Musk says Grok's costs now outstrip its revenue

Musk's admission is a rare public acknowledgement from a frontier-lab founder that serving a model at scale loses money. Users have reported that Grok's weekly limits drain quickly. That is consistent with a provider rationing expensive inference.

The statement comes as xAI pushes aggressively on product. Grok Build coordinates up to eight concurrent coding agents in the terminal and scores 70.8% on SWE-bench Verified, running on the Grok 4.7 flagship. Grok Imagine generates 15-second 720p videos. Each of these features is compute-heavy, which makes the cost gap harder to close.

xAI is not alone. Compute costs are this week's unifying theme. Fortune asks whether compute will curtail the push toward personal AI agents. OpenAI is routing tasks to a smaller GPT-6.1 Terra model to cut costs. Apple is trimming engineers' AI budgets. Anthropic has disclosed $895.4B in compute commitments. The economics of inference are now a strategic issue for every lab.

What to watch: whether xAI raises prices, tightens limits further or seeks new funding. xAI's reported defense work is a separate thread and may become its own revenue line.