NVIDIA (NVDA, Financial) CEO Jensen Huang announced the resolution of a design flaw in the company's latest Blackwell AI chip, achieved with the assistance of Taiwan Semiconductor Manufacturing Company (TSMC). The defect previously impacted production, leading to lower yield rates. Huang acknowledged, "This was entirely NVIDIA's fault."
Originally launched in March, the Blackwell chip's shipment was delayed from the planned second quarter, affecting clients like Meta (META), Alphabet (GOOGL), and Microsoft (MSFT). Recently, Huang described market demand for the Blackwell series processors as "crazy."
Last week, Dell revealed plans to ship devices featuring NVIDIA's Blackwell AI accelerators soon. Google has already started running servers based on these chips, and earlier announcements noted collaborations between OpenAI and Microsoft in this area.
Despite external reports suggesting tension between NVIDIA and TSMC due to production delays, Huang dismissed these claims as "fake news." He stated that NVIDIA had redesigned seven different types of chips for the Blackwell computer, with simultaneous production necessary. TSMC's role was crucial in overcoming production challenges and resuming Blackwell's manufacturing at an impressive scale.
The Blackwell chip combines two silicon pieces, enhancing performance by 30 times for tasks like chatbot interactions. At a recent Goldman Sachs conference, Huang confirmed that these chips will ship in the fourth quarter. In Denmark, Huang also unveiled a new supercomputer named Gefion, featuring 1,528 GPUs, built in collaboration with the Novo Nordisk Foundation, the Danish Export and Investment Fund, and NVIDIA.
