Nvidia is taking full advantage of the AI boom. Their latest Blackwell GPUs are in such high demand that they’re sold out for the next year. Jensen Huang, Nvidia’s CEO, shared this with analysts from Morgan Stanley.
This isn’t the first time Nvidia’s faced such demand. Morgan Stanley’s Joe Moore noted that a similar rush occurred with Hopper GPUs a few quarters back. Major tech players like AWS, Google, Meta, Microsoft, Oracle, and CoreWeave are scooping up every Blackwell GPU that Nvidia and their partner TSMC can produce.
This soaring interest strengthens Nvidia’s position in the AI processor game, even as competitors like AMD and Intel are vying for market share. Moore believes that in 2025, Nvidia will likely gain even more ground in AI processors, particularly as big users of custom silicon ramp up their adoption of Nvidia’s solutions. This sentiment is backed by insights shared during recent discussions.
The hype for AI chips is on the rise. Gartner recently forecasted a sharp increase in AI chip revenue by 2024.
Nvidia launched the Blackwell GPU platform in March, promoting its potential to revolutionize areas like data processing, engineering simulation, and generative AI. The platform includes the B200 Tensor Core GPU and the GB200 Grace “super chip,” crafted to manage the heavy lifting of large language model inference while also addressing the pressing concern of energy consumption. Nvidia designed these chips to enhance system uptime and affordability for extensive AI deployments by boosting diagnostic capabilities and reliability forecasting.
While Nvidia has tackled initial packaging challenges with its GPUs, questions linger about TSMC’s ability to meet CoWoS-L capacity demands. Memory supply is another concern, particularly regarding HBM3E memory for Blackwell GPUs. As of now, Nvidia hasn’t qualified Samsung’s HBM3E memory for these processors.
Despite acknowledging low yields for its Blackwell products back in August and a need for production tweaks on the B200 processor, Nvidia remains optimistic. They expect to ramp up production later this year and anticipate billions in sales from Blackwell GPUs in Q4.
The Blackwell architecture stands out as one of Nvidia’s most sophisticated designs for AI, aimed at future-proofing infrastructure for models with more complex performance needs. The company is also targeting key barriers in AI today, focusing on energy efficiency, latency, and precision. In the second quarter, Nvidia announced impressive data center revenue of $26.3 billion, reflecting a 154 percent increase from the same period last year.
