Personal AI agents that proactively handle tasks are drawing industry attention as a potential next major AI product, but their growth hinges on whether companies can secure computing power at an affordable cost.

OpenAI’s Dots illustrate the challenge: each assistant runs on its own cloud computer to manage tasks such as email and ticket bookings, making the service compute-intensive, according to Adam Fry.

Dots are currently available only to customers paying $100 a month for a Pro subscription, and OpenAI plans a background mode that would let assistants stay active and run errands.

To curb compute use, OpenAI is routing Dots requests across models, including the smaller GPT-6.1 Terra, instead of using the larger GPT-6 Astra for every task.

Fry expects computing efficiency to improve and lower costs as models get better, although agents that run continuously could drive up demand for compute.

The cost squeeze is already apparent at xAI: Elon Musk said running Grok Bot costs more than its revenue can sustainably support and suggested users may hit usage limits sooner as computing allocations shrink.

Anthropic’s amended S-1 lists $895.4 billion in long- and short-term purchase agreements, substantially all related to compute and infrastructure, and identifies difficulty securing more compute as a major business risk.

OpenAI faces similar infrastructure and financing pressures, relying largely on Nvidia GPUs while developing its own Jalapeno chip, with production scaling planned for 2027; its latest private funding round reportedly raised $11 billion against a $30 billion target.

Data-center expansion also faces community opposition that could delay or block permits, prompting OpenAI to call its facilities “compute campuses” and pledge to ease pressure on local power grids and avoid taking scarce electricity from families.

At the same time, OpenAI is building a broader consumer strategy: Fry said ChatGPT voice mode has 250 million weekly active users and search-intent queries equal one-fifth of Google’s volume, while consumer devices designed by Jony Ive are expected to debut this year.

Critics highlight nearer-term risks including cybersecurity failures, fraud, poor advice and concentrated industry power, while investors question whether efficiency gains can offset the cost of increasingly capable agents and reduce the need for more capital.

Even large private funding rounds may not cover the compute costs of personal agents, making a public offering a possible source of capital without guaranteeing that the economics will work.