- According to a report from Morgan Stanley, significant Chinese hyper scalers are buying NVIDIA’s H20 chips for AI computing requirements while abiding by export control laws. While smaller Chinese providers of AI data centers can rent the H100 GPU, bigger hyperscalers are choosing to purchase the H20 to comply with regulatory regulations.
- Analysts said that the H20’s single-chip performance is just 15% of the H100’s, based on their observations during the field trip. Moreover, the single chip performance of the H20 is just 3% of the H100’s when “Performance Density” based on the Float 16 data type is compared.
- Analysts noted that performance may be reduced to 50% of the H100 by joining a cluster of H20 GPUs because of the HBM density and increased networking bandwidth of H20. Therefore, it’s feasible that the cost of the H20 chip will be 50% that of the H100 chip.
- “Taiwanese supply chain members have informed us that H20 volume may reach one million units by 2024,” they said.
- In other news, Morgan Stanely said that its most recent data revealed Huawei’s 910B AI GPU only performs at a level comparable to NVIDIA’s (NASDAQ: NVDA) A100 and that if it weren’t for government subsidies, demand would not seem to be as strong.
- Other local Chinese GPU solutions, however, “seem to be good enough for AI inference,” according to experts. One such solution is that of Tencent-backed AI chip firm Enflame.
- Equipment manufacturers are seeing patterns in China related to high-bandwidth memory (HBM) and advanced packaging localization that resemble those that Nikkei Asia reported in February.
- China’s CXMT plans to go further into high-bandwidth memory technology and is certifying HBM capabilities. The report notes that SJ Semi’s ability to match TSMC’s CoWoS technology’s performance level in 2.5D packaging for AI chips is yet unknown.
Should I purchase or sell NVDA stock?
Are you feeling overloaded with financial data? You’re not alone yourself. The 2024 stock market sometimes resembles a rollercoaster, with opposing viewpoints and analyses appearing wherever you look. When there is a successful, tried-and-true method for determining NVDA’s fair value, it becomes simpler. Using artificial intelligence and machine learning to forecast potentially successful equities distinguishes it from conventional financial instruments that depend on past data. This system evaluates the degree of confidence in its forecasts, projects possible downside risk, and forecasts a stock’s performance about a primary index such as the S&P 500. The nicest thing about our research and stock selections is that we provide you with the essential information—that is, the fair value of each stock—by sorting through the noise. What is the NVDA’s bottom line, then?
