OpenAI’s Dots Agents Burn Through Compute, Raising Doubts About the Future of Personal AI

Image: Collected

OpenAI’s new Dots personal assistants are proving expensive to run, highlighting a growing industry-wide challenge: always-on AI agents may consume far more computing power than their revenue can support.

At a dinner with reporters in New York on Monday, November 2, Adam Fry, head of ChatGPT’s consumer business, acknowledged that Dots are “on the compute heavy side.” Each assistant runs its own cloud computer to carry out tasks across a user’s connected accounts, making the product unusually demanding even by AI standards.

A Cheaper Model in the Mix

On Tuesday, OpenAI said it was expanding the models behind Dots to include GPT-6.1 Terra, a smaller model that requires less computing power than GPT-6 Astra, which originally powered every Dots task. A new automatic router will now assign different models to different parts of a task, reducing the overall compute burden.

For now, Dots remain limited to OpenAI’s $100-a-month Pro plan. Fry said broader availability depends on a breakthrough in compute efficiency, which he expects will arrive and bring costs down quickly. He also wants Dots to become an always-available presence through features such as “background mode”—an ambition that would require even more computing capacity.

Funding Gets Harder

Money is the other constraint. OpenAI has secured about $11 billion in commitments for its latest funding round, well below its reported $30 billion target, at a $1.25 trillion valuation, according to The Wall Street Journal. The company had initially hoped to raise $1.4 trillion.

The Financial Times reported that OpenAI lowered its goal to $20 billion, with a later offering aimed at wealthy individuals intended to close the remaining gap. OpenAI has not publicly confirmed the figures.

Chips and Competition

OpenAI relies primarily on Nvidia GPUs and plans to scale production of its in-house chip, Jalapeno, in 2027, according to Fortune. Rivals face similar pressure: Anthropic, which uses four types of chips, has signed purchase agreements worth $895.4 billion—“substantially all of which is related to compute and infrastructure,” according to its amended S-1. Even so, the company lists difficulty securing additional compute as a major risk.

Elon Musk made a similar admission this week, saying the cost of running Grok Bot—the personal assistant SpaceX launched in August—is “unsustainably high relative to revenue.” He suggested SpaceX was reducing the computing power assigned to the assistant, causing users to hit their limits sooner.

The industry’s push into agents follows the success of Meta’s Muse assistant. But as Fortune put it: “The more powerful—and constantly running—personal AI agents become, the more compute they consume. Will OpenAI ever be able to catch up with itself?”