The STAR Market Daily reported on March 21 that the much-anticipated GTC conference speech was held as scheduled. During the event, Jensen Huang reiterated the necessity of future computing power construction needs, the demand for inference-side tokens, and provided$NVIDIA (NVDA.US)$a showcase of the product series. All indications suggest that the future demand for PCB is expected to expand further.

NVIDIA Groq 3 LPU is undoubtedly the new highlight of this conference. As a brand-new chip making its debut, it can be regarded as an ASIC specifically designed for AI inference, with its core focus on achieving extremely low latency and high throughput. According to the release, a single LPU server consists of 32 trays, each tray integrating eight LPU chips, with a single cabinet containing 256 LPU chips.

Compared to previous cabinet architectures, the number of trays in the LPU single cabinet has significantly increased (for instance, the GB300 NVL72 single cabinet consists of 27 trays). In fact, in server hardware architecture, the number of trays is closely related to the quantity and design of PCBs. Since integrating more electronic components often requires more space and circuit layout, typically, the more complex the electronic equipment, the greater the amount of PCB required.

According to Dongwu Securities, the orthogonal backplane solution in the Rubin Ultra architecture was showcased at this year’s GTC conference, which is also expected to increase the demand for PCBs. It is reported that in the Kyber architecture, by connecting vertically placed computing blades and switching blades via an orthogonal backplane, copper cables are replaced in the Scale-up layer, achieving higher computing power integration per cabinet.

In addition, this year’s GTC conference also showcased Vera CPU standalone cabinets and STX storage cabinets, which will similarly drive incremental growth in PCB segments.

The growth trend of PCB is gradually forming.

Against the backdrop of increasingly clear shipment expectations for NVIDIA’s Rubin architecture, numerous PCB manufacturers are also expected to enter a new growth cycle.

On March 17,$Olympic Circuit Technology (603920.SH)$stated on an interactive platform that after the company entered NVIDIA's server supply chain system through the OEM model, it relied on its core process advantages in high-multilayer boards and high-frequency, high-speed PCB, as well as stable mass production and delivery capabilities. As a result, the supply volume of NVIDIA server-related PCB products has achieved rapid growth at this stage.

At the same time, the expansion trend of the PCB industry is also gradually solidifying. Recently,$Avary Holding (002938.SZ)$announced that its wholly-owned subsidiary plans to invest 11 billion yuan in building a production base for high-end PCB projects. This investment aims to seize the wave of AI technology development, accelerate the production layout of high-end PCB products, which will help expand the company's business scale, promote technological upgrades and product iterations across various product lines, thereby enhancing the company's operational efficiency.

Guojin Securities' research report on March 15 indicated that robust AI demand has driven both price and volume growth for PCBs. Currently, many AI PCB companies are experiencing strong order flows, operating at full capacity, and actively expanding production, with high growth expected to continue. Guotou Securities noted that since 2026, domestic and international PCB manufacturers have been continuously increasing capital expenditures to expand production capacity, focusing primarily on advanced AI PCB projects such as high-layer count, high-level HDI, and high-frequency high-speed boards. Some manufacturers' investment scales have increased several-fold compared to 2025.

From a technology development perspective, some PCB manufacturers have begun advancing the verification of next-generation technologies:

$Victory Giant Technology (300476.SZ)$Recent research meeting minutes show that the company has completed electrical and thermal performance verification of M7 and M8 grade materials in its products and is actively promoting the certification of M9/M10 grade materials to improve the stability of high-frequency signal transmission and meet the requirements of the next-generation AI chip architecture. The company is deeply involved in the cooperative R&D of core customers' orthogonal backplane projects, entering the proprietary technology accumulation phase in advance, and closely aligning with customer mass production schedules.

According to a supply chain survey by analyst Ming-Chi Kuo, NVIDIA and$Wus Printed Circuit (002463.SZ)$have begun testing the next-generation CCL material M10. It is reported that Wus Printed Circuit has started sample submission and prototyping this quarter, with preliminary test results expected in the second quarter. If the tests proceed smoothly, mass production of M10 CCL and PCB is anticipated to commence in the second half of 2027.

CITIC Securities believes that in terms of application expansion, the orthogonal backplane solution is expected to see large-scale shipments starting in 2027 alongside Rubin Ultra 576 cabinets, significantly increasing GPU PCB ASP by approximately $400. Regarding specification upgrades, while the orthogonal backplane still uses the M9+Q cloth solution, other products like mid-plates and Switchtrays are seeing customers consider parallel interim alternative solutions due to tight capacity of advanced materials. This does not conflict with the ongoing trend of PCB specification upgrades under high-speed interconnect advancements. As capacity bottlenecks are resolved, the adoption of high-end processes remains highly predictable.

The institution further emphasized that as an industry with significant capital-intensive attributes, PCB production ramp-ups and incremental earnings releases exhibit stepwise growth characteristics. According to announcements from leading firms, new capacities from major PCB manufacturers will be released gradually starting in the second half of this year. By Q4 2025, capacity utilization is already near full, leaving limited room for product mix optimization, and a temporary slowdown in sequential growth rates is a reasonable industry characteristic. The industry cycle remains in an upward phase, and with additional capacity coming online in 2026, leading PCB manufacturers are expected to deliver strong earnings growth.

From an investment perspective, Guotou Securities pointed out that considering NVIDIA’s order scale and the upstream-downstream capacity expansions within the PCB industry, AI-driven demand remains robust. Additionally, the launch of new products such as LPU chips continues to bring incremental opportunities to the PCB sector. PCB equipment and consumables suppliers are expected to benefit significantly. Investors are advised to focus on relevant targets in the consumables and equipment segments.

Editor/Rocky