Google may be on track to build more AI accelerators in 2028 than Nvidia sells that year, according to a Tom's Hardware report summarizing research from Fubon Research.
The claim centers on Google's TPUs — the AI chips Google designs itself rather than buying from Nvidia. Fubon's projection, as relayed by Tom's Hardware, is that TPU production volume could exceed Nvidia's accelerator sales by 2028.
There's a catch, and it's a physical one: someone has to actually manufacture all those chips. Tom's Hardware reports that hitting those volumes could push Google toward Intel Foundry, Intel's contract chipmaking business, to add manufacturing capacity.
It's worth being clear about what this is. This is an analyst projection reported by a trade publication, not an announcement from Google, Nvidia, or Intel. No company named in the story has confirmed a manufacturing deal, and multi-year chip forecasts routinely move as demand and capacity shift.
Still, the direction of the claim is what makes it interesting. Nvidia has been the default supplier of AI computing power, and Google is its customer as well as, increasingly, its rival. A large cloud provider building enough of its own silicon to rival the market leader's shipments would mark a real change in who controls the supply of AI computing — and a Google–Intel Foundry tie-up would be a notable vote of confidence in Intel's push to manufacture chips for others.
It matters because the companies that make AI chips, and the factories that build them, shape the cost and availability of nearly every AI product downstream.