Shulou(Shulou.com)11/24 Report--

Nvidia will release its new quarterly results in the early hours of Thursday, when investors will be watching closely to see whether the huge demand for AI products will help offset the decline in global computer hardware sales.

In its last-quarter results, Nvidia expected revenue to grow by about 2gamma and diluted earnings per share to quadruple in the second quarter of fiscal 2024, which ended July 30, thanks to strong demand for its AI processors. At present, Nvidia's AI processor is widely used to train large language models such as ChatGPT.

Image source Pexels it is estimated that Nvidia's AI processor H100 currently meets only half of the market demand. Nvidia's shipments will reach 1.5 million to 2 million next year, at least triple the 500000 expected this year, according to three people familiar with the matter.

The huge demand for Nvidia AI chips is hitting the broader computing equipment market as large buyers invest heavily in AI and accordingly cut back on investment in the general server market.

Foxconn, the world's largest contract electronics manufacturer, last week predicted that demand for AI servers would be very strong in the coming years. But the company also warned that revenue in the overall server market would fall this year.

Lenovo, the world's largest calculator maker by shipments, reported last week that revenue fell 8% year-on-year in the second quarter of 2023, due to weak demand for servers from cloud service providers (CSP) and a shortage of AI processors (GPU).

Yang Yuanqing, CEO of Lenovo, said: "Cloud service providers are shifting their demand from traditional computers to AI servers. Unfortunately, the supply of AI servers is limited by GPU supply."

TSMC last month predicted that demand for AI server chips would grow by nearly 50 per cent a year over the next five years. TSMC is the world's largest contract chip manufacturer by revenue and the exclusive manufacturer of Nvidia AI processors. But TSMC also said this was not enough to offset the downward pressure from the global technology downturn caused by the economic downturn.

Cloud service providers such as Microsoft, Amazon and Google in the United States are shifting their focus to building their own AI infrastructure. Angela Hsiang, vice president of KGI, said: "the overall depressed economic environment poses challenges for cloud service providers in the United States. Because every component in the AI server needs to be upgraded, the cost is very high."

While cloud service providers are expanding aggressively in AI servers, this is not taken into account when setting a total capital expenditure budget, so this expansion is eating into other spending.

Globally, capital spending by cloud service providers is expected to grow by just 8 per cent this year, down from nearly 25 per cent in 2022, as interest rates rise and companies cut spending, according to Counterpoint Research, a research firm.

Figure: global server shipments are expected to fall 6 per cent this year and return to modest growth of 2-3 per cent in 2024, according to TrendForce, a research firm that accounts for AI spending as a share of the world's leading cloud giants. Jibang Consulting also made special mention of Meta and Microsoft. Among them, Meta decided to cut server purchases by more than 10 per cent and invest in AI hardware instead. Microsoft, on the other hand, delayed upgrading its general-purpose server to free up money to invest in AI servers.

Analysts also pointed out that in addition to the shortage of Nvidia chips, other bottlenecks in the supply chain are also dragging down the production of AI hardware products, as well as the benefits of artificial intelligence. "there are also gaps in advanced packaging and high-bandwidth memory (HBM), which limit the production of AI products," said Brady Wang, an analyst at Counterpoint Research.

TSMC has planned to double its CoWoS packaging capacity, but warned that bottlenecks would not be resolved until at least the end of 2024. CoWoS is an advanced packaging technology needed to manufacture Nvidia H100 processors. The two main suppliers of high-bandwidth memory are SK Hynix and Samsung.

Still, executives and analysts say companies in the server supply chain will benefit once the global economy improves and shortages ease.

Spanning AI servers consume much more power than general-purpose servers, which creates the need for different cooling systems and new power specifications.

Foxconn may be one of the main beneficiaries of this shift because it provides everything from components to final assembly, and its subsidiary Foxconn Industrial Internet Co., Ltd. is already the exclusive supplier of Nvidia GPU modules.

For WiWynn, a server-focused subsidiary of Foxconn's rival Weichuang, AI orders account for 50% of its revenue, more than twice as much as last year, according to Goldman Sachs.

In addition, parts suppliers also have strong upside. Goldman Sachs said in a report in June that Gold Circuit Electronic, a maker of printed circuit boards, would generate 38 per cent of its revenue from AI servers, up from less than 3 per cent this year. This is mainly because the amount of printed circuit boards in the AI server is 7 times that of the general-purpose server.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.