NVIDIA CEO Says Blackwell GPU Design Flaw Affecting Yields Is 100% Fixed
Perhaps the one and only thing that could potentially derail NVIDIA's ability to continue raking money in hand over fist in the AI era would be a design flaw that impacts chip yields. That's exactly what NVIDIA has been contending with in regards to its Blackwell GB200 silicon. However, NVIDIA's chief financial officer Colette Kress revealed during a recent earnings call that the company "executed a change" to improve yields, and now company co-founder and chief executive officer Jensen Huang is doubling down on the claim, saying it's no longer an issue.
In addition, Huang took full responsibility for the flaw, which reportedly caused tensions between NVIDIA and TSMC. Huang refutes those reports, though, calling them "fake news," according to Reuters.
"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," Huang said. "In order to make a Blackwell computer work, seven different types of chips were designed from scratch and had to be ramped into production at the same time.
Huang also credited TSMC with helping NVIDIA "recover from that yield difficulty and resume" manufacturing of Blackwell chips "at an incredible pace."
As far as design flaws go, this was pretty much a best-case scenario for NVIDIA, as it didn't affect the functionality of the silicon and was able to be fixed in a relatively short period of time. Both are important, as AI chip demand is booming and NVIDIA has positioned itself as the dominant player in the AI chip space.
Related, NVIDIA these days generates the bulk of its revenue from its data center division. Last quarter, NVIDIA's data center revenue swelled to record $26.27 billion, which is a 16% sequential increase and a monstrous 154% jump from the same quarter a year prior. In comparison, NVIDIA's gaming solutions raked in $2.88 billion. That's a ton of money, but also a distant second to the data center.
"Blackwell samples are shipping to our partners and customers. Spectrum-X Ethernet for AI and NVIDIA AI Enterprise software are two new product categories achieving significant scale, demonstrating that NVIDIA is a full-stack and data center-scale platform. Across the entire stack and ecosystem, we are helping frontier model makers to consumer internet services, and now enterprises. Generative AI will revolutionize every industry," Jensen said in late August.
While there had been reports of Blackwell shipments slipping into 2025, the expectation now the the design flaw is resolved is that they will ship by the end of this year.
