Nvidia Boosts Orders of Compute GPUs for AI: Report

Nvidia is placing more orders for advanced packaging services at TSMC due to increased demand for computing GPUs for AI applications. DigiTimes reports. Nvidia seems so optimistic about the demand for computing GPUs with CoWoS (Chip on Wafer on Substrate) packaging that it has placed additional orders throughout the year.
TSMC has reportedly pledged to process an additional 10,000 CoWoS wafers for Nvidia through 2023 to meet growing demand for its widely used AI chips. The report estimates that this means 1,000 to 2,000 wafers will be added each month for the rest of the year. The article doesn’t reveal which computing GPUs Nvidia plans to ramp up, but currently the company’s lineup includes his A100, A30, H100, and his China-specific A800 and H800. I have a GPU.
Given that TSMC’s monthly CoWoS capacity is 8,000-9,000 wafers, providing Nvidia with an additional 1,000-2,000 wafers per month will measurably increase the utilization rate of the foundry’s advanced packaging equipment. To do. As a result, competitors may find themselves in short supply of CoWoS services.
Due to the widespread adoption of Nvidia’s CUDA platform for AI and other high-performance workloads, there are dozens of large customers leveraging the company’s hardware to run AI applications. Just yesterday, Google announced a new A3 Nvidia H100-based supercomputer offering 26 ExaFLOPS of AI performance. Meanwhile, Microsoft, Oracle and even Elon Musk’s upcoming AI project have procured tens of thousands of Nvidia AI GPUs in recent quarters.
As the use of artificial intelligence continues to grow, demand for Nvidia’s high-end A100 and H100 chips for training and neural networks will grow. That said, the time has come for Nvidia to secure additional packaging power to meet the demand for high-end AI GPUs.
It’s worth noting that Nvidia continues to increase orders for computing GPUs even after losing its ability to ship its most powerful processors to Chinese companies without US government approval. Either the Chinese company is aggressively buying his less powerful A800 and H800 GPUs or demand from US, European and Japanese companies is making up for the drop in shipments to China.