Jason Hall

All roads don't necessarily have to lead to CUDA and Nvidia.

It's about a year ago that DeepSeek, the big thing happened there that rattled the Western AI capital markets, when we saw that there is a path to build really good, efficient

AI and LLMs, they can go beneath CUDA.

You don't have to necessarily use CUDA to build those systems.

Alphabet, I mentioned earlier, is beginning to sell their TPUs.

As we're starting to see more specific use case architecture, GPUs aren't necessarily always going to be the answer.

As we start to see some more of these products come into the market, AMD is starting to put some legitimate products into the market as well.

I think that there is a case that even if Nvidia continues to win, maybe it's not 90% of the market.

At the end of the day, no matter who wins, TSMC wins.

It is the road that everybody has to take.

to get to AI.

I don't think we can underappreciate that.

Its incentives are built for trust.

It has scale that makes it cheaper for everybody that's trying to build hardware and that needs hardware, even at a scale that it can make more money than anybody else that it's competing with.

I don't know that 2026 is the year.

I think it's going to take longer than that, because I think we've got to get to a point where monetization is more clear.

Because right now, this is just a land race.

This is a land grab.

Companies are trying to establish themselves as having dominant products that answer lots of problems, and that there's a path for not just people, but businesses to buy those AI capabilities

or the SaaS companies to start integrating AI based on those models.