ChatGPT’s debut has sparked a thrilling spec upgrade in the server market, which has breathed new life into the supply chain and unlocked unparalleled business opportunities. Amidst all this, the big winners look set to be the suppliers of ABF (Ajinomoto Build-up Film) substrates, who are poised to reap enormous benefits.

In the previous article, “AI Sparks a Revolution Up In the Cloud,” we explored how the surge in data volumes is driving the spec of AI servers as well as the cost issue that comes with it. This time around, we’ll take a closer look at the crucial GPU and CPU platforms, focusing on how they can transform the ABF substrate market.

NVIDIA’s Dual-Track AI Server Chip Strategy Fuels ABF Consumption

In response to the vast data demands of fast-evolving AI servers, NVIDIA is leading the pack in defining the industry-standard specs.

This contrasts with standard GPU servers, where one CPU backs 2 to 6 GPUs. Instead, NVIDIA’s AI servers, geared towards DL(Deep Learning) and ML(Machine Learning), typically support 2 CPUs and 4 to 8 GPUs, thus doubling the ABF substrate usage compared to conventional GPU servers.

NVIDIA has devised a dual-track chip strategy, tailoring their offerings for international and Chinese markets. The primary chip for ChatGPT is NVIDIA’s A100. However, for China, in line with U.S. export regulations, they’ve introduced the A800 chip, reducing interconnect speeds from 600GBps (as on the A100) to 400GBps.

Their latest H100 GPU chip, manufactured at TSMC’s 4nm process, boasts an AI training performance 9 times greater than its A100 predecessor and inferencing power that’s 30 times higher. To match the new H100, H800 was also released with an interconnect speed capped at 300GBps. Notably, Baidu’s pioneering AI model, Wenxin, employs the A800 chip.

To stay competitive globally in AI, Chinese manufacturers are expected to aim for the computational prowess on par with the H100 and A100 by integrating more A800 and H800 chips. This move will boost the overall ABF substrate consumption.

With the ChatBot boom, it is predicted a 38.4% YoY increase in 2023’s AI server shipments and a robust CAGR of 22% from 2022 to 2026 – significantly outpacing the typical single-digit server growth, according to TrendForce’s prediction.

AMD, Intel Server Platforms Drive ABF Substrate Demand

Meanwhile, examining AMD and Intel’s high-end server platforms, we can observe how spec upgrades are propelling ABF substrate consumption forward.

• AMD Zen 4:

Since 2019, AMD’s EPYC Zen 2 server processors have used Chiplet multi-chip packaging, which due to its higher conductivity and cooling demands, has consistently bolstered ABF substrate demand.

• Intel Eagle Stream:

Intel’s advanced Eagle Stream Sapphire Rapids platform boasts 40-50% higher computation speed than its predecessor, the Whitley, and supports PCIe5, which triggers a 20% uptick in substrate layers. This platform employs Intel’s 2.5D EMIB tech and Silicon Bridge, integrating various chips to minimize signal transmission time.

The Sapphire Rapids lineup includes SPR XCC and the more advanced SPR HBM, with the latter’s ABF substrate area being 30% larger than the previous generation’s. The incorporation of EMIB’s Silicon Bridge within the ABF substrate increases lamination complexity and reduces overall yield. Simply put, for every 1% increase in Eagle Stream’s server market penetration, ABF substrate demand is projected to rise by 2%.

As the upgrades for server-grade ABF substrates continue to advance, production complexity, layer count, and area all increase correspondingly. This implies that the average yield rate might decrease from 60-70% to 40-50%. Therefore, the actual ABF substrate capacity required for future server CPU platforms will likely be more than double that of previous generations.

ABF Substrate Suppliers Riding the Tide

By our estimates, the global ABF substrate market size is set to grow from $9.3 billion in 2023 to $15 billion in 2026 – a CAGR of 17%, underscoring the tremendous growth and ongoing investment potential in the ABF supply chain.

Currently, Taiwanese and Japanese manufacturers cover about 80% of the global ABF substrate capacity. Major players like Japan’s Ibiden, Shinko and AT&S, along with Taiwan’s Unimicron, Nan Ya, and Kinsus all consider expanding their ABF substrate production capabilities as a long-term strategy.

As we analyzed in another piece, “Chiplet Design: A Real Game-Changer for Substrates,” despite the recent economic headwinds, capacity expansion of ABF substrate can still be seen as a solid trend, which is secured by the robust growth of high-end servers. Hence, the ability to precisely forecast capacity needs and simultaneously improve production yields will be the key to competitiveness for all substrate suppliers.

Read more:

• HBM Supply Leader SK Hynix’s Market Share to Exceed 50% in 2023 Due to Demand for AI Servers, Says TrendForce
• Shipments of AI Servers Will Climb at CAGR of 10.8% from 2022 to 2026, Says TrendForce

(Photo Credit: Google)

Shipments of CPUs, GPUs, and chipsets have been falling due to the weakening demand for PCs, gaming devices, and cryptocurrency mining machines. This recent development has also constrained the growth of the market for ABF substrates. Currently, the demand situation for this material is exhibiting signs of uncertainty.

Regarding the distribution of the demand for ABF substrates, applications that are driving growth are cloud services, AI, and automotive electronics. CPUs, GPUs, FPGAs, and switch ICs are chips that are deployed in servers purposed for a wide range of applications related to cloud services and endpoint AI technologies. Meanwhile, other AI-related applications require high-end ASICs. At the same time, more and more high-end SoCs and MCUs are embedded in vehicles. All in all, these aforementioned applications will spur the demand for ABF substrates. Additionally, package size continues to increase for high-performance ICs. This trend, too, will sustain the demand for ABF substrates over the long haul. By contrast, the PC market has matured, so the related demand is shrinking. From a long-term perspective, the influence of the PC market on the demand for ABF substrates will gradually wane.

TrendForce forecasts that the scale of the global market for ABF substrates will expand from US$9.3 billion in 2022 to US$17.1 billion in 2026, thus showing a CAGR of 16.4%. Due to the influence of the US technology export restrictions against China, the demand for ICs purposed for HPC will be higher than expected for the period from 4Q22 to 3Q23. This, in turn, will also further raise the demand for ABF substrates.

Then, starting from 4Q23, exports of HPC chips to China will start to slow down. However, demand will continue to grow for ASICs, AI chips, SoCs, and MCUs at that time. The growth in these application segments will offset some of the negative effect of the US export restrictions on the market for ABF substrates. In terms of the supply-demand dynamics of ABF substrates, a balance will gradually be attained in 2024. However, demand will get stronger in 2025 and 2026, so supply could tighten during that two-year period.

Current U.S. sanctions on China have extended their reach to strike at HPC and sectors such as aerospace, automotive market, and military industry. TrendForce indicates, the market for high-end computing chips (including CPU, GPU, etc.) has borne the brunt of these restrictions at this stage, while those providing related storage such as DRAM and NAND Flash also face potential supply disruption. At present, this not only includes domestic companies in mainland China but also extends to related US-based suppliers. Among them, server companies that rely on high-intensity computing will face greater scrutiny.

Impact analysis on server terminal shipments

In terms of server terminal shipments, since relevant component suppliers have not yet been able to confirm whether services provided by the four major cloud service providers (CSPs) in China, Baidu, ByteDance, Alibaba, and Tencent, involve military use, before CSPs sign MOUs (memoranda of understanding), component manufacturers may temporarily delay shipments to the Chinese market. However, TrendForce believes, due to the fact that current CSP buyers’ component inventories remain sufficient, the short-term impact on global server market shipment performance is relatively low and long-term impact depends on the evolution of the US Department of Commerce’s rules.

Huawei and Sugon, two companies that have received attention at this stage due to the US ban, have previously withdrawn from the x86 server market and turned into cloud business providers and whole server delivery has been transferred to other domestic OEMs and outsourced computing power leasing, so as not to be affected by sanctions. However, due to the previous CPU ban, Sugon has turned to AMD to obtain authorization for localized chips, which may be significantly curtailed by this ban. In 2022, Sugon’s market share in the overall server market will be approximately 2.3% and 8.5% of the Chinese market.

TrendForce believes, it cannot be ruled out that relevant Chinese OEMs may have server products that may be rendered to government supercomputing centers in the future. Inspur, H3C, and Lenovo will face more exacting future scrutiny and, if consequences intensify, the mainland Chinese industrial chain may feel direct effects. Although commercial servers are not currently on the list of directly restricted items, if friction between the United States and China intensifies in the future, it cannot be ruled out that the U.S. Department of Commerce will add more potentially risky Chinese server OEMs and CSPs onto the UVL list. If certification cannot be realized within 60 days of being included in the UVL list, these entities will be included on the entity list. The worst case scenario will be a future trend of negative growth in Chinese server demand.

Since the restrictions enumerated in this ban are primarily concentrated in the HPC field, the greatest factor affecting Sugon is the company largely providing server OEM to government departments including in supercomputers, military aerospace, and government server farms. At present, there are 8 national-level supercomputing centers in mainland China and the supercomputer located in the center of Wuxi is the headquarters of China’s self-developed chips including the self-developed Sunway TaihuLight. As the U.S. Department of Commerce continues to strengthen its sanctions, China’s supercomputing technology and domestic research capabilities will be severely damaged in the future.

Impact analysis on GPU and CPU sectors

At present, companies utilizing high-end graphics cards are primarily concentrated in the HPC sector. In terms of CSPs, Alibaba and Baidu are the largest companies in mainland China. These two CSP companies account for up to 60% of the market share of GPU usage in China. Before the previous ban at the end of August, Chinese CSP operators had to submit purchase applications before procurement but they could not apply at all after the ban. However, based on the premise that buyer inventory levels on hand remain high and the supply of goods through distribution channels is sufficient, no effect on demand is forecast until 1H23. Nonetheless, it will be a challenge in the long-term. Since the ban expressly prohibits supercomputing center applications such as HPC, TrendForce assesses that GPU servers used by supercomputing centers will be directly affected, which accounts for up to 30% of China’s GPU market.

In terms of chip computing performance control, ECCNs 3A090 and 4A090 are newly added sanctioned items and chips with a total processing performance of more than 4,800 (inclusive) calculated by TOPS will be restricted. GPUs are usually used to directly assist in performing complex operations. Basically, NVIDIA’s A100 PCIe Gen4 and AMD’s MI250 OAM Module exceed the 4,800 limit. With new high computing performance products restricted in the future, development of server acceleration computing in China will take a hit.

However, the computing performance of most server CPU products is generally lower than the provisions of the ban. Only Chinese-made chips such as Tianjin Haiguang face direct restrictions and other CPUs such as Intel and AMD servers will not be subject to prohibition. At this stage, Intel and AMD will sign MOUs with relevant mainland Chinese manufacturers to ensure that related products cannot be used in military and supercomputing fields before shipment. In today’s server CPUs, the computing performance of the commonly used Intel Ice Lake CPU series does not reach the limit imposed by U.S. sanctions.

Impact analysis on the memory sector

At present, Samsung and SK hynix have also suspended their supply of product to Sugon. If Sugon can clarify procured memory is not used for supercomputing, domestic server products, etc., the parties will be able to reach a consensus for shipment. In the long run, Korean companies are evaluating whether they need a written commitment from each customer to disavow using purchasing memory products in supercomputers. Therefore, some memory shipments may be affected before documents are signed. The industry generally believes that market inventory remains relatively abundant and there will be no substantial damage to the market in the short term. As far as SSD is concerned, the greatest utilization remains in the category of AI/DL (Deep Learning), since most of the data trained from DL must be stored in faster and more convenient SSDs for use in inference scenarios. If the suspension of shipments caused by the current ban cannot be rectified by relevant buyer agreements, the development of Chinese server manufacturers in related AI/DL fields may be hamstrung and a calamitous decline in the market penetration rate of enterprise SSDs from international manufacturers cannot be ruled out.

Impact analysis on the networking sector

There are three reasons for a relatively minor impact assessment on the well-connected suppliers in the networking sector. First, there are numerous networking suppliers and many of them are in China. Since the demand for key components is relatively small, Chinese suppliers should be able to keep up. Second, the mainstream process in this field is a mature process and future expansion is less restricted. Third, from the perspective of supplier shipments, after foundry assembly, packaging, and testing, there are multiple distribution channels for the circulation of the final product and it will be difficult to determine whether terminals are military use. However, from the perspective of long-term impact, there is a high probability that Chinese manufacturers will give priority to China’s local supply chain in the future to ensure future supply. This move will undoubtedly deepen the resistance of other suppliers’ shipments to China, so it is necessary to open up multiple shipping channels to stabilize market share.

（Image credit: iStock）