This is precisely where the PCB (Printed Circuit Board) is changing status. Long perceived as a discreet, almost mundane component of electronics, it is becoming a performance asset in modern AI. As Nvidia increases compute density, internal bandwidth, and the number of interconnections in Rubin, the quality of the substrate carrying the signal becomes critical. In the Vera Rubin POD architecture, Nvidia is promoting a much more advanced rack-scale logic, with several specialized systems integrated into a single unit and a third generation of MGX architecture designed for the movement of massive data volumes at low latency. At this level, the PCB material is no longer a mere industrial detail: it becomes a prerequisite for the machine's proper operation.

Rubin is significant because it pushes the industry toward more distributed, denser, and more demanding computing. Nvidia speaks of agentic AI—systems that reason in multiple steps, handle long contexts, call tools, execute code, and multiply exchanges between GPU, CPU, memory, and storage. To absorb this load, Rubin does not settle for a faster GPU: the platform assembles several types of specialized racks within a single supercomputer, with an architecture designed for resilience, energy efficiency, and speed of deployment. As this architecture moves upmarket, signal transmission constraints become increasingly severe.

This is where ELL, or Extreme Low Loss, comes in. In simple terms, these are ultra-high-quality materials designed to allow very fast signals to pass with minimal loss. When a signal travels across an electronic board, part of its energy naturally dissipates as heat. As frequencies and bitrates increase, this phenomenon becomes more penalizing. ELL materials reduce this transmission loss, improve signal stability, limit heating, and help reduce the system's electrical consumption. This is exactly what modern AI infrastructures require, where performance no longer depends solely on chips, but also on the system's ability to cleanly circulate gargantuan volumes of data.

The industry now follows an increasingly clear hierarchy between standard, Low Loss, Very Low Loss, Ultra Low Loss, and Extreme Low Loss materials. This move up the value chain is not a theoretical refinement: it addresses a physical constraint. The denser the architecture and the faster the internal links, the more essential materials capable of preserving signal integrity become. ELL is therefore not a luxury for fastidious engineers. It is the entry ticket for the most advanced AI infrastructures.

The thesis becomes even more compelling with the next stage of Rubin. The market already anticipates that Rubin Ultra, and subsequent platforms, will require even higher-performance materials for key components such as backplanes and network cards. This is logical: the more Nvidia pushes its architecture toward the entire rack, the more the quality of internal interconnections becomes a competitive advantage. In the stock market, this type of mutation rarely benefits only the lead contractor. It also benefits—sometimes primarily—the suppliers who master the materials indispensable to this performance surge.

And in this technology, two or three Taiwanese companies stand out:

  • Taiwan Union Technology, first, because it is already very advanced in high-end materials.
  • ITEQ, next, because its exposure to infrastructure is massive and its product portfolio covers precisely the segments where value is concentrated today: high-speed, high-frequency, and low-loss materials. If Rubin accelerates the migration toward denser and faster architectures, ITEQ is one of the most logical names to capture the fallout.
  • A third name must be added: Elite Material. Less discreet than it appears, the group is also a key player in high-end laminates. For those seeking broader exposure to the theme of critical materials for AI servers, it constitutes another credible entry point.

Rubin is therefore not just a story of computing power. It is a story of infrastructure, and even of physics applied to infrastructure. In this new world, value flows toward those who enable the machine to deliver on its real, rather than theoretical, promises. High-end PCB material manufacturers, long relegated to the background, could well become some of the cleanest winners of the next leg of the AI cycle.