
Does China Wanting Jensen Huang’s Chips Hold Water?
November 13, 2026
Trouble in The Cloud: AI Storm With Microsoft
November 13, 2026Free Audiovisual Version Here

By C. Rich
I love Grok Bot. Apparently, enough other people love it that Elon Musk now has a problem most businesses would envy right up until the electricity bill arrives.
Build an assistant people actually want to use, and they use it. Make it better, and they give it more work. Let it finish something useful while they do something else, and pretty soon they discover fourteen other things it could be doing. Congratulations: you have invented an employee everybody wants, who never sleeps, and whose lunch consists of a data center.
That is the angle that interests me about Grok Bot. Its success could force Musk into an increasingly consequential decision: how much computing capacity does he devote to his own products, and how much does he sell to other companies?
This is already more than a theoretical dilemma. Musk has acknowledged that serving Grok Bot was costing too much relative to its revenue. Business Insider also reported that the company had reached 800,000 paid weekly users and that Musk anticipated the agent could eventually consume most of its advanced computing capacity. People appear to be finding plenty for it to do.
Meanwhile, there are customers for the machinery behind the magic. The company announced an agreement giving Anthropic access to Colossus 1, with that capacity intended to support Claude subscribers. The infrastructure can serve a competitor’s customers as well as Musk’s own.
Now put yourself in that allocation meeting. One department wants more capacity for an increasingly popular assistant. Another has customers willing to pay for compute. Both would like the same resources, preferably yesterday. Somewhere in the building, an engineer is explaining that enthusiasm cannot be plugged into a transformer.
Every useful action has a physical cost. A bot that searches, reasons, opens applications, checks results, corrects mistakes, and continues working may require many rounds of computation to finish one assignment. How much depends on the task and the system. But the work runs on processors, draws electricity, and produces heat that must be removed. We call it “the cloud” because “an industrial facility sweating through your spreadsheet” tested poorly with marketing.
This is where the public’s enthusiasm runs into thermodynamics. A monthly subscription can feel wonderfully abstract. You pay your money, open the assistant, and start handing over work. The equipment on the other end receives no such exemption from reality. It needs power, cooling, maintenance, and eventually replacement. Your digital teammate may have a delightful personality, but the utility company remains emotionally unavailable.
And success can increase demand faster than efficiency reduces the cost of each task. If a better bot finishes work more reliably, people can trust it with more assignments. Businesses can deploy it across more employees. Tasks that once seemed too troublesome become worth attempting. That is a good thing for users. It also means “we made it more efficient” does not necessarily translate into “we need less total compute.”
My view is that Musk’s decision becomes more interesting as Grok Bot improves. Selling capacity brings revenue from customers who need infrastructure. Keeping capacity for Grok Bot supports a product that could build lasting customer relationships and become central to how people work. There are contracts, technical constraints, and different workloads involved; this is not one giant extension cord Elon can move between rooms. But the underlying choice remains: what earns the best return from scarce capacity, now and over time?
For those of us who love these assistants, the consequences could show up as usage allowances, prices, waiting times, or different levels of service. The public’s appetite does not manufacture additional chips or energize a new power connection. We can fall in love with the assistant considerably faster than anyone can build its next home.
That is why I find Grok Bot’s popularity so revealing. Give people something useful enough, and the question shifts from persuading them to try it to finding the resources to keep it working. I want Musk to give it more compute. Naturally, so does everybody else with a job waiting in the queue. The bot may never need a coffee break. The power plant would still appreciate one.




