Nvidia CEO Addresses Blackwell AI Chip Production Issues

By Jacob Gronholt-Pedersen and Supantha Mukherjee
COPENHAGEN/STOCKHOLM (Reuters) – Nvidia (NASDAQ:NVDA) CEO Jensen Huang stated on Wednesday that a design flaw affecting production of their latest Blackwell AI chips has been resolved with the assistance of their long-time Taiwanese manufacturing partner, TSMC.

Nvidia's stock fell approximately 2% during early trading. The Blackwell chips were unveiled in March, with shipping originally scheduled for the second quarter, but delays have raised concerns for clients including Meta Platforms (NASDAQ:META), Alphabet (NASDAQ:GOOGL)'s Google, and Microsoft (NASDAQ:MSFT).

"We had a design flaw in Blackwell," Huang acknowledged. "It was functional, but the design flaw caused the yield to be low. It was 100% Nvidia's fault."

Although reports suggested tension between Nvidia and TSMC due to the delay, Huang dismissed those claims as "fake news."

Huang explained that creating a functioning Blackwell computer required designing seven different chip types from scratch and ramping them into production simultaneously. TSMC played a crucial role in overcoming the yield difficulties and accelerating Blackwell's manufacturing pace.

The new Blackwell chips integrate two squares of silicon, significantly improving performance—claiming to be 30 times faster in tasks like chatbot responses than previous offerings. Huang indicated that the chips will now ship in the fourth quarter during a Goldman Sachs conference.

On Wednesday, Huang also launched a new supercomputer named Gefion, featuring 1,528 graphic processing units (GPUs), developed in collaboration with the Novo Nordisk (NYSE:NVO) Foundation and Denmark's Export and Investment Fund.

(This story has been refiled to correct a typo in paragraph 7)