Nvidia’s Jensen Huang is warming up for GTC 2026 with a hype line designed to make the faithful swoon and everyone else roll their eyes.
Huang said: “We’ve prepared several new chips the world has never seen before. Nothing is easy because all technologies are at their limits.”
Nobody has said exactly what these mystery chips are, which is convenient because it lets the rumour mill do all the work.
One obvious option is more parts spinning out of the Vera Rubin platform Nvidia has been pushing, the kind of derivative lineup that fills gaps, refreshes slides and keeps procurement teams sweating. The other is a bigger reveal tied to the next-gen Feynman chips.
The reality Nvidia is wrestling with is that compute demand shifts fast. Hopper and Blackwell were all about pre-training bragging rights, while newer platforms have leaned hard into inference, where latency and memory bandwidth become the bottlenecks that ruin your day.
If Feynman is where Nvidia goes next, the chatter is that it could enable more SRAM-heavy designs and deeper integration techniques, possibly involving 3D stacking and even LPUs as part of the package, though nobody is confirming anything useful yet.
Huang did the usual ecosystem sermon, because the company stays ahead by buying, partnering and investing across anything that smells like AI infrastructure.
“Nvidia has great partners and amazing startups, and we are investing across the entire AI stack. AI is not just a model; it’s an entire industry encompassing energy, semiconductors, data centres, the cloud, and the applications built on top of it,” he said.
The keynote is mid-March in San Jose, with Nvidia pitching it as the next era of the AI infrastructure race, which usually means a blizzard of new silicon, bigger numbers and more reasons to upgrade your data centre before the last invoice is paid.
