With the release of the first quarter reports of semiconductor companies and the convening of earnings calls one after another, the current market picture of semiconductors has gradually become clear. On the one hand, the strong demand for generative artificial intelligence such as AI computing chips continues to lead, and some chip categories and manufacturing processes are in short supply, and the "visibility" of orders can reach 2025, while consumer electronics, automobiles, traditional servers and other markets still need to recover; On the other hand, there is a "temperature difference" between the Chinese market and the market in other regions, and the demand for mid-to-high-end smartphones, electric vehicles, and the Internet of Things in the Chinese market has been emphasized by international and domestic manufacturers many times. Linking the above characteristics and trends, we can see that the demand of the semiconductor market is staging a "quartet".
First, the demand for artificial intelligence computing power leads the semiconductor market. The beneficiaries are the industrial chain around artificial intelligence computing power acceleration chips, including Fabless, which designs and sells computing power chips such as NVIDIA and AMD, HBM (high-bandwidth memory) suppliers such as SK hynix, Samsung, and Micron, foundries such as TSMC, and advanced packaging manufacturers.
Specifically, AMD's AI accelerator MI300 has exceeded $1 billion in cumulative sales since its launch in the fourth quarter of last year, driving the company's data center GPU sales to a record high in the first quarter. Intel expects Gaudi3, an AI accelerator announced in April this year, to bring in more than $500 million in revenue in the second half of the year.
HBM, as the core of AI accelerators, has also seen a surge in demand driven by AI, driving SK hynix's first-quarter operating profit to soar by 734% quarter-on-quarter, while Samsung's first-quarter storage revenue increased by 96% year-on-year. It is reported that SK hynix's HBM production capacity in 2025 is close to being sold out, Micron's HBM production capacity in 2024 has been allocated, and most of the HBM production capacity in 2025 has also been booked.
TSMC expects AI processor-related revenue for servers to more than double this year and contribute more than 20% of TSMC's revenue by 2028 due to its reliance on advanced manufacturing processes. At the same time, AI processors have also driven TSMC's demand for CoWoS packaging, and spilled over to TSMC's OAST (semiconductor packaging and testing manufacturer) partners. TSMC expects to more than double its CoWoS production capacity in 2024 compared to last year, but it is still not enough to meet customer demand, and will do its best to increase production capacity with OSAT partners.
Second, the demand for AI infrastructure and edge AI has brought new increments. This part of the demand does not rely on cutting-edge chips, and related suppliers, foundries, and packaging companies with mature process semiconductors can also benefit.
On the one hand, AI servers require high-speed IO chips and memory controllers that handle data transmission, as well as power management chips that optimize the power consumption of computing and memory units, so as to bring increments to mature process products. On the other hand, the innovative applications brought about by edge AI are also accelerating, which is good for SoCs, MCUs, memory and other chips required for edge machine learning, as well as related manufacturing processes and 3D packaging. UMC, a mature process foundry, expects its potential market share of AI semiconductors to reach 10%~20%.
The third is the leading demand of the Chinese market. The demand for mid-to-high-end smartphones, electric vehicles, industrial and IoT has played a significant role in boosting the revenue of international and domestic semiconductor manufacturers.
In terms of smartphones, China's mobile phone production is recovering. According to data from the Ministry of Industry and Information Technology, China's smartphone production in the first quarter was 276 million units, a year-on-year increase of 16.7%. Cristiano Amon, President and CEO of Qualcomm, said on the earnings call that the proportion of high-end and mid-to-high-end products in China's mobile phone market continues to increase, which has become the driving force for Qualcomm's performance growth. According to the data, Qualcomm's revenue from Chinese OEMs (original equipment manufacturers) increased by more than 40% year-on-year in the first half of fiscal year 2024 (ending March 24, 2024).
In terms of electric vehicles, the retail penetration rate of electric vehicles in the Chinese market has grown steadily, bringing market opportunities to international and domestic semiconductor manufacturers. In an analyst call for the second fiscal quarter (Q1) 2024, Infineon CEO Jochen Hanebeck called the launch of the Xiaomi SU7 series "one of the highlights of the quarter". Infineon will provide silicon carbide power modules and chips for the Xiaomi SU7 family, as well as a system solution with more than 60 different components, which is expected to be used in more than 10 applications. Hassane El-Khoury, president and CEO of onsemi, who has just returned from the 2024 Beijing Auto Show, also said that it will continue to expand the market share of automotive-grade products for the top 10 Chinese automotive OEMs.
The recovery and development of China's manufacturing industry has also driven the demand for industry and IoT. NXP President and CEO Kurt Sievers said in the first quarter earnings call that the industrial and IoT terminal demand in the Chinese market is improving, NXP's industrial IoT business is mainly oriented to the Chinese market, and also noted that China's manufacturing purchasing managers' index (PMI) situation is better, NXP will continue the distribution channel of the IoT business to occupy a favorable competitive position in the second half of the year.
Fourth, other markets that are still in the stage of destocking. Including consumer electronics, automobiles, traditional servers, etc., although the inventory level is still high, there are also some bright spots, such as this year's America's Cup, Asian Cup, European Cup, Olympic Games and other sports events, driven by sports events such as set-top boxes, TVs and other consumer goods will usher in an increase, driving the demand for related semiconductor stocking; Driven by mobile phone replacement and automotive ADAS, the demand for high-resolution CIS in some regional markets has increased. The revenue of RF communication and radio frequency identification related products of European semiconductor companies has improved.
The "quartet" of demand in the semiconductor market means that the future market demand will be further structurally differentiated, which will also deeply test the ability of semiconductor companies to respond to market changes.
For advanced computing chip manufacturers, although the wind of AI is strong, it is not constant and stable. As SMIC co-CEO Zhao Haijun said on the May 10 earnings call, AI-related capacity demand is an "uneven process" – uneven demand from year to year, and uneven demand from individual companies. For example, if IDMs and foundries build capacity based solely on AI market demand or expectations, rather than specific customer needs, there will be an oversupply. This oversupply can be regional or cyclical, such as capacity being fully loaded in years of high demand and being constrained at other times, increasing operating costs.
Leading mature process foundries such as SMIC, UMC, GLOBALFOUNDRIES, and Hua Hong Semiconductor all lowered their ASPs slightly in the first quarter. For related enterprises, the first thing to focus on is the hot application of the most dynamic regional market. Secondly, it is necessary to form a good interactive relationship with downstream customer enterprises, and for local customers, it is necessary to use the advantages of being close to the market to form customized service capabilities, or to intervene in the early stage of product planning for customers and carry out joint research and development. For example, power management chip suppliers can consider further increasing power density, or applying silicon carbide and gallium nitride technologies to meet AI's demand for higher power supply power and energy conservation and carbon reduction in digital centers. MCU manufacturers can consider integrating machine learning acceleration capabilities into their products to develop the demand for edge AI devices and enhance product competitiveness.
The cockpit and driver integration puts forward the "core" requirements
AI phones are on fire, and these chips are assisting!
