Musk: Grok Bot costs 'unsustainably high relative to revenue'

The personal AI agent push is hitting a cost wall just as the industry treats proactive assistants as the next step after chatbots. Meta's Muse is credited with convincing the industry of that direction, but Musk's remarks on Grok Bot and Fry's on Dots point to the same bottleneck: each assistant runs its own cloud computer to execute tasks, and that consumption scales with how always-available the agent becomes.

OpenAI's response—routing tasks to smaller models such as GPT-6.1 Terra instead of the larger GPT-6 Astra—is a compute-efficiency move aimed at bringing Dots beyond the Pro tier. Broader access depends on Fry's expected breakthrough in efficiency, even as more capable agents consume more compute.

The compute constraint extends to infrastructure itself. Local pushback against data centers has led OpenAI to rebrand them "compute campuses," and Anthropic's S-1 amendment lists $895.4 billion in purchase agreements, substantially all related to compute and infrastructure, with inability to secure more compute flagged as a major risk. OpenAI's latest funding round landed at $11 billion, short of its $30 billion target, per the Wall Street Journal—leaving going public as a possible route to fund the agent push.