Author: Zhou Yuan / Wall Street Watch

AI-specific Chips (ASIC) and Industry Chain companies have become dark horses at the end of the year.

The related discussion topics stem from the value extension direction of AI-specific Chips in Datacenters: AEC, a special short-distance data transmission communication cable between server and server, and between server and workstation.

The related sector of this cable saw significant increases in May and August of this year. Subsequently, these sectors experienced a long period of adjustment. Now, with the performance of ASICs proving themselves, AEC has once again become the focus.

On December 16, several AEC Concept stocks hit their daily limit. Among them, Shenzhen Woer Heat-shrinkable Material continued to rise during Regular Trading Hours on December 17 (as of 11:01).

The extended value of dedicated AI Chips.

Broadcom in the USA announced astonishing performance for fiscal year 2024 on December 12: in fiscal year 2024, the revenue from Broadcom's AI ASIC (custom chips or dedicated chips) business increased by 220% year-on-year, reaching 12.2 billion USD.

At the same time, Broadcom claims that it is currently developing AI-related ASIC chips with three major clients, expecting the market size for its AI chips to reach 15 billion to -20 billion USD by 2025.

This achievement and the projection for next year's market size have shocked the global market.

As of the close of Regular Trading Hours on December 13, Broadcom recorded an astonishing increase of 24.43%, with its Market Cap breaking 1 trillion USD, joining the Global Semiconductor Market Cap club (alongside NVIDIA, Taiwan Semiconductor, and Broadcom); on December 16, Broadcom opened higher, with intra-day gains exceeding 6%, increasing its Market Cap to 1.12 trillion USD; on December 17, Broadcom closed at 250 USD, surging 11.21%, with its Market Cap touching 1.2 trillion USD.

From the release of the Earnings Reports on December 12 to the close on December 17, Broadcom’s stock price rose nearly 36.5%.

The emergence of GenAI (Generative AI) with ChatGPT has not only allowed NVIDIA CEO Jensen Huang to achieve a turnaround with GPUs and become a 'master Huang,' but has also spawned endless imaginative possibilities in the market for FPGAs and ASICs.

However, before Broadcom released its Earnings Reports this time, the market had not seen performance verification for ASICs or FPGAs, thus the entire AI acceleration card Industry Chain revolved around NVIDIA GPUs.

The difference between GPUs, FPGAs, and ASICs compared to CPUs mainly lies in their technical architecture. Traditional CPUs are slower than GPUs, FPGAs, and ASICs when processing vast amounts of data.

GPUs are standard (general-purpose) computing chips, representing companies besides NVIDIA, including AMD and Intel.

Compared to standard chip GPUs, ASICs are custom chips, also known as dedicated chips, mainly designed for specific user requirements and scenarios; FPGAs, or Field-Programmable Gate Arrays, are semi-custom chips that fall between GPUs and ASICs.

Broadcom's impressive performance comes from ASIC-based AI chips.

ASIC chips can also be divided into three categories: TPU, DPU, and NPU.

Given that the technical names of these three types of chips are quite complex, it's easier to summarize their respective uses: TPU is mainly used for machine learning; DPU provides computing engines for Datacenters (IDC).

NPU, this one is relatively familiar, right?

Huawei first launched the Da Vinci architecture in 2018, along with the AI accelerator card based on this architecture—Ascend 310; in 2019, Huawei introduced the first SoC chip integrating an NPU based on the Da Vinci architecture—Kirin 810.

The next generation, the 'Kirin 990' (the first SoC with an integrated 5G modem launched by a Chinese company), also achieved global prominence for Huawei smart phones due to its outstanding computing power performance.

So what 'work' does NPU mainly handle? It's evident that NPU is primarily used for processing data in deep learning instruction sets, which can be understood as mainly providing computing power.

ASIC chips have stronger performance and lower energy consumption compared to GPUs, but they have a high R&D threshold and long design cycles, making them more expensive than GPUs. However, large-scale mass production can solve the cost issue.

As a result, the new AI 'rising star' Broadcom is experiencing a surge, while the previous 'old-school' AI giants, NVIDIA, seem somewhat lonely: on December 16, during Regular Trading Hours, NVIDIA's stock price hit a two-month low since October 15.

Currently, the leading companies in the Global ASIC market mainly include Google, Intel, and Broadcom; companies like IBM, Samsung, Apple, and NVIDIA are also laying out plans for ASICs.

With ASIC dedicated chips, the data processing speed in Datacenters has correspondingly become faster.

However, this raises a problem: the data transfer speeds between the Servers and workstations that make up the Datacenter also need to be synchronously improved to fully leverage the speed advantages of ASIC dedicated chips.

In other words, the demand for short-distance high-speed data transfer (while the optical module mainly addresses long-distance high-speed data transfer) will grow rapidly based on ASIC.

At this point, it's time for AEC (Active Electrical Cable) to take the stage.

The domestic ASIC market is weak, but the AEC is still acceptable.

The reason it seems a bit lengthy and only now gets to the point is that without discussing the above background, it is difficult to understand why Shenzhen Woer Heat-shrinkable Material has become the topic of discussion.

AEC is a type of cable that differs from traditional common cables (wires), as AEC is an active power cable.

Inside conventional cables, there may only be a certain type of Metal (like Copper), but AEC cables integrate not only Copper but also some electronic Components that enhance signal transmission (such as signal amplifiers and equalizers).

What is the function of this cable?

The core function is singular: to support high-speed data transmission. What is the market scale of active AEC cables?

According to Light Counting's data forecast, during the period from 2023 to 2027, the global high-speed Copper cable market will continue to expand at a compound annual growth rate of 25%; it is expected that by the end of 2027, global shipments of high-speed Copper cables will reach the scale of 20 million units.

The basic signal transmission path for AEC is provided by Copper cables; in addition, there is a core module specifically designed for connecting active Copper cables, which is known as Copper cable interconnect.

Compared to some other high-speed transmission solutions, Copper cable interconnect is relatively low in cost. When constructing a Datacenter or other network systems, using AEC with Copper cable interconnect can satisfy performance requirements while reducing overall construction and Operating costs.

Previously, NVIDIA introduced at GTC 2024 that NVL72 significantly reduces costs by using Copper cables for interconnection instead of optical modules.

Domestic A-share companies with AEC related products mainly include: Shenzhen Woer Heat-shrinkable Material, Broadex Technologies, Zhaolong Interconnection and Anhui Xinke New Materials.

Shenzhen Woer Heat-shrinkable Material's 2024 semi-annual report shows the technical performance and application areas of the company's high-speed communication lines as shown in the figure:

On August 3, 2023, Shenzhen Woer Heat-shrinkable Material stated on InteractEase, "The 400G and 800G high-speed communication lines produced by our subsidiary Letin Intelligent Connection are DAC Copper cables used for data signal transmission in Datacenters, Servers, and Switches/Industrial Routers."

The AEC products of Shenzhen Woer Heat-shrinkable Material are mainly produced by its subsidiary Letin Intelligent Connection, including SAS series, PCIe series, and QSFP series; major clients include Amphenol Corporation from the USA, Volex from the United Kingdom, Molex from the USA, and TE Connectivity from Ireland, among other international cable connector leaders.

What is DAC?

It is a type of passive Copper cable (passive Copper cable) that requires appropriate connectors at both ends to connect to devices.

Therefore, the AEC cable (Active) Copper Connector is a key component for DAC to achieve device connections, and the two work together to complete data transmission between devices.

In addition to Shenzhen Woer Heat-shrinkable Material, Broadex Technologies also has high-speed cable products, categorized into two types: Active Optical Cables (AOC) and Active/Passive Copper Cables (DAC Passive, ACC, and AEC Active):

Broadex Technologies mainly focuses on optical modules; the company also offers ACC (Active Copper Cable), which is another type of active copper cable, similar to AEC, but the latter has relatively more comprehensive performance; for example, certain performance metrics of AEC are closer to those of Active Optical Cables (AOC).

Anhui Xinke New Materials is a newcomer in the high-speed copper connection field: on September 28 this year, the company initiated an "Investment Cooperation Agreement for High-Speed Copper Connection Project," just starting to lay out the high-speed copper connection project.

Zhaolong Interconnect has a similar breadth of product layout in high-speed cable business as Broadex Technologies: it includes high-speed cables (DAC and ACC), connectors, and connection modules.