NVIDIA CEO: Design flaw in Blackwell chips has been fixed and expected to ship in Q4

NVIDIA CEO Jensen Huang said on Wednesday that the design flaw in the company's latest Blackwell AI chips, which had affected production, has been fixed with the help of TSMC. "We had a design bug in Blackwell, it was functional, but the design bug caused very low yield. It was 100% NVIDIA's fault," the CEO said.

NVIDIA released its Blackwell chip in March and had said it would ship in the second quarter, but that was delayed, which could affect customers including Meta, Alphabet and Microsoft.

Earlier this month, Huang said demand for the new Blackwell series processors was "crazy." Last week, Dell said it would soon start shipping devices with NVIDIA's Blackwell artificial intelligence accelerators, and Google said it had started running servers based on Blackwell chips. Earlier, OpenAI and Microsoft had also announced a partnership.

Earlier reports of production delays had led to tensions between NVIDIA and TSMC, but Huang said that was "fake news."

"We designed seven different types of chips from scratch to make Blackwell computers work, and we had to go into production all at the same time," he said. "What TSMC did was help us recover from the yield issues and get Blackwell production back up in an incredible place."

NVIDIA's Blackwell chips combine two pieces of silicon, the size of the company's previous products, into a single component that runs 30 times faster when performing tasks such as providing answers for chatbots.

At a recent Goldman conference, the CEO said the chips would ship in the fourth quarter.

On Wednesday, Huang unveiled a new supercomputer called Gefion in Denmark, which has 1,528 graphics processing units (GPUs) and was built in collaboration with Novo Nordisk Foundation, the Danish Export and Investment Fund and NVIDIA.