Headline: Nvidia Overcomes Production Challenges for Blackwell AI Chip

Nvidia CEO Jensen Huang announced today that the design flaw affecting the production of the company's Blackwell AI chips has been resolved. Introduced in March and initially scheduled for shipment in the second quarter, the chips faced delays due to the discovered issue, which could have had potential implications for major clients like Meta Platforms, Google, and Microsoft. Huang took full responsibility for the error, stating, "We had a design flaw in Blackwell. It was functional, but the design flaw caused low yield. This was entirely Nvidia's fault." Addressing rumors of tension between Nvidia and its production partner TSMC, he labeled such reports as "fake news."

Huang also elaborated on the complexity of the Blackwell chips, which comprise seven different types of chips designed from scratch and manufactured concurrently. He praised TSMC's role in overcoming the production challenges, saying, "What TSMC did was help us overcome this yield challenge and restart Blackwell production at an incredible level."

The Blackwell chips represent a significant advancement in Nvidia's technology. By combining two silicon squares into a single component, these chips deliver performance up to 30 times faster than previous models for tasks like running chatbots. Speaking at a Goldman Sachs conference, Huang noted that Blackwell chips are now expected to be shipped in the fourth quarter.

In related news, Huang was in Denmark today to introduce a new supercomputer, Gefion, which features 1,528 GPUs. The project is a collaboration between the Novo Nordisk Foundation, Denmark's Export and Investment Fund, and Nvidia.