NVIDIA Rubin Ultra PCB Material Selection: PTFE Offers Superior Performance but Yield Concerns Remain, HC Solution Emerges as Transitional Option
nashnova research
Nvidia is testing two glass-free PCB material systems for the Rubin Ultra NVL576 due in late 2027 — PTFE delivers the best signal performance but faces serious manufacturing-yield hurdles, while HC offers easier fabrication at the cost of lower performance. The outcome will reshape the upstream supply chain.
What is Nvidia deciding?
The core question is which PCB material to use for the next-gen AI server Switch Tray: PTFE — polytetrafluoroethylene, an ultra-low-loss polymer — or HC — hydrocarbon resin, a more process-friendly alternative.
The original plan paired a PTFE copper-clad laminate (CCL, the base board) with HC prepreg (PP, the bonding layer). The new plan under test switches to an all-HC system — HC-based CCL plus HC PP — with a glass-reinforced HC PP as a compromise option.
This means → Nvidia has not picked a winner. It is weighing peak performance against manufacturing readiness, running both tracks in parallel.
Where does PTFE excel — and where does it stall?
PTFE's dielectric loss — signal energy lost as it travels through the material — is significantly lower than HC's. After formulation tuning, it can even exceed the M10 spec, a high-end transmission benchmark. On paper, full PTFE glass-free is the best answer.
But PTFE is soft and extremely non-stick, making PCB lamination and drilling far harder. In plain terms = signals run fast on this material, but factories can't press it flat or drill it clean — scrap rates spike.
With less than a year before Rubin Ultra ships, Nvidia cannot afford a long yield-ramp cycle. Production readiness now outranks raw performance.
Is the HC fallback actually viable?
HC's lower manufacturing bar is driving its appeal — yet some suppliers warn that the glass-free HC CCL-plus-PP combination's yield is still well below mass-production requirements.
Whether it can pass Nvidia's qualification is uncertain; every candidate formulation requires ongoing sampling and certification.
This means → HC is not a "proven backup." It is a slightly easier-to-make alternative that also hasn't cleared the bar — neither path is production-ready yet.
How are suppliers placing their bets?
Four Taiwanese firms — TaiFlex Scientific, EMC, Asia Electronic Material, and Microcosm Technology — are all validation-stage and betting primarily on PTFE, treating it as a major product-transition opportunity. Their HC investment is limited.
China's Shengyi Technology takes a different approach. It is the lead CCL supplier for the current Vera Rubin NVL144 Switch Tray and is developing both PTFE and HC in parallel, with relatively advanced progress on glass-free certification.
This means → if Nvidia ultimately shifts to HC, the four Taiwanese makers heavily committed to PTFE face shipment delays, while Shengyi's dual-track positioning gives it more flexibility.
Why is Nvidia stripping out glass fiber at all?
Signal quality: conventional glass cloth is woven from glass yarn. The weave pattern causes signal-phase distortion and dielectric loss. In plain terms = the texture of the glass weave makes electrical signals wobble and slow down — the faster the AI server, the worse this gets.
Supply bottleneck: upgrading to M9-grade PCB requires quartz cloth and other premium glass fabrics. Only a handful of global suppliers exist, capacity is persistently constrained, and even expansion may fail to keep pace with AI server demand growth.
This reflects a dual motive: Nvidia is not chasing performance alone — it is de-risking the supply chain, solving the signal problem while sidestepping an upstream capacity ceiling.
What does the final pick mean?
The PTFE-versus-HC decision will directly determine whether the four Taiwanese suppliers can deliver on this product-transition cycle on schedule.
It will also test Shengyi Technology's dual-track strategy — if the HC route wins, the supplier that clears certification first takes the largest share.
In plain terms = this is not just a materials-engineering question. It is a betting contest: the supplier on the right track eats; the one on the wrong track may miss the entire Rubin Ultra cycle.
市场有风险，内容仅供研究参考，不构成投资建议。