The business world is increasingly relying on AI to be the next big technological advancement, and NVIDIA is one of the key players that is powering this revolution.
With its market capitalization hitting $3 trillion earlier this year, the California-based chip manufacturer has witnessed a meteoric rise since the rise of GenAI. NVIDIA's revenue has surged dramatically, increasing by 262% year over year, highlighting the massive demand for its AI chips. However, the remarkable rise of NVIDIA might be experiencing a serious obstacle.
According to a report by The Information, NVIDIA AI chips will be delayed by three months or more due to design flaws that were discovered “unusually late in the production process.”
Customers who have ordered the new Blackwell flagship line of chips were expected to receive their shipments later this year but now may have to wait until 2025. The delay is expected to cause major disruption for customers such as Google, Microsoft, and Meta who already ordered tens of billions of dollars worth of NVIDIA chips.
Several NVIDIA customers are expected to launch new AI products and set up data centers by early next year, and this delay can cause significant disruption to their plans.
Google is estimated to have placed an order of more than 400,000 GB200 chips. Meta has also placed a similar order, while Microsoft wants at least 55,000 chips for OpenAI by the first quarter of next year.
The Blackwell AI chips are the successor to the hugely popular NVIDIA H100 chips, which are powering the majority of GenAI applications around the globe. NVIDIA unveiled the Blackwell GPUs in March, claiming they deliver a performance boost of up to 30 times compared to the H100 while reducing energy consumption by as much as 25% on some workloads.
The design flaw was reportedly identified by NVIDIA's chip producer Taiwan Semiconductor Manufacturing Company (TSMC). A fresh set of test runs are being conducted on the AI chips.
According to The Information, the design flaw is with the processor die that connects the Blackwell GPUs on a single NVIDIA GB200 Superchip. NVIDIA is now working on redesigning the die and will likely need a few months before it can move to production testing with TSMC.
There are reports that NVIDIA is considering producing a single GPU version of the Blackwell series to fulfill its commitment to some customers. However, the performance impact of such a design would be risky.
NVIDIA may prefer to delay the shipments rather than risk shipping out faulty or underperforming products, which is what appears to have happened with Intel's 13th and 14th Gen Core processors. The instability issues of the Intel processors have plagued gamers since December 2023 and have become a major headache for the tech giant.
While NVIDIA may be able to afford the delay, any performance failures of its highly anticipated Blackwell chips could be catastrophic. Not only would it be a disaster for its reputation, but also be extremely costly.
GB200 compute tray featuring two Grace Blackwell Superchips (Image courtesy Nvidia)
A single GB200 Blackwell superchip can cost up to $70,000, while a complete server rack can cost more than $3 million. Given the volume of chips and servers NVIDIA has committed to producing, any issues with the Blackwell chips could have severe financial repercussions for the company.
The delay in the AI chip arrives as NVIDIA faces another setback. The chip manufacturer is subject to investigation from the U.S. Department of Justice over its AI practices. The investigation is probing into claims that NVIDIA unfairly pressured cloud computing companies to buy its AI chips, and is also exploring potential antitrust violations related to NVIDIA’s acquisition of the AI startup Run:ai.
Several of NVIDIA’s customers are investing in hardware research to build their own AI products in-house. As a result, NVIDIA must minimize the impact of the AI chip delay to maintain its leading position in the market. There is a lot at stake for NVIDIA, and it must carefully navigate these challenges to ensure it continues to dominate the AI hardware industry.
Related Items
NVIDIA Prepares New AI Chip for China Amid Ongoing US Export Controls
Google’s Controversial AI Chip Paper Under Scrutiny Again
Nvidia’s Superlove: First Superchip, Now Supermode
