Nvidia’s Blackwell AI GPUs reportedly delayed due to design flaw

A design flaw may have resulted in the delay of Blackwell GPUs for professional AI use. According to unofficial reports, Nvidia’s next-generation server cards won’t see mass distribution until the first quarter of 2025.

Nvidia has reportedly already warned Microsoft about the delay in the distribution of the cards, according to anonymous sources who told The Information. The delay is expected to affect not only the Windows owner, but also other big tech companies, such as Google and Meta. As usual, none of the companies involved wanted to comment on the information.

Data Center Dynamics claims that the design flaw was discovered by TSMC, involving the die that connects two Blackwell GPUs in the GB200 units. Nvidia is reportedly re-adjusting the design, which would result in another production test with TSMC before resuming mass manufacturing.

To get around the delays, Nvidia is reportedly considering a new version of the GB200, with just one Blackwell GPU.

More than one reason for the delay

Adding to the rumors about the delay, Dylan Patel took to Twitter to talk about another possible reason for the delays. The SemiAnalysis analyst claims that supply will be much lower than anticipated in the first half of 2025:

Patel, however, does not mention any design flaws on Nvidia’s part. Instead, he says it is a transition from the CoWoS technology used by TSMC in chip packaging. The Taiwanese manufacturer currently focuses on CoWoS-S, while Blackwell chips require CoWoS-L, which is still being installed in factories.

Blackwell raises strong expectations

Nvidia’s dominant position in the AI ​​GPU market has led to sky-high expectations for its Blackwell generation. That means a lot of money, of course.

The leaks also hint at the manufacturer’s production strategy for its GB200 chips. TechPowerUp reports that Nvidia has prioritized NVL72 units over NVL36. As the name suggests, NVL72 units incorporate 72 GPUs, while NVL36 units use 36.

Read more:

Although there is no consensus on the justifications for the delay in the rumors, it seems that Nvidia will really need to wait a little longer to make its new “industrial revolution”.