Nvidia Boosts TSMC Orders by 25% to Meet Soaring AI Demand

This year, Nvidia will release a new generation of gaming graphics cards, the GeForce RTX 50 series based on the Blackwell architecture. Additionally, the Blackwell architecture data center, announced earlier, will also be shipped in the second half of the year. These new products will continue to be manufactured by TSMC, utilizing a custom 4NP process, an improved version of the existing 4N process.

According to a report by TrendForce, industry insiders have revealed that TSMC is preparing to start production of Blackwell architecture GPUs. To meet strong customer demand, Nvidia has increased its order volume by 25%. This underscores the unprecedented boom in the artificial intelligence (AI) market and provides significant momentum for TSMC’s performance in the second half of the year, laying the foundation for an optimistic adjustment in annual revenue forecasts.

TSMC plans to hold an earnings call on July 18 to announce its financial results for the second quarter of 2024 and provide guidance for the third quarter. The upcoming mass production of Blackwell architecture GPUs is likely to be a focal point of the meeting, as these are considered among the most powerful AI chips. Tech giants such as Amazon, Dell, Google, Meta, and Microsoft intend to equip their AI servers with Blackwell architecture GPUs, and due to the higher-than-expected demand, Nvidia has had to increase its order volume.

Reports indicate that the shipment volume of terminal server cabinets, including models GB200 NVL72 and GB200 NVL36, has also seen a significant increase, surging from the initially expected 40,000 units to 60,000 units, a 50% growth. The GB200 NVL36 model accounts for the majority, reaching 50,000 units. The average price of the GB200 NVL36 is approximately $1.8 million, while the GB200 NVL72 reaches as high as $3 million, all contributing to TSMC’s revenue growth.