What Does Elon Musk Think About AI on Smartphones?

Author: Peter Jarich, Head of GSMA Intelligence

Meta’s AI supercluster. Indonesia’s new AI Center of Excellence. Google’s $3 billion hydropower deal with Brookfield. A quick scan of my AI newsfeed perfectly reflects the “bigger is better” and “big brains” view of AI compute espoused by Jensen Huang back at GTC Paris in June. And, yet, on the topic of AI scale and distribution, I’ve been thinking about the comments and views of another tech bajillionaire: Elon Musk. 

Now, to be fair, the comments on my mind were completely unconnected and not necessarily about AI. But the man knows a little bit about the topic and, frankly, if you throw out enough wacky ideas, there are bound to be some connections that materialize. And, intended or otherwise, I think that’s the case with…

• The Distributed Tesla Supercloud. On a Tesla earnings call last year, Musk mused about the latent compute power of the company’s vehicles when idle. “If you imagine the future perhaps where there’s a fleet of 100 million Teslas and on average, they’ve got like maybe a kilowatt of inference compute. That’s 100 gigawatts of inference compute, distributed all around the world.” Interesting thought exercise? Sure. Commercially or technically workable? Not clear. Regardless, it got the fanboys all riled up.
• The Magnitude of Big Numbers. In an effort to call out his objection to the spending in President Trump’s Big Beautiful (Budget) Bill, Musk took to X to highlight the difference between one million, one billion, and one trillion. The short story: it’s a difference of weeks vs. years vs. millennia.

Where’s the connection? 

We’re not in a world where there are 100 million Teslas on the road. It’s closer to 5 million at the end of 2024. With AI factory plans looking at 10,000 GPUs per, we might get to a point where there are just as many GPUs in them worldwide. And that all compares with about 5 billion smartphones in operator as of mid-year 2025 (rounded up from 4.5 to make the math easy). With top-end smartphone processors claiming 40+ TOPS of AI processing and Nvidia’s H200 coming in at 4,000 (FP8 or INT8 Tensor Core) I suspect you can see where I’m going. 

A universe of smartphones running high-end silicon would deliver 10X the AI compute capacity of some future state where a bunch of AI factories are brimming with top=-of-the-line GPUs.

I also suspect you have some questions for me. Smartphone NPUs and datacentre GPUs are good at different tasks, can you really compare them? Very few of today’s smartphones actually sport high-end, AI-optimized silicon; how quickly will that change? Will the continuing democratization of Aid drive AI capabilities into the low-end? What’s the trajectory for performance of datacentre-focused GPUs compared with smartphone silicon – one is surely going to outpace the other? If the idea of Elon pooling a bunch of Tesla compute seems wacky, how feasible would it be to tap into billions of smartphones? 

You’re 100% right on all of this. 80% if I dock you for a bad attitude. And that’s the point (the questions not the attitude). 

Most of today’s on-device AI focus has been about use cases that can sell new phones. In a market where sales are growing 1% year-on-year, this is totally logical; customers need a reason to splash some cash on a new slab of glass. But as more and more smartphones include increasingly powerful AI capabilities, we cannot deny the power this puts into the hands of people, particularly as workloads move from training to inference. How will this be put to use? Will it impact demand for AI Factory compute? What will it mean for network capacity? These are the questions my customers are asking. The ones which will define network investment strategies, smartphone roadmaps and potential new monetization paths. 

As to the question at the top of this page. No, I don’t know what Elon thinks. If he has thought about the intersection of AI and smartphones, I suspect his viewpoint is similar to most things he opines on; somewhere between brilliant and hazy. Apropos for the topic, I suppose.