Nvidia has received incredible orders in the AI application frenzy, accelerating the transformation.
(Taipei News) Recently, there has been a significant and rapid increase in global demand for GPUs, fueled by the growing popularity of AI applications like ChatGPT. As the leading provider of AI technology worldwide, Nvidia acknowledged in a recent press conference that they are experiencing an overwhelming volume of orders and emphasized that this marks the beginning of a transformative phase for data centers.
Despite the challenges Nvidia faced last year, such as a decline in the PC market and the diminishing interest in cryptocurrency mining, resulting in substantial order revisions for their consumer product, the RTX 40 series, there has been a notable surge in demand for AI servers in data centers in recent times.
Prominent international clients like Microsoft, Google, and Oracle have remained steadfast in their purchases of Nvidia's DGX A100 servers and are even planning to introduce the new Nvidia DGX H100 model in the second half of the year.
The projected average selling price of the new model is estimated to be two to two-and-a-half times higher than its predecessor. This significant price increase is expected to have a positive impact on the overall revenue scale in the AI server market, leading to growth and expansion.
Nvidia emphasizes that we are currently at the beginning of a decade-long transformation period for data centers, during which the application of AI is poised to flourish.
As AI applications like ChatGPT gain wider acceptance, the demand for high-performance GPUs will continue to grow. Nvidia's leading position in AI technology and its robust product line enable the company to benefit from this trend.
Nvidia's success in the AI server domain has not only propelled the company's financial growth but has also presented opportunities for Taiwan's semiconductor industry.
With the flourishing development of AI applications, the global demand for high-performance AI chips and related technologies is expected to continue growing. Taiwan's semiconductor industry, including ASIC design, advanced wafer manufacturing, heterogeneous integration and packaging of small chips, ABF chip substrates, and advanced packaging equipment supply chain, has already established relevant capabilities and is committed to providing end-to-end supply chain services.
In the semiconductor manufacturing industry, TSMC, the leading semiconductor foundry, holds advanced process technology and manufactures many AI chips. Major American companies like Nvidia and AMD actively deploy AI chips and also utilize TSMC's advanced manufacturing processes.
Particularly, there has been a sudden increase in AI orders, leading to a surge in demand for advanced packaging, necessitating a rapid capacity expansion. TSMC's President, Wei Che-Chia, anticipates that with the launch of new customer products, the company's performance in the second half of the year will surpass that of the first half. With the AI trend driving the market, advanced packaging is expected to become a new growth engine for TSMC.
In terms of packaging technology, the demand for AI chips is constantly increasing in addition to wafer manufacturing. The rapid increase in AI demand has led to a rise in the demand for CoWoS (Chip-on-Wafer-on-Substrate). CoWoS is an advanced packaging technology that consists of two parts: Chip-on-Wafer (CoW) and Wafer-on-Substrate (WoS). CoW involves stacking multiple chips together, and WoS refers to packaging the stacked chips on a substrate. Together, they form the CoWoS technology. In simple terms, it involves stacking chips and then packaging the stacked chips on a substrate, which can help save power consumption and reduce costs. The CoW process involves interposer handling and bonding the chips to the interposer, while the WoS process involves cutting and packaging the chips on a substrate. It is expected that TSMC will capture the majority of the orders for both CoW and WoS processes, while companies like ASE may primarily focus on the WoS process.
ASE (Advanced Semiconductor Engineering) has been deeply involved in AI chip packaging and testing. Its subsidiary offers a range of packaging technologies, including FCMCM (Flip Chip Molded Chip on Module), 2.5D & 3D IC, and FOCOS (Fan-Out Chip on Substrate). These technologies have been applied in high-performance central processors, FPGA chips, network chips, and more. With the increasing popularity of AI applications, there is also increasing attention on small chip packaging technology.
In PCB technology, ABF substrate is considered a crucial material for enhancing the packaging efficiency of AI chips and improving their operational performance. It can effectively improve the functionality of AI chips. Market expectations indicate that ABF substrate suppliers such as Xinxiang, Nanya, and JingShuo can ride the wave of AI chip applications.
In terms of critical equipment, with the increasing demand for high-end AI chips, market expectations suggest that backend advanced packaging equipment suppliers such as Hong Shu, Xin Yun, Wan Run, and Tai Sheng, are expected to benefit. Packaging equipment manufacturers have indicated that the second half of this year will be better than the first half, but next year is expected to be even better. This is primarily due to a shortage of certain critical components, resulting in equipment lead times of over six months. The delivery of expansion equipment may be delayed until the end of this year or next year.
In the field of heterogeneous integration, the adoption of heterogeneous integration in packaging has been accelerating in order to meet the growing demand for more complex functionality and lower power consumption. Heterogeneous integration allows IC manufacturers to stack and integrate more silicon devices within a single package. This may involve combining chips from different wafers, different semiconductor technologies, and different suppliers. Heterogeneous integration also enables the integration of smaller chips to overcome yield challenges associated with larger bare dies and mask size limitations.
The advantage of heterogeneous integration in advanced packaging lies in the ability to encapsulate different functional chips, such as memory, logic, and sensors, into a single chip. Customers can mix and match processes, utilizing 3/5-nanometer process technology only for critical functions while using mature processes for the rest. This approach not only enhances chip performance but also reduces costs. Because the significant increase in demand for advanced packaging, Laser Debonding developed by Kingyoup Optronics Co.,Ltd, a key technology in heterogeneous integration advanced packaging, has been attracting more and more attention.
Laser Debonding technology offers fast and efficient separation of wafers and packaging substrates, offering benefits such as low maintenance costs and high wafer yield. With the technical support of Kingyoup Optronics Co.,Ltd., customers can accelerate the production of high-quality AI chips and meet the demands of the market more quickly. In addition to AI chips, Laser Debonding technology can be applied to other advanced packaging processes such as 3D IC/2.5D IC.
Kingyoup Optronics’s patented high-efficiency laser debonding system adopts a 355nm solid-state laser and incorporates cold cutting, non-thermal effects, and high-frequency rapid scanning techniques. It is suitable for 12" wafers and square fan-out wafer-level packaging.
Image：Yole Group /Kingyoup Optronics Co.,Ltd. is already one of the world's top three suppliers of Laser Debonding equipment solutions.
Overall, the recent wave of AI applications has driven a rapid growth in global GPU demand, and Nvidia, as the world's largest AI technology provider, has received incredible orders amidst this trend.
Meanwhile, Taiwan's semiconductor industry is actively positioning itself in the AI chip domain, creating more opportunities and challenges for industry development. With the continuous advancement of AI technology, this trend will provide greater momentum for the global development of smart technology.
Reference: