550,000 Units Contracted This Year... 1.5 to 2 Million Units Expected Next Year
Prices Soar to $40,000 Amid Shortage Crisis
In its Q2 earnings announcement this year, Nvidia achieved a groundbreaking 88% year-over-year revenue growth and plans to triple its GPU shipments next year. All GPUs produced this year have already been sold out.
According to the UK media outlet Financial Times (FT) on the 23rd (local time), Nvidia plans to set the shipment volume of next-generation GPUs, which will be outsourced to Taiwan's TSMC next year, at 1.5 to 2 million units.
Nvidia's latest AI and data center GPU, the 'H100,' will have 550,000 units produced this year. However, due to the surge in demand driven by the popularity of generative AI, all units have already been sold out through the end of this year. This means GPU shipments will increase at least threefold next year.
The severe shortage of H100 GPUs has even sparked a 'stockpiling' competition among companies and institutions. Government agencies are also eager to secure H100 GPUs in advance.
Previously, the Kingdom of Saudi Arabia and the United Arab Emirates (UAE) governments agreed to purchase thousands of Nvidia GPUs, and recently, the UK government reportedly plans to spend ?100 million (approximately 170 billion KRW) to procure 5,000 H100 GPUs.
The initial price of the H100 was set at $36,000 (approximately 48 million KRW). However, as competition among consumers to secure the limited GPUs intensified, prices have now approached $40,000 (53.4 million KRW). On secondhand online platforms such as eBay, some chips have even been resold for $50,000 (66.75 million KRW) each.
The unexpected price surge has also worked to Nvidia's advantage. In the earnings report released on the 24th, Nvidia posted an 88% year-over-year revenue increase and an earnings per share (EPS) of $2.70, far exceeding Wall Street expectations in an earnings surprise. Nvidia even projected Q3 revenue of $16 billion (approximately 21.36 trillion KRW), 2% higher than Q2.
NVIDIA's GH200, set to be released next year, connects the latest ARM CPU, Grace Superchip, with the H100 GPU to provide immense AI processing power. [Image source=NVIDIA]
Meanwhile, Nvidia is expected to launch the 'GH200' next year, a computing system specialized for large AI model computations. The GH200 is a full-fledged AI computer that combines the so-called 'super CPU' Grace chip with the H100 GPU.
In particular, the Grace chip is expected to deliver overwhelming AI processing performance compared to competitors by integrating 144 of the latest CPU cores, 'Neoverse V2,' designed by the UK semiconductor company ARM.
Additionally, it is equipped with the next-generation high-bandwidth memory (HBM) product, HBM3E, to further increase data transfer speeds.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
