PCB, CCL and substrate supply chain for Nvidia AI-server platforms Report Interpretation

Overview

This supply-chain report assesses whether PTFE adoption in Nvidia's Rubin Ultra NVL576 architecture disrupts the established high-speed CCL upgrade cycle. J.P. Morgan concludes that PTFE is likely to be selective and technically challenging, while the wider migration from M8+ toward M9 and M10 remains the more important industry trend.

Core views

J.P. Morgan addresses investor questions about whether Rubin Ultra could adopt PTFE in its PCB materials. Nvidia is exploring multiple material and board-manufacturing options to meet an electrical-loss target of Df 0.0004 or below for extreme-connectivity projects such as backplanes and NVL576 switch boards. Most M10 samples tested at roughly Df 0.00043-0.00045, short of the target. Glass-less M10 laminate samples came within 5% of the target because removing glass improves signal transmission, but this sacrifices the mechanical support that glass provides. The report considers an M10/PTFE hybrid the more likely Rubin Ultra NVL576 solution rather than a pure PTFE or fully glass-less construction. PTFE can achieve roughly Df 0.0003-0.0004, while retaining glass in the PTFE component can preserve structural support. A full PTFE stack-up faces a shortage of sufficient high-temperature lamination capacity and weaker drilling performance; glass-less boards also present challenges in metal-to-laminate attachment, high-voltage durability, drilling soft materials and limited supply-chain experience. Full glass-less PTFE could become an option for more extreme-connectivity applications over the longer term if these barriers are solved, but J.P. Morgan believes it is premature to conclude that it can overcome all practical constraints for large switch boards. On market size, the report estimates that if 20% of Rubin Ultra GPUs are used in NVL576, the switch-tray CCL opportunity would be about US$400 million, roughly split 50:50 between PTFE and M10. Its scenario table places NVL576 switch-tray CCL TAM at US$200 million, US$400 million and US$800 million for 10%, 20% and 40% penetration, respectively; corresponding PTFE TAM is US$100 million, US$200 million and US$400 million. Actual material demand could be higher because board yield remains under development and is viewed as unlikely to reach 50% even by late-2027 mass production. Removing glass would lower CCL ASP by more than 20% relative to typical material, although the report estimates margins could be around 60% versus about 50%. J.P. Morgan reiterates that high-speed CCL remains one of the fastest-growing industries as data-transmission requirements accelerate. PTFE may be used in advanced AI-cluster switch boards and backplanes, but mainstream Nvidia NVL72 compute and switch boards should remain M8+ for the next two to three years. The firm therefore sees the PTFE discussion as confirmation that the CCL upgrade cycle is accelerating: compute trays, switch trays and midplanes could migrate from M8+ to M9 and then M10 in Rubin Ultra or Feynman generations. Rubin Ultra specifications are expected to be finalized in 1Q/2Q27, one to two quarters before expected mass production in 2H27, creating potential share-price volatility around Nvidia's material updates. For Elite Material Co (EMC), J.P. Morgan says PTFE would affect less than 3% of its 2027E revenue. It nevertheless identifies EMC as the leading beneficiary of high-end CCL upgrades, citing more than 70% M8 and more than 90% M9 market share and a positive view of its M10 position. The report also points to EMC's resources to develop M11 and a PTFE portfolio if necessary. For Nan Ya Plastics (NPC), the report highlights an integrated operating model that can adapt material formulas to customer specifications and sees potential M8-M9 order wins in 2H26-2027 as upside to estimates. Its existing M10 formulations contain glass cloth and are epoxy-based, so a PTFE choice could require further modification. NPC has previously produced low-Df PTFE CCL, although it has not been a main product because of high cost, processing difficulty and low board yield. J.P. Morgan expects most margin improvement over the next two years to come from M7-M8 orders from large hyperscalers, with an M10 win representing additional upside optionality. It expects AI-grade CCL to exceed 50% of revenue mix by 2028, versus less than 10% in 2025, and raises its December 2027 target price to NT$350 while forecasting ROE above 40% by 2027. For Nittobo, J.P. Morgan expects broader NER glass-cloth use in M8 and M9 to delay NEZ commercialization to 2028E from as early as end-2027E. It believes PTFE-based solutions will generally still use NE/NER glass cloth to improve stiffness and mechanical strength, limiting the direct earnings impact even if glass-less technologies emerge. Nittobo is estimated to be planning roughly 60-70% year-on-year NER capacity growth from FY26-end to FY27-end, supporting revenue growth and its leading position, though the report will monitor pricing because entry barriers appear lower than for T-glass. For Victory Giant Technology, the report cites a stronger order outlook into 2027 and management's plan to bring forward its RMB100 billion annual value-output target by one year to end-2027E. As Nvidia platforms evolve, J.P. Morgan expects PCB content value to rise through both higher-end material use and more complex fabrication. It argues that capacity scale, proven high-volume delivery and close collaboration with Nvidia should be important determinants of wallet share and reinforce existing suppliers' technology moats.

Analysis framework

J.P. Morgan compares electrical performance, structural support, drilling and thermal-management trade-offs across glass fiber, M-series CCL and PTFE. It then applies NVL576 adoption scenarios to estimate CCL and PTFE TAM, before assessing the transmission of material changes to leading CCL, glass-cloth and PCB suppliers.

Methodology notes

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  AI-server material specifications are traced through CCL, glass cloth, copper foil and PCB fabrication suppliers.

  The report links Nvidia's connectivity requirements and material choices to demand, product mix, manufacturing complexity and competitive positioning across the supply chain.

• Industry AnalysisSupply-demand framework

  Scenario analysis of NVL576 penetration, switch-tray demand and CCL/PTFE market size.

  J.P. Morgan assumes GPU shipments, rack counts, NVL576 penetration, material use per tray and CCL-sheet pricing to derive potential market size.

• Industry AnalysisVolume-price decomposition

  Material ASP, board yield and margin comparison between glass-containing and glass-less CCL.

  The report separates volume and yield effects from pricing and margin effects to explain why material-market value may differ from the baseline TAM.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Elite Material Co (2383.TW)

  Primary beneficiary of the high-speed CCL upgrade trend; PTFE exposure is limited.

  Strengths

  More than 70% M8 and more than 90% M9 market share; resources to develop M11 and PTFE products.

  Comparison

  J.P. Morgan views EMC positively in M10 as well as its dominant M8/M9 positions.

  Risks

  PTFE and M10 specification updates could cause near-term volatility; PTFE impact is estimated at less than 3% of 2027E revenue.

• Nan Ya Plastics Corp (1303.TW)

  CCL supplier positioned for M8-M9 orders and potential M10 optionality.

  Strengths

  Integrated model, formula flexibility, low-Df PTFE CCL production track record and new copper-foil capacity.

  Weaknesses

  Current M10 formulas are glass-cloth, epoxy-based and may require modification if PTFE is selected.

  Comparison

  Unlike most Taiwanese CCL peers, the company has a track record in low-Df PTFE CCL.

  Risks

  PTFE's high cost, processing difficulty and low board yield may constrain adoption; M10 order wins remain optional.

• Nittobo (3110.T)

  Glass-cloth supplier benefiting from broader NER use in M8 and M9 materials.

  Strengths

  NER capacity expansion and a leading market position.

  Weaknesses

  NEZ launch has been postponed to 2028E.

  Comparison

  Entry barriers in NER appear lower than in T-glass.

  Risks

  Pricing dynamics and the commercialization timing of next-generation glass require monitoring.

• Victory Giant Technology-H (2476.HK)

  PCB fabricator positioned to benefit from rising material and manufacturing complexity in Nvidia platforms.

  Strengths

  Stronger order outlook into 2027, accelerating capacity expansion and high-volume delivery capability.

  Comparison

  J.P. Morgan sees close collaboration between incumbent AI PCB vendors and Nvidia as reinforcing their technology moat.

  Risks

  Wallet share remains dependent on capacity scale and demonstrated delivery capability.

Key data

• Electrical-loss targetDf 0.0004 or belowTarget cited for extreme-connectivity projects including backplanes and NVL576 switch boards.
• M10 sample performanceDf 0.00043-0.00045Most M10 samples fell modestly short of the stated loss target.
• PTFE electrical performanceDf 0.0003-0.0004Estimated range cited as a reason PTFE is being considered.
• NVL576 switch-tray CCL TAMUS$200mn / US$400mn / US$800mnJ.P. Morgan scenarios at 10% / 20% / 40% NVL576 penetration.
• Nan Ya Plastics AI-grade CCL revenue mix>50% by 2028 versus <10% in 2025J.P. Morgan estimate of product-mix improvement.
• Nittobo NER capacity growth~60-70% YoYEstimated increase from FY26-end to FY27-end.

Impact & implications

The report sees selective PTFE use as an incremental technology development rather than a break in the high-speed CCL upgrade cycle. Material qualification, yield and manufacturing feasibility will determine whether PTFE expands beyond selected NVL576 applications, while M8-to-M10 migration continues to underpin demand for established CCL, glass-cloth and PCB suppliers.

Risks

• Glass-less and PTFE solutions face unresolved issues in structural support, metal-to-laminate attachment, high-voltage durability and drilling soft materials.
• Pure PTFE deployment is constrained by high-temperature lamination capacity and processability challenges.
• Board yield is still under development and is considered unlikely to reach 50% even by late-2027 mass production.
• Nittobo faces potential pricing pressure because barriers to entry in NER glass cloth appear lower than for T-glass.

What to watch

• Nvidia's final Rubin Ultra CCL specification, expected in 1Q/2Q27 before projected 2H27 mass production.
• Progress in PTFE lamination capacity, drilling and board-yield improvement ahead of the Feynman generation.
• Nan Ya Plastics' M8-M9 order wins and any M10 qualification outcome.
• Nittobo's NER pricing dynamics and next-generation glass commercialization timing.
• Victory Giant Technology's order trajectory and capacity expansion toward its end-2027E output target.