Recently, NVIDIA's latest generation of Blackwell GPUs has sparked unprecedented demand. CEO Jensen Huang revealed at an investor meeting that supplies for the next 12 months have been completely sold out. This phenomenon mirrors the surge in demand for the Hopper GPU a few months ago, indicating a strong market appetite for high-performance computing.
This surge in sales is primarily driven by traditional clients, including major tech companies such as AWS, Google, Meta, Microsoft, Oracle, and CoreWeave. It can be said that every Blackwell GPU produced by NVIDIA and its manufacturing partner TSMC over the next four quarters has been pre-purchased by these giants.
Analyst Joe Moore noted that NVIDIA's share in the AI processor market is expected to further expand by 2025. Large users are rapidly increasing their demand for NVIDIA's solutions, a trend that has been fully confirmed in recent discussions. Meanwhile, Gartner predicts a significant rise in AI chip revenue in 2024.
NVIDIA launched the Blackwell GPU platform in March, highlighting its breakthrough capabilities in data processing, engineering simulation, and electronic design automation. The Blackwell series includes the B200 Tensor Core GPU and the GB200 Grace "superchip," designed specifically for large language model (LLM) inference while significantly reducing energy consumption, which is particularly important in the current industry.
Although NVIDIA has addressed the packaging issues of the B100 and B200 GPUs, there are still some concerns. For instance, whether the world's largest chip foundry, TSMC, can provide sufficient CoWoS-L packaging capacity, and whether memory manufacturers can supply enough HBM3E memory for Blackwell. Notably, NVIDIA has not yet verified Samsung's HBM3E memory, which also affects the supply to some extent.
Despite these challenges, NVIDIA is confident in ramping up Blackwell production by the fourth quarter of 2024 and expects to ship billions of dollars' worth of Blackwell GPUs in the final quarter of this year. The Blackwell architecture is considered the most complex ever built for AI, designed to meet the demands of today's models and provide infrastructure for organizations to handle future larger parameters and performance needs.
At this launch event, NVIDIA also mentioned that they are not only focusing on computational processing needs but also addressing the three major bottlenecks limiting AI development: energy consumption, latency, and precision. The Blackwell architecture excels in performance while enhancing energy efficiency, showcasing NVIDIA's strong capabilities in this field.
In the second quarter, NVIDIA's data center revenue reached $26.3 billion, a significant year-over-year increase of 154%, marking another milestone in the company's continuous growth.
Key Points:
- 💡 Demand for Blackwell GPUs has exploded, with supplies for the next 12 months completely sold out.
- 🔍 Major clients include tech giants like AWS and Google, driving market growth.
- 📈 NVIDIA expects to ship billions of dollars' worth of Blackwell GPUs in 2024, with significant growth in data center revenue.
