During a Taiwan visit, Nvidia chief executive Jensen Huang said: “We have already seen Rubin on the production line.”

He said: “Nvidia is also manufacturing CPUs, network chips, switches, and many other chips related to Blackwell.”

TSMC is cranking three-nanometre output to keep pace with demand that has not eased since Blackwell, which is about as subtle as a brick through a window.

Rubin will pair with next-gen HBM4, with samples already secured from SK Hynix, Samsung and Micron. That spread should blunt the usual memory bottlenecks that turn big launches into small trickles.

Mass production is targeted for Q3 2026, though the current pace suggests it could be earlier if the fabs behave themselves.

Vera Rubin binds hulking twin GPUs to a Vera CPU with heaps of LPDDR, a design cooked for hyperscalers that measure progress in petaflops rather than patience.

When asked how many wafers Nvidia wanted, TSMC president C.C. Wei called it a “Secret”, which, in Fabland, usually means more than anyone wants to admit.