Nvidia’s CEO, Jensen Huang, announced on Wednesday that a design flaw in the company’s latest Blackwell AI chips, which had affected production, has been resolved. The fix came with the assistance of TSMC, Nvidia’s long-standing Taiwanese manufacturing partner.
The issue had caused Nvidia’s shares to drop by around 2% in early trading. The Blackwell chips, initially unveiled in March, were expected to ship in the second quarter but were delayed. This delay potentially impacted major clients such as Meta Platforms, Alphabet’s Google, and Microsoft.
Huang acknowledged the flaw, explaining, “We had a design flaw in Blackwell. It was functional, but the design flaw caused the yield to be low. It was 100% Nvidia’s fault.” While some media reports suggested tensions between Nvidia and TSMC due to the production delay, Huang dismissed these reports as “fake news.”
Huang further elaborated on the complexity of the Blackwell chips, which required seven different types of chips to be designed from scratch and ramped into production simultaneously. He credited TSMC with helping Nvidia recover from the yield issue and resume production at an accelerated pace.
The Blackwell chips represent a significant advancement in AI processing, combining two silicon squares into a single component that delivers 30 times the speed of previous models for tasks such as powering chatbots. The chips are now expected to ship in the fourth quarter, according to Huang’s remarks at a recent Goldman Sachs conference.
Huang made the announcement while in Denmark for the launch of a new supercomputer, Gefion. This machine, built in collaboration with the Novo Nordisk Foundation, Denmark’s Export and Investment Fund, and Nvidia, features 1,528 graphic processing units (GPUs).
