Anthropic and SpaceX Reach Full Power Cooperation on Colossus 1
Anthropic announced an agreement with SpaceX to utilize all computing resources of the Colossus 1 data center, adding over 300 megawatts (over 220,000 NVIDIA GPUs) of computing power within a month.
The collaboration will directly enhance the user experience for Claude Pro and Claude Max. The company also announced that the five-hour rate limit for Claude Code will be doubled for Pro, Max, Team, and enterprise users, and the peak limit for Pro/Max will be removed, while significantly increasing the Claude Opus API limits.
Anthropic also disclosed several infrastructure collaborations, including a $100 billion agreement with Amazon and a 5GW agreement with Google, and is exploring joint development of multi-gigawatt "orbital AI computing power" with SpaceX.
Source: Public Information
ABAB AI Insight
Anthropic has aggressively secured computing power through a multi-cloud + multi-vendor strategy since 2025, having previously signed substantial long-term contracts with Amazon, Google, Microsoft, and NVIDIA. Taking over all capacity of xAI's Colossus 1 is the fastest implementation of this strategy, continuing Dario Amodei's execution style of "safe but cost-no-object expansion of computing power."
In terms of capital, Anthropic is directly converting its latest financing into ready-to-use GPU rights for Colossus, stimulating subscription and API usage growth by immediately relaxing user limits, while exploring orbital computing for future space infrastructure. This creates a three-tier reserve of "spot computing + long-term contracts + frontier exploration," rapidly narrowing the capacity gap with OpenAI.
Similar to OpenAI's previous large prepayments to Oracle and Microsoft, and xAI's rapid deployment of Colossus, leading AI labs are currently in a fierce competition phase, transitioning from "waiting for new factories" to "taking over existing superclusters."
Essentially, this is about capital concentration: by taking over Colossus 1, Anthropic quickly converts the massive capital already deployed by xAI into its own capacity, restructuring resources from a single Musk ecosystem into a shared AI training infrastructure across companies. Mechanically, this rapidly enhances user experience and revenue through the existing 300MW+, while paving the way for future orbital computing, accelerating the structural shift of AI computing from ground clusters to a multi-level global layout.
ABAB News · Cognitive Law
The fastest computing power is not about building new facilities, but directly taking over existing superclusters. In a heated competition, it doesn't matter who your partner is; the key is who can immediately provide you with more GPUs. When two seemingly opposing companies share computing power, the AI arms race enters a new phase.