Musk says Grok Bot costs are 'unsustainably high' as AI agents hit compute wall
Elon Musk has said the cost of running Grok Bot, the personal AI assistant SpaceX introduced in August, is "unsustainably high relative to revenue", and suggested the company was cutting back the compute allocated to it, so users hit their limits sooner. Days later, OpenAI's head of consumer business, Adam Fry, told reporters at a dinner in New York that the company's Dots assistants are "on the compute heavy side". Each Dot runs on its own cloud computer to carry out tasks on the user's behalf, from reading and answering emails to booking event tickets.
The admissions land as the industry bets that proactive personal agents are the next stage of AI. Meta's Muse assistant is credited with convincing rivals of that, but the compute needed to run agents that are always on and constantly working is the constraint. Dots are limited to Pro subscribers paying $100 a month; Fry said he expects a breakthrough in compute efficiency to bring costs down quickly even as models improve, and OpenAI is introducing a router that sends simpler tasks to a smaller model, GPT-6.1 Terra, rather than the larger GPT-6 Astra that originally handled every Dots task.
OpenAI is also trying to head off resistance to the data centres those agents depend on. This week it began calling them "compute campuses", promising in an inaugural post to ease strain on local electricity grids and not to draw on power needed by local families, after bipartisan pushback across the US. Anthropic's updated S-1 filing shows the scale of the spending: purchase agreements totalling $895.4 billion, "substantially all of which is related to compute and infrastructure", with failure to secure more compute listed as a risk to its success.
OpenAI faces its own constraints. Its fleet leans heavily on Nvidia GPUs, though it is developing an in-house chip called Jalapeno, due to scale up in 2027. It is still raising money and does not yet make a profit; its latest funding round landed at $11 billion against a $30 billion target, according to Token Times. With private funding possibly nearing its limit, a listing may be the only way to pay for the agent boom, and even that offers no guarantee.
