Telecommunications | Increased Investment in ASICs Accelerates AI Network Development

Written byLi Heran, Zhou Hanlin, Wei Pengcheng

With the resonance of AI inference and training demands, along with the maturity of ASIC chips, the wave of AI network construction has been reignited. Major international chip leaders Marvell and Broadcom continue to increase their ASIC projects, suggesting that ASICs may drive the next wave of Capex investment and network construction alongside NVIDIA’s computing power cards. A virtuous cycle has formed from AI investment to AI revenue and back to AI investment, making the trends for upgrading interconnect components like optical modules and copper cables clearer, ushering in a period of high prosperity. It is recommended to pay attention to leading manufacturers in the optical module supply chain and copper cable/AEC (Active Electric Cable) sectors.

Telecommunications | Increased Investment in ASICs Accelerates AI Network Development

Overseas cloud vendors are increasing their investment in self-developed AI chips (ASICs).

As the demand for AI training and inference increases and diversifies, various cloud vendors are accelerating their ASIC chip projects. According to company websites and Bloomberg news, Google is set to launch its seventh-generation TPU (TPU v7) in 2024; Meta will release its second-generation self-developed chip MTIA in 2024, along with a large rack system integrating 72 chips; Amazon’s Trainium/Inferentia series chips are currently being shipped in bulk and will continue to iterate. We predict that self-developed ASIC chips from cloud and AI vendors will see large-scale shipments in 2025 and 2026, driving increased demand for optical modules and copper cables.

Marvell and Broadcom accelerate chip business, increasing network-side investment driven by ASICs.

Broadcom’s earnings call reflects the high prosperity of AI networks and ASICs: According to Broadcom’s public earnings call on June 8, the company has seen strong demand in both ASICs and AI networks, with AI network revenue growing approximately 170% year-over-year in the past quarter, indicating a resonance between inference and training demands.

Marvell is optimistic about AI Capex outlook and the rapid growth of the AI network market: At the Marvell Custom AI Investor Event on June 18, the company provided a positive outlook on the prosperity of AI Capex investment, ASICs, and supporting equipment among North American vendors. Regarding AI Capex, the company mentioned that with new vendors or projects increasing their investments, along with the already strong Capex from Google, Microsoft, Meta, Amazon, and Oracle, overall AI capital expenditure in North America is expected to rise from $435 billion in 2024 to $593 billion in 2025, with a projected CAGR of 20% from 2025 to 2028, reaching $1 trillion by 2028. In terms of cluster scale, the company believes that future AI clusters may expand to a million cards.

ASICs and NVIDIA chips drive growth, with continued optimism for future optical module demand.

Driven by the explosion of AI applications and the rapid iteration of foundational models, the demand for AI computing infrastructure remains robust. We expect that shipments of optical modules in 2025/2026 will be driven by the dual forces of NVIDIA’s accelerated cards and the volume production of ASIC chips. We anticipate that the increase in Ethernet penetration and the large-scale deployment of ASICs will continue to drive growth in 800G optical module shipments. Currently, leading optical module manufacturers have a solid industry position, and we are optimistic about the sustainability of their performance growth. Additionally, the strong demand for optical modules is also driving growth in upstream optical devices, optical chips, and optical engines, and we are optimistic about the performance of related manufacturers.

ASICs and high-density AI cluster construction drive AEC demand.

According to the Q1 2025 earnings call of overseas AEC copper cable leader Credo, the demand for customized XPUs (X Process Units) is driving the need for copper connections due to the increase in large-scale AI clusters. The company’s management noted a significant increase in customer diversification compared to the previous quarter, with three customers now accounting for over 10% of revenue. Additionally, the company is set to add two more large cloud vendor customers that will also account for over 10% of revenue. We believe that as Ethernet penetration increases and the density of AI cluster scale-up improves, the prosperity of AEC will continue.

Risk Factors:

AI inference may fall short of expectations; demand for new products such as high-speed optical modules and copper cables may not meet expectations; the development progress of new solutions/technologies like CPO may not meet expectations; market competition may intensify; technological path risks; geopolitical risks.

Investment Strategy.

As the demand for AI inference and training begins to resonate and ASIC chips mature, we believe that the future demand for computing power and network devices will remain strong, and the trends for the volume and upgrade of interconnect components like optical modules and copper cables will become clearer. It is recommended to focus on leading manufacturers in the fields of high-speed optical modules, optical devices, optical engines, optical chips, and AEC/copper cables.

Telecommunications | Increased Investment in ASICs Accelerates AI Network Development

This article is excerpted from the report “Significant Comments on the Telecommunications Industry – Increased Investment in ASICs Accelerates AI Network Development” published by CITIC Securities Research Department on June 19, 2025. For detailed analysis (including related risk warnings, etc.), please refer to the report. In case of any ambiguity arising from the excerpt, the complete content of the report published on the day should prevail.

The securities market research information contained in this material is prepared by the Research Department of CITIC Securities Co., Ltd. (hereinafter referred to as “CITIC Securities”). The CITIC Securities Research Department is positioned as a sell-side research team aimed at professional institutional investors. This material, produced in the form of WeChat, is intended only for professional investors among CITIC Securities’ clients; please do not forward this material in any form. CITIC Securities does not consider any subscriber of this public account as a client of CITIC Securities due to the act of subscribing to this public account. If you are not a professional investor among CITIC Securities’ clients, to ensure service quality and control investment risks, you should first contact CITIC Securities’ institutional sales service department or the client manager of the brokerage business system to complete investor suitability matching and fully understand the nature, characteristics, precautions, and potential risks or losses of the service before subscribing, receiving, or using the information in this public account. This material is difficult to set access permissions, and we apologize for any inconvenience caused! Thank you for your understanding and cooperation. If you have any questions, please send an email to [email protected].

Important Statement:

This material is positioned as “investment information reference service” rather than specific “investment decision service” and does not involve judgments on specific securities or financial instruments at specific prices, specific times, or specific market performances. It is particularly important to note that (1) recipients of this material should carefully read the attached statements, information disclosure matters, and risk warnings, pay attention to the key assumptions under which the relevant analysis and forecasts can hold, and accurately understand the meaning of investment ratings and the forecast time frame for target prices. (2) The information contained in this material is considered reliable; however, CITIC Securities does not guarantee its accuracy or completeness, and the related analysis opinions and speculations may be changed without notice based on subsequent securities research reports published by CITIC Securities Research Department, and may differ or contradict the opinions given by other departments, units, or affiliates of CITIC Securities in producing similar other materials due to the use of different assumptions and standards, or different viewpoints and analytical methods. (3) Before making specific investment decisions, investors should consider their own risk preferences, capital characteristics, and other specific situations, and combine various other auxiliary analysis methods, including “stock selection” and “timing” analysis, to form independent decisions. To avoid misunderstandings regarding the research methods, investment ratings, target prices, etc., involved in this material, which may lead to investment losses, professional investment advice should be sought when necessary. (4) The risk factors listed above do not encompass all risks associated with the improper use of this material. Investors should not solely rely on the information received but should make independent judgments, fully understand various investment risks, and make their own investment decisions while assuming investment risks.

All materials in this public account are copyrighted by CITIC Securities. No content in this public account may be modified, sent, or copied in any way. Except in cases recognized by CITIC Securities, all other forms of reproduction are illegal. All rights reserved, violators will be prosecuted.

Leave a Comment