).
So Elon faced two options: lease compute to Dario at Anthropic (a guy he’s publicly trashed) or to Sam Altman (a guy he’s actively suing). The expert reads it as pure spite math. Help the enemy you dislike less.
Elon’s tweet defending the deal is the most reluctant praise Berman has ever seen. “No one set off my evil detector.” Compare that to past Elon tweets calling Anthropic hypocritical, accusing them of stealing training data, and saying their employees should quit. The contrast is wild.
The bigger thesis
The creator floats an interesting idea: if models are commoditizing and chips are commoditizing (Anthropic happily uses Trainium, TPUs, and Nvidia interchangeably), then the real bottleneck at the bottom of the stack is energy. Not silicon. Not algorithms. Electricity.
That’s why these compute partnerships keep stacking. Anthropic also has 5GW with Amazon, 5GW with Google and Broadcom, $30B in Azure capacity, and $50B in U.S. AI infrastructure with Fluidstack.
Go watch the full video for Berman’s full reaction, the receipts on Elon’s old anti-Anthropic tweets, and his take on whether Cursor should be panicking right now.