Nvidia’s latest artificial intelligence (AI) chip, Blackwell, has been delayed from the tip of this 12 months to the primary quarter of next 12 months because of a design flaw. This has also caused problems with GPU supply plans for OpenAI, Meta, and xAI.
The Information reported on the 2nd (local time), citing two people in control of chip and server hardware production, that Nvidia’s next AI chip will likely be delayed by greater than three months because of a design flaw. The Blackwell lineup, which was unveiled at GTC in March, includes the ‘B100’ and ‘B200’.
Bloomberg also reported on the identical day that Nvidia notified Microsoft (MS) employees and others last week that the discharge of its latest chip had been delayed.
Those that are greatly affected by this issue are cloud corporations akin to Microsoft, Google, and Amazon, in addition to corporations that give attention to developing next-generation models akin to Meta and xAI. As well as, Microsoft plans to support OpenAI with the acquired GPUs. These corporations have already ordered a whole lot of billions of dollars price of chips.
The brand new chips usually are not expected to be released until the primary quarter of next 12 months. Provided that it’ll take months to get them into actual servers, it’s unlikely that the brand new chips will likely be used for training large language models (LLMs) until the second quarter of next 12 months.
But major AI corporations are already using existing “H100” GPUs to coach next-generation models. OpenAI looks set to release GPT-5 this summer, and Elon Musk’s xAI has built a Memphis data center entirely with H100s for its Grok 3 training without waiting for Blackwell supplies.
Meta is claimed to be planning to secure a complete of 600,000 GPUs by the tip of this 12 months, and it just isn’t certain whether this includes the Blackwell lineup. Nevertheless, Rama 4, which is scheduled to be released early next 12 months, has already began training with the H100.
Nvidia declined to comment directly on the report, but said production would begin later this 12 months.
Tom’s Hardware points out that delaying the launch before the issues occurred was the correct decision. The truth is, AMD’s ‘Ryzen 9000’ and Intel’s thirteenth and 14th generation Core processors were rushed out, but suffered from instability issues.
If the Blackwell line has serious problems after launch, it could possibly be an enormous problem. Because it seems, Blackwell servers cost greater than $3 million per rack, and Nvidia plans to ship 60,000 to 70,000 racks of B200 servers.
Reporter Park Chan cpark@aitimes.com
