Source: Caixin.
① Nvidia's H200 chip will be delivered in large quantities after Q3; ② The GB200 server will also enter mass production in Q4, and the zero-component factory has been requested to complete production preparations in Q3; ③ According to brokerage views, demand for AI computing industry chain is highly prosperous, and the subsequent surge in GB200 is expected to bring further growth momentum.
$NVIDIA (NVDA.US)$Multiple AI GPUs and servers will be shipped soon.
According to Taiwan's Commercial Times, the upstream chip of Nvidia's AI GPU H200 has entered mass production in the late Q2, and large-scale delivery is expected after Q3.
Some supply chain sources pointed out that the H200, which will be delivered in large quantities in the third quarter, is mainly Nvidia's DGX H200, but the order backlog is still concentrated on H100, and the proportion of H200 is still limited.
Nvidia said that the shortage of H200 supply and even the situation of supply shortage may continue until 2025. In fact, in March this year, TSMC's 4nm production capacity was almost full due to the production of H200, which directly reversed the expectation of a decline in the utilization rate of 4nm production capacity at the beginning of the year.
At the same time, with the early launch of Nvidia Blackwell, terminal customers such as Gigabyte and Asus' purchasing willingness for H200 was also affected, and they chose the next generation of B100/B200 series instead. It is reported that B100 has now reached some visibility, and it is expected to be shipped in the first half of next year.
B100 and B200 belong to the Blackwell architecture GPU. The latter is an important part of the 'super chip' GB200 (two B200 GPUs plus a Grace CPU based on Arm architecture). In fact, this chip is usually integrated into an AI server named GB200 NV series for customer use.
Today's MoneyDJ news also reported that Nvidia's GB200 NVL36 server will also enter mass production in Q4, and some component factories have been requested to complete mass production preparations in Q3.
This GB200 server mass production news confirms the statement by Foxconn Industrial Internet Chairman Zheng Hongmeng not long ago. He said that the progress of GB200, whether it is the 36 or 72 version, is smooth, and it is expected to be officially launched this year. It is expected that the contribution of AI to the company's cloud computing revenue will reach 40% in 2024, and AI servers will account for 40% of the global market share.
The latest research report from TrendForce shows that server demand will show a growth trend from the second quarter to the third quarter. In addition to the robust demand for AI servers, it is also driven by bidding projects and storage server demand driven by AI, which will boost second-quarter shipments and continue to run in the third quarter. It is estimated that the third quarter will increase by about 4-5% compared with the second quarter.
According to the institution's prediction, the global AI server (including AI training and AI inference) will exceed 1.6 million units in 2024, with an annual growth rate of 40%.
Zhejiang Securities' research report on June 24 stated that servers with GB200 are expected to start production and delivery in September this year. The demand for related orders in foundries is strong, which is bullish for the performance of the AI server industry chain. The high prosperity of demand in all aspects of the AI computing power industry chain, and the further growth of GB200 volume is expected to bring growth momentum. At the same time, the volume of end-side AI products is expected to promote AI applications and further strengthen the demand for cloud computing power. It is recommended to pay attention to investment opportunities related to optical modules, liquid cooling, and power supplies.
Editor / jayden