TSMC reportedly plots 2027 start date for its 3 nm US fab, but will that be in time to save next-gen GPUs from tariffs?

Maker of most of the world’s cutting-edge chips, TSMC, is reportedly accelerating its plans to produce modern 3 nm chips in the USA. Originally pencilled in for 2028, TSMC is now said to be aiming to pull in production to 2027 in response to tariffs threatened by the Trump administration.

But will that be soon enough for next-gen GPUs?

MoneyDJ (via TrendForce) claims that TSMC is bringing forward its second chip fab in Arizona. TSMC’s alleged new plan is to install equipment in the new facility next year and begin volume production of chips in 2027. That’s a year ahead of TSMC’s current publicly stated schedule.

The reason for the accelerated time table is said to be new tariffs. As we reported recently, President Trump has threatened up to 100% tariffs on chips from Taiwan, which would directly impact TSMC’s output and make imports of components like GPUs massively more expensive.

If TSMC could produce those chips in the US at one of its Arizona fabs, then it would sidestep the tariffs entirely (though its customers would still need to think of their supply chain and packaging). The question then becomes a matter of timing.

TSMC’s first Arizona fab is already cranking out chips on the N4 node, a derivative of N5, broadly referred to as 5 nm, reportedly including CPU dies for AMD. That’s the node also used by both Nvidia for its latest RTX 50 family of GPUs and AMD for its upcoming RDNA 4 graphics cards.

It’s extremely likely that both companies will move to 3 nm or N3 for its next-gen cards, codenamed Rubin for Nvidia and UDNA for AMD. Given brand new GPUs from both outfits have just been released at the beginning of 2025 and that two-year cycles for GPU families are the norm, 2027 for the new 3 nm Arizona fab seems like it could be a good fit.

However, the timings may be a little tighter than that. To allow for a January 2025 launch, for instance, Nvidia will have been manufacturing RTX 50 GPUs many months earlier in order to allow time for the chips to be packaged and fitted to graphics cards in sufficient volumes to supply retailers.

It’s also something of a big ask to expect a brand new fab to be manufacturing very large GPUs from the get go. You might normally expect a new facility to target smaller chips that are less sensitive to yields as the facility builds volumes and irons out production kinks.

Of course, neither Nvidia nor AMD need stick with a two-year schedule. If 100% tariffs or anything even close really are imposed on Taiwan-made chips, it could well be worth delaying release until TSMC’s Arizona fabs can take responsibility for production.

As for the future, TSMC has a third fab planned for Arizona which will produce chips on TSMC’s next-gen nodes, likely to be 2 nm or 16A. However, Fab 3 as it’s known isn’t expected to come online until around 2030.

Anyway, it’s usually a long and expensive process building fabs and bringing up production volumes and yields. So, if TSMC really can pull in its 3 nm Arizona facility from 2028 to 2027, that would be some achievement. And it might just help prevent graphics cards from becoming even more ridiculously expensive.