Nvidia has been riding high due to the sales of its GPUs made for AI data centers. However, a new report claims that the release of one of its upcoming AI chips could be delayed by three months or even longer.
The report comes from The Information, citing unnamed sources. The story says Nvidia has informed its many AI GPU customers, including Microsoft, that its Blackwell B200 chips have a design flaw that will cause the company to delay the mass production and shipments of the processor.
Nvidia first announced the Blackwell GPU architecture in March during its GTC conference. The Blackwell B200 processor will pack in 208 billion transistors. The GPU could offer performance of up to 20 petaflops with 4-bit floating point (4FP) support. They are designed to replace the Hopper H100 GPUs that are currently used by Microsoft, OpenAI, Google, Meta, and others in their AI data centers.
The report claims that this design flaw was found very late in the production of the Blackwell B200 chips. Nvidia is reportedly going through a new round of production tests with the chip. It was supposed to begin mass shipments of the GPU by the end of the year, but The Information says Nvidia is now planning on mass production to start in the first quarter of 2025.
Officially, a spokesperson for Nvidia told The Information that "production is on track to ramp" sometime later in 2024 for the Blackwell chips, but added, "We don’t comment on rumors.”
Nvidia"s many AI chip customers have reportedly already ordered "tens of billions of dollars” of the upcoming Blackwell B200 GPUs, according to the Information. A lengthy delay in the release of these chips could affect not only Nvidia in terms of revenue but also its customers, who have likely been counting on the big performance boosts from the chip to help add new and improved AI features.