Huaxi Securities: NVIDIA announces AI sovereignty with continuously increasing GB200 expectations
Huaxi Securities released a research report stating that NVIDIA has successfully transformed into a server manufacturer, enhancing its commercial value and changing the competitive landscape of the supply chain. The shipment expectations for GB200 are optimistic, and it is expected that OEM manufacturers will further increase their share in NVIDIA's system. North America's four major cloud companies are optimistic about investing in artificial intelligence, with optimistic expectations for NVIDIA's performance. In addition, companies such as Google, Microsoft, and Amazon also have ongoing capital expenditure plans
According to the information from Zhitong Finance and Economics APP, Huaxi Securities published a research report stating that from HGX to DGX to now MGX, NVIDIA (NVDA.US) has successfully transformed from a chip module manufacturer to a server manufacturer. The logic of MGX is similar to that of DGX. NVIDIA independently designs servers and then outsources production to OEM manufacturers, with NVIDIA dominating the industrial chain discourse. Against the backdrop of the revolutionary significance of GB200, OEM manufacturers' share of production in the NVIDIA system is expected to further increase. North America's four clouds are optimistic about Capex budgets, firmly focusing on artificial intelligence; on the supply side, TSMC and HBM manufacturers are providing optimistic guidance for AI this year. In addition, against the backdrop of the trillion-parameter AIGC, shipment expectations for GB200 remain optimistic.
From HGX to MGX, NVIDIA emphasizes absolute sovereignty over the supply system: From HGX to DGX to now MGX, NVIDIA has successfully transformed from a chip module manufacturer to a server manufacturer, further enhancing its own commercial value. Additionally, due to the change in its commercial product form, the competitive landscape of its supply chain has also changed. The logic of MGX is similar to that of DGX. NVIDIA independently designs servers and then outsources production to OEM manufacturers, who then ship to CSP or non-CSP manufacturers, with OEMs holding a larger share in this regard. At this point, NVIDIA dominates the industrial chain discourse. Therefore, OEM manufacturers are seeing positive developments, and against the backdrop of the revolutionary significance of GB200, OEM manufacturers' share of production in the NVIDIA system is expected to further increase.
Optimism remains on the supply and demand sides: North America's four clouds are optimistic about Capex budgets, firmly focusing on artificial intelligence. NVIDIA's performance expectations remain optimistic: Microsoft expects to continue expanding its investment scale and capacity in artificial intelligence by 2024, with a significant increase quarter-on-quarter. Google expects to maintain capital expenditures of no less than $12 billion in the remaining quarters of 2024. Meta has raised its Capex guidance from $30-35 billion to $35-40 billion, with capital expenditures expected to continue increasing in 2025. Amazon expects a significant increase in overall capital expenditures in 2024. On the supply side, TSMC is providing optimistic guidance for AI this year. On April 18, 2024, the company announced its Q1 2024 quarterly report, clearly stating strong demand for artificial intelligence, with AI revenue expected to account for over a dozen percentage points in 2024, and the company expects its revenue compound annual growth rate to reach 50% over the next five years. HBM manufacturers are also providing optimistic guidance.
Continuously raising expectations for GB200: In addition, according to Semianylasis' calculations, in large models like GPT-4, B200 brings about 4 to 7 times performance improvement in GPT-4 reasoning, depending on the selected interaction curve points when quantifying fairness. In contrast, the improvement range of GB200 ranges from about 5 times to 21 times—a larger improvement. Individuals lagging behind in model generation from large companies are expected to purchase GB200 for their large model training and reasoning purposes Therefore, against the backdrop of the trillion-level AIGC parameters, the shipment expectations for GB200 remain optimistic.
Investment Advice:
Seize the core supply chain changes, grasp the elasticity: 1. Changes in supply chain share, such as the increase in OEM share; 2. New entrants in the supply chain system, such as copper links, liquid cooling, power supply, etc.
Beneficiaries:
Contract manufacturing: Foxconn, Inspur Information, etc.;
Power supply: Delta Electronics, etc.;
PCB: Shanghai Electric, Sheng Hong Technology, etc.;
Liquid cooling detection equipment: Chunzhong Technology;
Server operation: Inspur.
Risk Warning: 1. Risks of policies falling short of expectations; 2. AI ethical risks; 3. Risks of technological upgrades falling short of expectations; 4. Risks of escalation in China-US trade frictions