Elon Musk’s xAI has signed a deal to lease its Colossus 1 data center, which consists of over 220,000 mixed-generation NVIDIA GPUs, to Anthropic for exclusive use in inference workloads, a move valued at $5 billion. This decision comes as xAI’s mixed GPU architecture was proving inefficient for training models due to the “straggler effect,” where differing GPU speeds hinder progress. By shifting its training operations to the more efficient Colossus 2, built entirely on homogeneous Blackwell chips, xAI can reframe itself as a profitable infrastructure provider, potentially easing its path to a successful IPO by showcasing a stable revenue stream amidst its ongoing financial challenges.

xAI: xAI is an AI company building Grok, a chatbot designed to accelerate human scientific discovery and understand the universe. Recently integrated into SpaceX following its acquisition, xAI manages large-scale data centers including the Colossus clusters for advanced AI workloads. In this news, xAI leased its entire Colossus 1 cluster to Anthropic for inference use, shifting its own training to the more homogeneous Colossus 2 to optimize efficiency ahead of the SpaceXAI IPO.2546
NVIDIA: NVIDIA dominates AI computing with GPU architectures like the Hopper series including H100 and H200, and Blackwell featuring GB200 for training and inference. Recent advancements emphasize inference optimization and agentic systems. The Colossus 1 cluster employs a mix of NVIDIA’s Hopper and early Blackwell GPUs, which pose training challenges due to heterogeneity but suit inference well.93
SpaceX: SpaceX is an aerospace company pioneering reusable rockets like Starship for space transportation, satellite internet via Starlink, and human spaceflight. It acquired xAI in February 2026, creating the combined SpaceXAI entity focused on integrating AI with space infrastructure. The company is preparing for an IPO as early as June, with the Colossus 1 lease to Anthropic serving as financial engineering to demonstrate revenue from AI compute assets.6091
Anthropic: Anthropic is a public benefit corporation conducting AI safety research and developing the Claude models for reliable, interpretable AI systems. It recently released advanced versions like Claude Opus emphasizing coding and complex tasks. In the news, Anthropic leased xAI’s full Colossus 1 capacity to address surging inference demand for Claude, marking a pragmatic partnership with a competitor.9248

Compute Deal: xAI announced a partnership granting Anthropic exclusive access to Colossus 1 as a single tenant for inference, resolving capacity constraints amid AI demand.
Technical Rationale: Colossus 1’s mixed GPU generations create straggler effects that hinder distributed training but align well with inference’s parallel nature.
Financial Positioning: Leasing repurposes underutilized training assets into a revenue stream, reframing SpaceXAI as an infrastructure provider ahead of its IPO.