NVIDIA Just Dropped Two Major Chips – PCB Changes Every Hardware Engineer Must Know

June 1, 2026, Taipei Nangang Exhibition Center. Jensen Huang walked on stage and dropped two "nuclear bombs" in under an hour.

One is called RTX Spark – NVIDIA's official entry into the PC chip market.
The other is Vera Rubin – AI superchip now in full production, already onboarded by OpenAI and SpaceX.

Within hours, TSMC's 3nm lines ran at full capacity, SK Hynix's HBM supply tightened, and the PCB industry – often the overlooked "supporting actor" – suddenly found itself at a major technology inflection point.

1. AI PC is here – PCB precision now rivals chip packaging

The RTX Spark superchip (flagship N1X) is co‑developed with MediaTek. It packs a 20‑core Grace CPU and a Blackwell GPU with 6144 CUDA cores, delivering 1 PFLOP of AI compute and supporting 128GB of unified memory. This marks the moment when running large language models locally no longer requires the cloud – your PC can do it.

What does this mean for PCBs?

1.1 HDI: no longer a "nice‑to‑have", now a "must‑have"

In the past, HDI (high‑density interconnect) was mostly for high‑end smartphones. Now AI PC motherboards demand the same – or even higher – routing density. An AI PC board must pack more power layers, denser high‑speed signal lines, and more complex BGA fan‑out into the same footprint.

The industry is rapidly moving from basic HDI (1‑2‑1) to advanced HDI (3‑2‑3, any‑layer). Advanced HDI almost always requires mSAP (modified semi‑additive process). Traditional subtractive etching suffers severe side‑etching when line width shrinks below 30µm, destroying line profile and impedance consistency. mSAP creates more vertical, uniform copper traces with precise control down to 15‑20µm, meeting the density demands of AI chips.

1.2 Materials: from ordinary FR‑4 to ultra‑low‑loss laminates

AI PCs run LLMs and real‑time inference locally, which means CPU/GPU composite chips draw significantly more power and push massive data throughput. PCB materials must upgrade from standard FR‑4 to ultra‑low‑loss laminates (M7, M8 grades) to maintain signal integrity under sustained high loads.

Whether it's high‑order HDI or mSAP, the conclusion is the same: the PCB is no longer just a "carrier board" – it's becoming the core interconnect medium.

2. AI data centers: PCB value per server has tripled

While the PC‑side changes are still unfolding, demand for PCBs in data centers has already exploded.

2.1 Orthogonal backplanes: replacing copper cables with PCBs

NVIDIA's Rubin platform (Vera Rubin, Rubin Ultra) ramps production in 2026‑2027, massively increasing per‑rack compute density. A key trend: orthogonal backplane technology replaces traditional copper cables – backplane PCBs have jumped from 20‑30 layers to 78 layers. The PCB is no longer auxiliary; it's the backbone.

78 layers means micron‑level manufacturing precision, extremely tight registration tolerance, and a completely different class of process control. Industry estimates suggest the PCB value in a single Rubin server has more than doubled compared to the previous generation.

2.2 ASIC explosion: new demands on PCB design flexibility

Beyond NVIDIA GPUs, custom ASICs from major North American cloud providers are accelerating. In 2026, ASIC‑based AI servers are expected to account for roughly 27.8% of the market.

ASICs don't come with well‑tuned reference designs. Every customer's chip layout, pinout, and power distribution is different. This creates a new requirement for PCB manufacturers: high customization, quick‑turn prototyping, flexible stackups – exactly where small‑to‑medium batch PCB makers excel.

2.3 CPO arrives: PCBs must now work with light

Alongside Rubin, NVIDIA announced volume production of Spectrum‑X Ethernet silicon photonics – the world's first CPO (co‑packaged optics) based Ethernet switch.

In simple terms, CPO places the optical engine and switching chip on the same substrate. The electrical signal distance from chip to optical interface shrinks from centimeters to hundreds of micrometers, and port power drops from 14‑16W to about 5.5W. The goal: push AI cluster energy and latency to their limits.

This means PCBs must maintain exceptional signal integrity at the electrical‑optical conversion boundary. High‑speed signal pins on optical modules are extremely dense, demanding strict length matching, impedance control, and advanced materials. Standard FR‑4 is no longer enough – the PCB must simultaneously be an electrical bridge and optical infrastructure.

3. How PCB manufacturers can ride this wave

For both AI PC and AI data center tracks, the PCB supply chain faces a forced technology upgrade. Here's what it takes:

3.1 Process: embrace advanced HDI and mSAP

Advanced HDI is now the entry ticket to AI hardware. Major Asian PCB makers are aggressively expanding mSAP, CoWoP, and high‑order HDI capacity. The global mSAP PCB market is projected to grow at a CAGR of roughly 17% from 2025 to 2028.

3.2 Materials: get comfortable with ultra‑low‑loss laminates

Moving from FR‑4 to M6/M7/M8 grade ultra‑low‑loss materials changes PCB cost structures, but the performance threshold is now fixed. Manufacturers must work closely with material suppliers to master the processing characteristics of each high‑speed/high‑frequency laminate – including specialized lamination cycles, drilling parameters, and surface finishes.

3.3 Positioning: find your niche

For a small‑to‑medium batch manufacturer like AnyPCBA, competing head‑to‑head with giant Taiwanese PCB houses on volume isn't realistic. The real opportunities are:

• Fragmented ASIC orders – diverse designs that don't fit mass production lines.

• AI hardware startups – rapid iteration needs small, flexible batches.

• Traditional industrial equipment makers – just beginning their AI upgrade journey and needing a partner who understands PCB materials and design optimization.

Final thoughts

2026 is being called the first year of CPO commercialization and the year AI PC goes mainstream. The PCB industry is moving from the background to center stage – but unlike GPUs or HBM, PCB manufacturing is highly fragmented. No single giant can serve everyone.

For small‑to‑medium batch PCB makers, the opportunity is clear: the diversity of AI hardware, fragmented ASIC orders, and fast‑turn prototyping needs are a natural fit for flexible, responsive, non‑standardized manufacturing.

AnyAnyPCBA, founded in 2011, specializes in small‑to‑medium batch (5–5,000 pieces) PCB fabrication and PCBA assembly. Whether you need an AI PC motherboard, an ASIC evaluation board, or a CPO optical module carrier, we provide flexible production – from DFM review to fully traceable data.

👉 AnyPCBA website: https://www.anypcba.com/
Small‑to‑medium batch PCB & PCBA | Flexible · Traceable