Cerebras Systems Inc. CEO Andrew Feldman revealed that SpaceXAI, formerly xAI, moved into the business of renting AI computing because its processors were not busy enough. Speaking on the Sourcery podcast on Monday, Feldman attributed the excess capacity to weaker-than-expected early adoption of the Grok model, which left billions of dollars in hardware sitting idle.

Feldman explained that SpaceXAI could not afford to leave equipment unused. "They had these GPUs sitting around, and that’s a bad idea," he said. This situation led SpaceXAI to lease a block of GPUs to Anthropic for use at SpaceX’s Colossus 1 data center in Memphis, Tennessee. Feldman noted that the company realized it could build a strong business by stepping into a constrained market despite its own model's initial struggles.

Anthropic Agreement Terms

Anthropic’s agreement involves the Colossus and Colossus II facilities. The company said the site provides more than 300 megawatts through over 220,000 Nvidia GPUs. This capacity allows Anthropic to double Claude Code limits, remove peak-hour reductions, and raise API ceilings.

Grok 4.5 Launch and Market Position

The leasing deal has not ended the competition between the firms. SpaceXAI launched Grok 4.5 on July 8, describing it as its strongest model for coding, agentic tasks, and knowledge work. Musk characterized the model as "Opus-class."

Despite the rivalry, Musk has softened his public stance toward Anthropic, recently acknowledging it as "obviously currently the leader in AI." He stated he would not weaponize SpaceXAI's compute access against the rival, emphasizing that sabotage was "not my style."

Cerebras Systems Inc. (NASDAQ: CBRS) announced a major expansion of its European infrastructure footprint, planning to increase total capacity to 200 MW by the end of 2027. The company will bring its first European data center capacity online by the end of 2026, with a rapid build-out across France and the Nordics. This expansion aims to deliver faster response times for increasingly complex AI workloads by moving high-speed AI inference infrastructure closer to European users.

A portion of the expanded capacity is expected to support OpenAI workloads as part of the companies’ existing partnership. The deployment addresses surging demand from European enterprises, research institutions, and governments seeking local, low-latency AI infrastructure alternatives to compute capacity concentrated in the U.S. and Asia.

Cerebras co-founder and CEO Andrew Feldman stated that the company is contracting significant capacity for 2027, with data centers slated for Norway and Finland. He emphasized that these deployments enable the company to meet customer requests for fast, high-performance AI compute located in Europe. Feldman added that the expansion reflects confidence in Europe as a long-term growth market for Cerebras.

Expansion Details

The expansion plan includes specific targets and locations to bolster Cerebras's presence in the region:

Cerebras's wafer-scale architecture is designed to deliver industry-leading inference and training performance. The European build-out positions the company to serve regional demand directly from within the market.