Author of this article: Zhang Yifan

Editor: Shen Siqi

The weather is good today The weather is good today.

With the increase of chip design and system complexity, technology giants will cooperate more with ASIC manufacturers.

Morgan Stanley predicts that the size of the high-end customized ASIC chip market will be between $20 billion and $30 billion with a compound annual growth rate (CAGR) of 20%.

As leaders in the ASIC market, Broadcom and Marvell may become the winners.

Currently, the two companies hold over 60% of the market share.

Broadcom holds 55-60% of the market share, while Marvell closely follows with 13-15%.

In addition, with cloud providers and large OEMs entering the ASIC market, the supply chain may shift from Nvidia's dominance to diversification.

1. ASIC and computing cards

The competition between ASICs and general-purpose computing cards has existed for a long time, and it has intensified with the entry of cloud providers and large OEMs.

Currently, the main manufacturer of general-purpose computing cards is Nvidia, which holds nearly 70% of the AI computing market share; the main manufacturers of ASICs are Broadcom and Marvell, which together hold over 60% of the ASIC market share.

ASIC has the advantages of high performance, low power consumption, cost-effectiveness, confidentiality, security, and reducing circuit board size in specific task scenarios.

This advantage is mainly due to:

• ASIC: an integrated circuit designed for specific applications, optimized for specific tasks, usually with higher performance, lower power consumption, and low advantages than GPUs on these tasks. However, the disadvantage is that it is not universal.

• General-purpose computing cards: providing standardized high computing performance, but not focusing on specific task scenarios, suitable for a wide range of applications, with universality;

In other words, ASIC sacrifices universality for high performance in specific scenarios; general-purpose computing cards have universality, but their performance is inferior to ASIC in specific scenarios.

In fact, different computing card customers have different requirements.

Cloud providers may focus more on elastic computing, while enterprises may be more concerned about cluster computing power, etc. In the face of specific needs, ASIC is more advantageous than standard computing cards and more in line with customers' usage scenarios.

Currently, Google, Meta, Microsoft, Amazon and other cloud and large-scale companies are leading the trend of ASIC.

For example, Google's TPU, Meta's MTIA, Microsoft's Maia, Amazon's Trainium2, etc.

It should be noted that the cost of ASICs may be higher than that of general-purpose computing cards. According to Deutsche Bank's calculations, the TCO (total ownership cost) of GB200 is 44% lower than that of TPUv5 and 30% lower than that of Trainimium 2. Product structure, 10-30 billion yuan products operating income of 401/1288/60 million yuan respectively.

II. Broadcom and Marvell

As chip design and system complexity increase, large cloud computing and device OEM vendors will work more with ASIC design partners.

Broadcom and Marvell, as leaders in the ASIC market, may be the winners.

1) Broadcom's Cooperation and Development

Broadcom has always been the main manufacturer of Google's self-developed AI chip TPU, and this cooperation has lasted for about 10 years.

So far, the two sides have cooperated in the design of six generations of TPU and are promoting the mass production of the sixth generation TPU (3nm process).

Although the market has always rumored that Google will abandon cooperation with Broadcom and choose to self-develop to save costs.

However, recently, Broadcom revealed at an analyst conference that it has acquired the contract to provide Google with multiple generations of TPU. Credit Suisse believes that this contract includes the upcoming TPU seventh generation (v7), and it is expected to be released in 2026/2027.

In the past, the TPU fees Google paid to Broadcom were estimated to be $2 billion per year, which reached $3.5 billion in 2023 and is expected to reach $7 billion in 2024, mainly due to the rapid expansion of AI demand.

In addition, Broadcom's cooperation with Meta in its AI infrastructure is also expected to generate considerable revenue. Credit Suisse predicts that this cooperation may reach several billion dollars in the next two years.

Broadcom's customer base is not limited to Google and Meta, but also includes numerous customers in various industries such as Apple, Cisco, Fujitsu, Ericsson, Nokia, HPE, NEC, Juniper Networks, Ciena, Volkswagen, and Western Digital.

2) Marvell's Prospects

Marvell has years of ASIC cooperation experience with Amazon, Google and Microsoft.

Currently, Marvell is accelerating the production of its first two AI ASIC projects, reportedly Amazon's 5nm Tranium chip and Google's 5nm Axion ARM CPU chip.

In addition, there are several larger projects underway: 1) Amazon Inferentia ASIC, expected to launch in 2025; 2) Microsoft Maia, expected to launch in 2026.

Credit Suisse predicts that Marvell will experience strong growth in 2026.

And predicted Marvell ---

• AI revenue was between $1.6 billion and $1.8 billion in 2024, and will increase to between $2.8 billion and $3 billion in 2025;

• Can achieve accelerated computing/AI ASIC revenue of $7 billion to $8 billion by 2028.

In addition, Goldman Sachs mentioned that the surge in custom chips (ASICs) is a bullish message for companies that provide EDA software (tools required for chip design) and IP (pre-designed components that can be integrated into chips) such as SNPS, CDNS, and ARM.

3. Diversification of the supply chain

With cloud service providers and large OEMs entering the ASIC market, the supply chain may move towards diversification away from the dominance of Nvidia.

At present, almost 70% of AI computing in the market uses Nvidia's computing card, and the focus of the AI supply chain has always been on Nvidia's supply chain.

However, as cloud service providers gradually adopt ASIC chips, the supply chain may show a trend towards diversification.

In the ASIC chip supply chain, the choice of vendors mainly depends on their developers (cloud service providers, OEMs), rather than Nvidia.

For cloud service providers, they have enough strength to independently develop ASIC chips; however, for OEMs that are relatively lacking in R&D capabilities, Nvidia or IP licensing can enable them to independently develop based on Nvidia's computing card.

However, no matter which scenario, it will have an impact on diversifying the supply chain.

Looking at it from another perspective, Nvidia still has an advantage with some customers, such as sovereign states and small and medium-sized enterprises that do not have the advantage of self-development.