In today’s era of rapid advancement in large AI models, as we focus on GPUs, optical modules, and switch chips, a fundamental shift at the underlying level is often overlooked. In the latest AI server architectures, the cost share of PCBs (printed circuit boards) is rapidly approaching that of core chips. In high-end InfiniBand network switches, PCB costs now account for 18% of the total bill of materials, while ASIC‑based switch chips make up 23%. Industry consensus has emerged: PCBs are no longer mere “connectors”; they are evolving into “semiconductors.”

This is by no means an exaggeration; AI computing power has brought the “physical limits” of PCBs into sharp focus. As signal rates surpass 224 Gbps, with 800G scaling up and 1.6T entering mass production, every trace on a PCB has become a high‑frequency transmission line. Conventional FR‑4 materials and millimeter‑level manufacturing tolerances can no longer meet the stringent signal‑integrity requirements of AI workloads. In response to this fundamental shift, Shenzhen LeaKin Technology has already taken the lead in achieving a comprehensive upgrade of both its technology and capabilities.

A qualitative leap from “circuit boards” to “transmission line systems”

In the 224G SerDes era, signals are no longer simply “traveling from point A to point B”; instead, they propagate across the PCB as electromagnetic waves. The traditional design concerns—such as whether trace widths are sufficient or whether vias can be properly aligned—are no longer adequate. We must now confront semiconductor‑level physical considerations: at 224G, the surface roughness of copper foil in standard FR‑4 materials can no longer be ignored; for every additional 1 μm of roughness, insertion loss may increase by more than 10%. Variations in the dielectric constant of glass fibers can lead to severe impedance fluctuations, and a poorly handled via stub can consume up to 10% of the signal margin.

To address these challenges, LeaKin Technology has fully adopted M8/M9‑grade ultra‑low‑loss CEM laminates and quartz‑fabric materials, keeping line‑width tolerances within ±10% and achieving differential pair length‑matching accuracy of ±1 mil. With manufacturing precision approaching that of IC package substrates, we ensure high‑speed signals can traverse the PCB with virtually no loss.

Reimagining R&D and Delivery with “Simulation-Driven” Approaches

AI server PCBs often exceed 32 layers, with single‑run prototyping costs reaching hundreds of thousands and lead times surpassing one month. LeaKin Technology has completely abandoned the traditional “prototype‑and‑trial‑and‑error” approach, adopting a fully simulation‑driven verification process instead. Early in the project, our signal integrity (SI) engineers step in to build transmission‑line models using 3D electromagnetic field simulations, enabling precise assessments of impedance, loss, and crosstalk. By leveraging tens of millions of virtual computations, we trade for a single successful real‑world outcome, significantly shortening the product’s time‑to‑market.

Crossing Boundaries: Becoming an Expert in the Underlying Architecture of AI Hardware

The semiconductorization of PCBs has blurred the boundaries between design and manufacturing. At LeaKin Technology, our hardware team not only masters circuit‑board layout but also understands electromagnetic fields; we go beyond routing to grasp material properties. We can engage in seamless collaboration with leading upstream material and substrate manufacturers, optimizing every minute detail that impacts high‑frequency performance—from laminate stack‑up and copper‑foil selection to solder‑mask processes.

In the arms race of AI computing power, the “semiconductorization” of PCBs is an irreversible trend. Choosing LeaKin Technology means choosing a hardware partner that understands high-frequency design, material science, and the fundamental principles underlying AI computing. Whether you’re an in-house engineer or an external collaborator, we apply semiconductor‑grade, rigorously stringent standards to build for you a high‑speed information highway to the future. Let’s shift gears together, embrace this hardcore evolution tailored for hardware engineers, and thrive in the AI era!