Anthropic Secures 220,000 GPUs in Deal to Use xAI’s Colossus Supercomputer
SAN FRANCISCO — Anthropic has secured access to the full computing capacity of xAI’s Colossus-1 data center, gaining more than 220,000 NVIDIA GPUs and over 300 megawatts of power, according to reports published this week.
The agreement, first reported by Wired, pairs Anthropic — maker of the Claude family of AI models — with infrastructure built by Elon Musk’s xAI, two companies that compete in the frontier AI market. The Colossus-1 facility is located in Memphis, Tennessee, according to The Decoder.
Anthropic expects the additional compute capacity to come online within approximately one month, according to reports. The deal expands the company’s infrastructure footprint at a time when demand for AI training and inference compute outstrips available supply.
Immediate Impact on Developers
Alongside the infrastructure deal, Anthropic announced it is doubling rate limits for Claude Code, its agentic coding tool, and raising API limits for its Opus models, according to Engadget. The changes indicate that the additional compute capacity is already being factored into the company’s product roadmap and enterprise commitments.
The rate limit increases are expected to benefit both individual developers and enterprise customers who rely on Claude’s coding capabilities for software development workflows.
An Unusual Alliance
The partnership links two competing AI labs. Anthropic, founded by former OpenAI researchers Dario and Daniela Amodei, has positioned itself as a safety-focused AI lab. Musk’s xAI built the Colossus supercomputer to train its own Grok models and competes directly in the frontier AI space.
With GPU clusters representing billions of dollars in capital expenditure, monetizing idle or underutilized capacity through partnerships may offer a near-term alternative to building new data centers, according to Wired. The Colossus-1 cluster, at 220,000 GPUs, ranks among the largest AI computing installations in the world.
Business Insider separately confirmed the partnership, reporting that Anthropic is tapping SpaceX — which shares infrastructure ties with xAI — for additional AI compute power as part of the company’s expansion strategy.
Industry Context
The agreement comes amid an industrywide scramble for compute resources. Major AI labs including OpenAI, Google DeepMind and Meta have all announced data center buildouts in recent months, while the U.S. government has signaled support for expanding domestic AI infrastructure capacity.
For Anthropic, the deal provides a near-term solution to compute constraints without the yearslong lead time required to build new data centers from scratch.
The partnership also raises questions about data security and competitive dynamics. How Anthropic and xAI will manage the separation of proprietary model training data and intellectual property on shared infrastructure was not immediately clear from available reports.
Financial terms of the deal were not disclosed.
