Why does xAI rent its data center Colossus 1 to Authropic? This tweet should be the most insightful analysis, with the core logic being: XAI currently holds approximately 550000+GPUs (based on H100 equivalent performance), while Colossus 1 (220000) only accounts for about 40% of the total available capacity and is a hybrid H100/H200/GB200 training cluster. This hybrid cluster is not suitable for training (different generations of GPUs have significant communication delays), but it is very suitable for inference (inference requires much less tightly synchronized communication between GPUs). Coincidentally, Authropic currently requires the most inference computing power, and one family can consume all of Colossus 1's computing power. Moreover, Anthropic, as a single tenant, occupies all 220000 GPUs, and the network switching jitter (unexpected delay) that occurred in multi tenancy has disappeared. The technical weaknesses of both sides ultimately complement each other almost perfectly. Lao Ma left Colossus 2, a data center built entirely on Blackwell, for himself to train the next generation of large-scale models for xAI. Lease out older, mixed generation Colossus 1. As a mixed H100/H200/GB200 training cluster, Colossus 1 can only achieve an MFU (utilization rate) of 11%. However, once it is handed over to a single inference client, this asset transforms into a cash flow asset and is rented out at a price of approximately $2.60 per GPU hour (weighted average of GPU type leasing rates). For xAI, what was originally a "hell cluster" for training has become a "golden goose" when redeployed for inference, bringing in $5-6 billion in revenue annually. When comparing this $6 billion with xAI's income statement, its weight becomes clearer. Annualized xAI's 1Q26 net loss to approximately $6 billion annually. In other words, leasing Colossus 1 to Anthropic generated an annual revenue of 5-6 billion US dollars, almost perfectly offsetting xAI's loss figures It has to be said that this is a perfect collaboration: Authropic has gained much-needed reasoning and computing power; SpaceXAI has obtained cash flow that can make up for its annual losses in AI business.