NVIDIA Blackwell AI Chips Reportedly Delayed By Several Months, Culprit Being A Design Flaw

NVIDIA's Blackwell AI chips have reportedly been delayed by months, as industry sources quote that the architecture has become a victim of "design flaws."

NVIDIA's Blackwell Debut Might Be Pushed Into 2025, As The Firm Encounters A Major Roadblock In Mass Production

Update [8/3/2024] - An NVIDIA spokesperson has provided the following update to us:

Related Story NVIDIA To Discontinue GeForce RTX 3060 GPUs Soon As Production Dries Out, Still The Most Popular GPU On Steam

It looks like Team Green's "most successful product" might face a considerable delay, causing drastic effects in the supply chain. The information reports that NVIDIA's Blackwell AI chips are expected to be pushed ahead by the firm for three months or more, citing sources within the supply chain.

\While the media outlet hasn't quoted a specific reason, it is said that NVIDIA is facing design issues with its Blackwell AI chips and has already informed about the situation to its mainstream partners, such as Meta, Microsoft, and Google.

Well, given that the above-mentioned tech giants had ordered billions worth of Blackwell AI chips, the delay is certainly going to have an adverse effect, not just on client relations but also on acquiring the computing power that is said to be gained with NVIDIA's next-gen AI products.

Initially, it was reported that Team Green's Blackwell AI products would debut in the markets by Q4 2024, it looks like the first chips might be pushed to late 2024 but the company is still taking orders and recently confirmed that it will start shipping its Blackwell GPUs for sampling around the globe this week. The company has also reportedly increased its Blackwell orders at TSMC by 25% while the $1.8 million NVL36 servers featuring several GB200 chips will account for the vast majority of sales.

In terms of the actual story behind the design flaw, Ctee reports that NVIDIA's key ally TSMC had seen a design flaw with Blackwell associated with the "interconnect" technique, but since