Same technology, different economics: personal AI assistants and business AI workers

Two posts from Elon Musk, three days apart. First: saving money for the distant future is pointless, because super-intelligent abundance is coming. Then: the cost of serving Grok Bot was unsustainably high relative to revenue.

Both can be true. The second is the one worth reading closely, because it says more about how personal AI assistants are built than about whether AI agents work.

## The personal assistant problem

A personal assistant in the cloud is an expensive thing to sell for a flat monthly fee. Every open-ended task runs on someone's GPUs, the heaviest users cost the most, and they all pay the same. Grok Bot, OpenAI's Dots and Meta's Muse are all managing that with tiers and caps.

For personal use, the device may be the better home: a capable model running on your own phone or laptop costs nothing to serve, and your data stays with you. Let's be honest, most people never had a human personal assistant either. The consumer budget for this is small, and it's discretionary.

## Business is a different calculation

The difference is which budget it's measured against. Judged as IT spend, next to a software licence, an AI worker can look expensive. Judged against what it costs today to get a defined piece of work done (calls answered, approvals chased, payments reconciled), it often isn't. The work is bounded, the output can be counted, and a person still checks it.

## Do you need your own GPUs?

The infrastructure question everyone in the Gulf asks: do we need our own GPUs to keep data safe? At today's prices, almost never. A single 8-GPU server lists at roughly $250,000 to $340,000, before power, hosting and the people to run it. A leading provider costs far less, as long as you choose four things carefully:

- the right model for the task, not the biggest one - where it runs: in the UAE when the data has to stay here - how long prompts and outputs are retained - a contractual commitment that your data is never used for training

Get those right, add logs and a human who can override, and you have the compliance case without a data-centre bill.

Where does your organisation file AI: under IT, or under the cost of the work?

---

*I build AI workers for businesses at Xamun, so read this with that in mind. Sources: Musk's post on Grok Bot's serving cost (1 Nov 2026), [as reported by Fortune](https://fortune.com/2026/11/03/grok-bot-is-growing-fast-but-elon-musk-says-its-costs-are-unsustainably-high-compared-to-revenue/). Server prices are public dealer listings, not quotes. For what UAE rules say about where a clinic's data and AI may live, see [Xamun Position Paper No. 12](https://www.xamun.ai/whitepaper-uae-clinic-data-and-ai).*

*First posted on [LinkedIn](https://www.linkedin.com/feed/update/urn:li:activity:7524307334739976192/).*