This might be the most dramatic turn of events in the AI industry recently. Elon Musk’s AI company xAI’s Colossus 1 was just built and started running in an extremely short timeframe, only to be hit with negative reports that despite possessing 550,000 NVIDIA GPUs, its utilization rate (MFU) was only 11%. Just as the outside world was watching xAI’s development, tonight news broke that xAI has reached an agreement with rival Anthropic: to rent out all the computing power of the Colossus 1 data center to Anthropic, turning the world’s largest AI supercomputer located in Memphis overnight from 11% low utilization to full-capacity operation.
xAI’s Embarrassment: 550,000 GPUs Only Running at 11%
According to foreign media《The InformationAn internal xAI memo shows that xAI President Michael Nicolls admitted to the team that the company’s current MFU is only about 11%. This figure means that hardware that could theoretically produce 100 units of training compute is actually only producing 11 units, equivalent to effective compute from 550,000 GPUs being reduced to only around 60,000 GPUs.
Nichols described the results as “embarrassingly low” in the memo, and set a target for the team: to raise the MFU to 50% within the next few months.
Brilliant move: Anthropic fully covers all Colossus 1 compute
However, just as xAI was grappling with low utilization rates, an unexpected opportunity arose. Just now, SpaceXAI (In February 2026, SpaceX officially acquired xAI, with the merged entity being SpaceXAI) announced a partnership agreement with Anthropic (the developer of Claude) to provide all computing power of the Colossus 1 data center to Anthropic.
SpaceXAI will provide @AnthropicAI with access to Colossus 1, one of the world’s largest and fastest-deployed AI supercomputers, to provide additional capacity for Claude → https://t.co/nfDR9S822L pic.twitter.com/EQAz0S84m2
— xAI (@xai) May 6, 2026
This partnership brings Anthropic over 300 megawatts (MW) of new power capacity, equivalent to more than 220,000 NVIDIA GPUs worth of computing resources, expected to be connected within this month.
Celebrating massive computing power gains! Anthropic raises Claude usage limits in sync
To celebrate this major expansion of computing power, Anthropic simultaneously announced that starting todayAcross-the-board increase Usage limits for the Claude series, specifically including:
- Claude Code usage doubledThe 5-hour usage limit for Pro, Max, and Team plans is doubled, with Enterprise following suit.
- Cancel peak-hour reductionPro and Max plans no longer reduce usage limits during peak hours in Claude Code
- Opus API significantly improvedOpus model’s API usage limits have received a significant boost
For engineers and developers who rely on Claude Code for their daily development work, this is undoubtedly a godsend.
From 11% to Full Load: xAI’s Perfect Comeback
The brilliance of this deal lies in:
- the 89% of computing power that xAI itself doesn’t useNo more wasteful idling
- SpaceXAI sells capacity to AnthropicTurn idle assets into substantial income
- Anthropic secures much-needed computing power, easing the pressure from Claude’s surging demand
- Colossus went from just 11% utilization to running at full capacity
In other words, Musk found an incredibly clever solution to xAI’s underutilized compute problem—if he can’t use it all himself, he’ll just sell it to people desperately in need of compute power (Anthropic). All in all, it’s a genuine win-win, and congratulations to Musk and developers worldwide who’ve been suffering from Claude’s degraded performance. That said, things would’ve been pretty awkward for him if Anthropic hadn’t stepped in to save the day. He really wants to build more data centers and construct his own Terafab—does he not fear ending up with idle capacity again?
Source: KOCPC Chinese