In times of turmoil, strongmen emerge, which is prominently reflected in the 2023 IC design market.

On May 9th, TrendForce released a list of the top 10 global IC design companies by revenue for 2023. According to the statistics, the combined revenue of these 10 companies is approximately $167.7 billion, a year-on-year increase of 12%. Against the backdrop of a sluggish global economy and semiconductor industry, the key to achieving double-digit percentage growth is NVIDIA, as the company's revenue saw a staggering year-on-year increase of 105%, while most other companies experienced negative growth.

It can be observed that among these 10 companies, only NVIDIA, Broadcom, Will Semiconductor, and MPS have achieved positive annual revenue growth, while the revenue of the other six companies has declined.

Overall, companies that have achieved high growth are mostly closely related to AI server products and services, while those with negative growth are mostly not closely related to AI servers.

Advertisement

NVIDIA and Broadcom are the two companies that have benefited the most from the AI server market.

As giants in the mobile chip industry, Qualcomm and MediaTek have been greatly affected by the sluggish smartphone market.

Qualcomm's revenue for 2023 was $30.913 billion (only QCT is calculated), a decrease of 16% year-on-year. In this situation, the company is actively expanding into the automotive chip market, but it is still difficult to play a key role in revenue in the short term.

MediaTek's revenue for 2023 was $13.888 billion, a decrease of 25% year-on-year, with declines in its smartphone, power management IC, and Smart Edge (smart terminal platform) businesses.Realtek's revenue for 2023 was approximately $3.053 billion, a year-on-year decrease of 19%, primarily affected by factors such as a significant decline in PC market shipments.

01

In turbulent times, innovation is the key to victory

Among the four companies that achieved positive growth, two are ranked in the top three, while the other two are at the bottom two positions. The companies with negative growth are all in the middle of the list. It is evident that in the overall industry downturn, innovation is the key to success, and companies with relatively conventional products and business operations are generally not faring well.

Let's take a look at the performance of the products and business of the top and bottom four companies on the list with good performance.

Thanks to the strong sales of the AI GPU H100, NVIDIA's revenue for 2023 reached as high as $55.268 billion, a 105% increase compared to 2022, and the company currently holds more than 80% of the AI server GPU market. In the second half of 2024 and the beginning of 2025, with the mass production of the upgraded version of H100, H200, as well as the new generation of products B100/B200/GB200, the company's revenue is expected to continue to grow.

Broadcom's revenue for 2023 reached $28.445 billion (only calculating the semiconductor business), with a year-on-year increase of 7%, and AI chip revenue accounts for 15% of its total semiconductor solutions revenue.

In addition to selling chips, a large part of Broadcom's semiconductor business revenue comes from IC design services. As the industry leader in global IC design services, Broadcom has caught a good time. To meet the needs for diversified and cost-effective high-performance AI server chips, CSPs (Cloud Service Providers) represented by Amazon, Google, and Meta are all developing their own AI chips. These big companies lack IC design technology, experience, and talent accumulation. To produce qualified products within a specified time frame, they must seek external help, which is when Broadcom can leverage its strengths.

Taking Google as an example, the company has developed its own TPU. The chip blueprints designed by Google are physically realized by Broadcom. Physical realization is the process of converting logical circuits into circuit diagrams with physical connections. After Broadcom completes the physical layout, it is then sent to the wafer foundry for tape-out. Once the tape-out is successful, the chip officially enters the manufacturing phase, and the entire process requires deep involvement from Broadcom.

In data centers, hundreds or thousands of high-performance processors work together, and their communication has become a significant issue, which is also the main source of performance loss in today's data centers.Broadcom is a communications giant, best known for solving communication bandwidth issues. In the global 50GB/s SerDes market, Broadcom holds a 76% share. Its SerDes interfaces convert low-speed parallel data into high-speed serial data, which is then converted back to parallel data at the receiving end. This process allows data to be transferred efficiently from one TPU to another at high speeds.

With Broadcom's assistance, Google's in-house chip project has accelerated significantly, with the TPU being deployed in data centers just 15 months after its design began.

As the competition for large model markets unfolds rapidly, Google has significantly increased its orders for TPU design services, propelling Broadcom to become the second-largest AI chip manufacturer after NVIDIA. Semianalysis estimates that AI chips will bring in $8 to $9 billion in revenue for Broadcom by 2024.

Not just Google, but also major companies like Meta, Amazon, and Microsoft are increasing their investments in developing their own AI server chips. The demand for outsourcing IC design services is growing, and opportunities for IC design service companies, represented by Broadcom, are becoming more abundant.

Shanghai WeiLai Semiconductor and MPS, which rank lower, also have their unique strengths that have enabled them to achieve year-on-year revenue growth.

WeiLai Semiconductor has three main business segments: image sensors, touch and display solutions, and analog solutions, with other businesses accounting for a very small percentage. In 2023, the company's image sensor revenue was 15.536 billion yuan, accounting for 73.91% of the total annual revenue; semiconductor agency sales revenue was 2.97 billion yuan, representing 14.13%; touch and display solutions business achieved revenue of 1.25 billion yuan, accounting for 5.95%; and analog solutions business revenue was 1.154 billion yuan, representing 5.49%.

In 2023, WeiLai Semiconductor's overseas revenue was 18.331 billion yuan, representing 87.20% of the total revenue; domestic revenue was 2.636 billion yuan, accounting for 12.54%. In the current trade environment, the majority of the company's revenue comes from overseas, which is an important factor in ensuring its year-on-year revenue growth.

MPS's revenue in 2023 was approximately $1.821 billion, a year-on-year increase of 4%, mainly benefiting from automotive, enterprise data, and storage computing businesses, while the company's communication and industrial application product revenue declined.

02

AMD and MarvellIn this list, AMD and Marvell are somewhat unique among the companies featured.

Their uniqueness stems from the fact that while their products and businesses are closely related to data centers and AI systems, their revenues have declined.

AMD's revenue for 2023 was $22.68 billion, a year-on-year decrease of 4%. Due to the sluggish PC market and decreased demand, and since most of the company's divisions and products are targeted at the PC market, despite the growth in revenue from its data center and embedded businesses, it was not enough to offset the decline in PC sales. As a result, AMD is looking to gain more market share in AI.

In the fourth quarter of 2023, AMD launched the AI GPU MI300 series, and the company is pinning its hopes for revenue growth in 2024 on this line of products.

Beyond data centers, AMD is also expanding into the edge AI market. Not long ago, the company introduced two new FPGA products—the second-generation Versal AI Edge and Versal Prime series adaptive SoCs, primarily aimed at edge computing devices such as subsystems in smart cars. Both of these chips are FPGA-based and are an extension of AMD's Xilinx Versal series. It is reported that Subaru has chosen the second-generation Versal AI Edge series for its next-generation ADAS vision system called "EyeSight."

Marvell, ranked sixth on the list, has maintained a relatively stable industry position. Over the years, the company's ranking among global IC design companies has not changed significantly. A key reason for this stability is that the company's products and businesses rarely involve the consumer (C) end and are almost entirely business (B) end, where market fluctuations are relatively smaller. In 2023, Marvell's revenue declined by 7% year-on-year, which, compared to the other five declining companies on this list, is neither the highest nor the lowest, giving an overall impression of being average.

Marvell's main products and businesses are communication interconnects and storage for data centers and cloud computing. The company is a long-standing competitor with Broadcom in the industry, and their products and businesses are very similar, especially in communication. However, over the years, Marvell's communication business has been overshadowed by Broadcom, making it difficult to achieve breakthrough revenue growth. Additionally, Broadcom has a more extensive product and business line, particularly in the area of RF front-end chips, which target the smartphone market and supply to Apple, opening a revenue door for Broadcom. In contrast, Marvell is more focused on data center and cloud computing businesses, with a relatively limited range of products and business lines.

In recent years, amid the AI boom, Nvidia and Broadcom have benefited significantly, with Marvell and AMD following closely behind. However, in the second tier, their revenues are much lower compared to the first tier.

Marvell also has a significant business in IC design services, where the company ranks in the top two globally with Broadcom. However, in terms of IC design service revenue, there is a clear gap between Marvell and Broadcom. Marvell's CEO, Matt Murphy, stated that currently, the company's IC design service gross margin is lower than that of its off-the-shelf commercial products. Over time, the operating profit margins of these two businesses are expected to even out. To achieve this goal, Marvell's approach is to encourage customers to use standard Marvell chips and then add custom interfaces, accelerators, or custom algorithms suitable for the required workload. Customization typically takes 12 to 18 months.

In comparison, Broadcom's IC design service business currently has a profit margin that is on par with its average level, indicating that Marvell's design service business is not yet fully mature, and the service fees charged are lower than those of Broadcom.Based on the aforementioned factors, Marvell's annual revenue has experienced a certain degree of decline, but the extent is not significant.

In order to seize the opportunities in the AI market, especially the "purse" of various Cloud Service Providers (CSPs), Marvell is striving to enhance its IC design service capabilities.

In recent years, Marvell has been developing custom chips for Amazon, which is a significant financial contributor and has brought in considerable revenue for the former. Additionally, recently, Google has developed two Arm server CPUs, one of which, code-named "Maple," is derived from Marvell technology.

03

More Business Opportunities in the AI Wave

There are numerous business opportunities in the AI server and AI PC market. Currently, the most apparent ones are related to processors (GPUs, CPUs) and memory, which are digital and logic chips. In fact, opportunities exist in a much broader space, especially for analog and mixed-signal chip manufacturers who rank lower in the global IC design company rankings. As the AI market expands in the future, there are increasingly more opportunities for businesses, particularly in power management for AI systems.

Due to the power consumption of AI servers being 6 to 8 times higher than that of regular servers, the requirements for power supplies also increase accordingly. At present, general-purpose servers typically require two 800W power supplies, while AI servers may need up to four 1800W power supplies.

Furthermore, in AI servers, to address thermal dissipation and loss issues, ultra-thin applications and power module-like power supplies will become more widespread.

Data centers require an increasing number of AI accelerator cards, which need to be equipped with a large number of processors (xPU). Compared to regular CPUs, xPUs have a multitude of cores that aid in neural network training and AI inference. However, when xPUs perform AI computations and transmit data, they generate significant power consumption. In other words, xPU chips are very power-hungry, and their stringent power consumption requirements pose new challenges for AI accelerator cards.

Currently, the requirements for xPU voltage regulators (VR) are significantly different from standard Point-of-Load (PoL) voltage regulators. Some applications demand providing over 1000A of current to xPU at voltages less than 1V. One of the issues brought about by the trend towards low voltage and high current is how to enhance the rapid response capability to load fluctuations. At this time, it is essential to control power consumption; otherwise, it will be challenging for the system to operate stably.How to reduce the power consumption of AI systems has become an industry challenge. Currently, there are two main approaches: First, reduce the power consumption of the core processors of AI systems; second, optimize the power supply management system to improve the efficiency of power management for AI core processors. However, with the popularization of emerging applications such as AI, the efficiency of traditional computing systems using AC/DC, DC/DC, multi-phase power controllers, and DrMOS power stage combinations has reached its ceiling, necessitating more advanced power management solutions.

These issues cannot be solved by popular manufacturers such as NVIDIA, AMD, and Broadcom; they require solutions from various analog and mixed-signal chip manufacturers. The development space in this area is promising.

04

Impact of 2024 Market Trends on the IC Design Industry

According to WSTS statistics, the global semiconductor industry grew by 6% year-on-year in the fourth quarter of 2023, laying the foundation for growth in 2024.

In 2024, the high-performance computing (HPC) market, represented by AI, will continue to be the locomotive of industry development, posing more and higher demands on processors and memory. Edge computing is also becoming increasingly important, as the demand for local processing and analysis of massive data at the network edge continues to grow. Edge devices require high-performance chips with low power consumption and AI capabilities, and many IC design companies are developing dedicated chips for edge computing to meet this decentralized computing demand.

In addition, the two key drivers of the semiconductor market (smartphones and PCs) are recovering. IDC predicts that after a 5% decline in smartphone shipments in 2023, there will be a 4% increase in 2024, and after a significant 14% drop in 2023, PC shipments will grow by 4% in 2024. At the same time, AI-related applications will penetrate personal devices, with AI smartphones, AI PCs, and other products becoming the new driving force for the development of the chip industry, and IC design companies will welcome more development opportunities.