NVDA
TSM
Feynman is already calling the shots
Nvidia is reportedly leaning harder into its next-gen Feynman AI platform even as Vera Rubin is still ramping. Translation: the company is basically designing the next rocket while the current one is still warming up on the launchpad.
The reported goal is a much more ambitious architecture, with 3D chiplets, TSMC’s SoIC packaging, HBM, and co-packaged optics — the kind of hardware buffet that says, “yes, the AI arms race is still very much on.” Industry chatter also pegs Feynman to TSMC’s upgraded 2nm-class A16 process, which is not exactly casual weekend-craft-level manufacturing.
Why TSMC is feeling the pressure
If Nvidia keeps sprinting, TSMC has to keep up. The report says the foundry is accelerating work at its AP7 site in Chiayi and AP8 in the Southern Taiwan Science Park, while building out capacity for SoIC, CoWoS-L, and next-gen CoPoS tech.
That matters because advanced packaging is one of the bottlenecks in AI hardware right now. If you can’t stack, package, and ship the fancy chips fast enough, all the model-training dreams in the world just sit there looking expensive.
The bigger investor takeaway
This isn’t just a “cool chip roadmap” story. It’s another sign that Nvidia’s demand curve is still so aggressive it’s dictating supplier investment decisions years ahead of launch.
And just to keep the pressure cooker vibe going, Nvidia is also set to report second-quarter results on August 26th, with Bank of America expecting revenue above the company’s own guidance. So yes — the market gets the roadmap drama now, and the numbers show up next.
Big picture: Nvidia’s next move is already rippling through the supply chain, which is great news if you own the AI boom. Less great if you’re a competitor trying to catch up while the finish line keeps moving.