In August 2026, supply-chain news within the industry stated that NVIDIA plans to introduce PTFE material into its NVSwitch boards, and to make it the primary material choice for the AI-accelerator-tray orthogonal backplane of its new-generation server platform Rubin Ultra (expected to launch in 2027). This news quickly fermented in the capital markets, and PTFE overnight became the hottest concept on the A-share market. The collective explosion of the sector stems from the market’s imagination of a “materials revolution”; beneath the clamor, the more central question is: is this actually a real industrial upgrade, or an over-embellished technological narrative? When the concept hype recedes, who exactly are the real beneficiaries?
Why Does PTFE Win? — From “King of Plastics” to “Signal Highway”
The direct reason PTFE was chosen by NVIDIA is its dielectric performance in ultra-high-frequency signal transmission scenarios. Under 10GHz conditions, PTFE’s dielectric constant Dk is about 2.1, and its dielectric loss Df is as low as 0.0004. These two indicators are at a leading level among commercial polymers. More importantly, the variation of PTFE’s dielectric parameters with frequency and temperature is extremely small, which means that the loss and distortion of a signal during transmission can both be effectively controlled.
Understanding this question requires returning to the physical structure of the AI server rack. Taking NVL72 as an example, one rack integrates 72 GPUs, achieving collaborative computing through NVLink high-speed interconnection. In the center of the rack a piece of orthogonal backplane is inserted vertically; this backplane does not carry any chips, and its only function is to transmit high-speed signals. The backplane is the component with the longest wiring distance in the entire rack, and the signal-crossing distance can reach several dozen inches. On the Rubin Ultra platform, the backplane signal rate will rise from the current 224Gbps to more than 337Gbps. The higher the frequency, the more significant the dielectric loss, and the physical limits of traditional copper-clad laminate materials begin to show.
The M9-resin-with-quartz-cloth and M10-hydrocarbon-resin solutions have already approached the theoretical ceiling on dielectric-loss indicators. PTFE’s advantage lies in its molecular structure — the carbon backbone is uniformly wrapped by fluorine atoms, the molecule has almost no net polarity, and it is not prone to polarization loss in an alternating electric field. This physical property makes PTFE a low-loss channel for high-frequency signal transmission.
But the scope of PTFE’s application needs to be accurately defined. This solution does not replace the entire backplane with PTFE material, but rather adopts a hybrid-lamination structure: PTFE is used to replace the electronic cloth only in the backplane’s high-speed signal transmission layer, while the outer layers and structural layers continue to retain traditional electronic cloth and M9 material. The choice of a hybrid-lamination solution itself shows that what PTFE solves is the signal-loss problem in a specific scenario, not all of the engineering problems of the entire board. For non-signal-transmission core areas such as GPU computing boards, power layers, and auxiliary PCBs, traditional materials are still in use.
“Replacement” vs. “Coexistence” — the Misread Technology Route
The market’s most core misjudgment lies in interpreting “PTFE used in the orthogonal backplane” as “PTFE comprehensively replacing electronic cloth.” The reality is far from that dramatic. The actual winning solution for PTFE is the hybrid-lamination structure — PTFE only replaces the electronic cloth in the backplane’s high-speed signal transmission layer, and the board’s outer layers and structural layers still retain traditional electronic cloth and M9 material. To make an analogy, this is like building a racetrack: the straights and high-speed curves are paved with the most top-grade special asphalt to guarantee extreme speed, while the buffer zones and roadbed still use conventional materials to guarantee strength and cost. The two are complementary in their division of labor, not a life-and-death struggle.
If you take the PCB apart by function, the replacement boundary is very clear: PTFE takes the “high-speed signal channel” slice of the pie, not all of the electronic cloth’s application scenarios. Even if Rubin Ultra introduces a PTFE hybrid-lamination backplane, the traditional-material demand for GPU computing boards, power layers, and auxiliary PCBs is still fully retained. Calculated at a 30% hybrid-lamination penetration rate, the traditional-material demand of the entire rack drops only 8% to 11%, far from a “cliff-edge decline.” What is more, the cost of PTFE copper-clad laminate is about three times that of the traditional electronic-cloth solution. Under the premise of such a huge cost gap, PTFE will only penetrate the very few high-end scenarios where “the performance premium can cover the cost increment.” For the vast majority of AI servers — including the Rubin standard version — the glass-fiber-reinforced solution is still the most cost-effective solution. Stretching out the time dimension, the conclusion becomes even clearer: even in the most optimistic scenario, PTFE’s large-scale mass production will only come in the second half of 2027, still at least 12 to 18 months away from contributing significant performance. During this period, the boom cycle of traditional electronic-cloth materials is being realized at an accelerated pace. The exponential growth of AI computing-power demand itself is continuously enlarging the total pie of the entire PCB-materials market, and PTFE’s diversion effect is far from enough to shake the fundamental base.
The Real Bottleneck of Technology Implementation — Why It’s Not “Done in One Step”
Even though PTFE has obvious performance advantages, going from the laboratory to mass-production implementation, there are still three engineering obstacles that need to be overcome.
First, the metal-lamination bonding-strength problem. In traditional copper-clad laminate, glass cloth not only undertakes the dielectric function but also plays a mechanical-reinforcement role. After removing the glass cloth, the adhesion between the copper foil and the PTFE resin declines, and delamination risk easily arises during the processing and use of large-size rack backplanes.
Second, the drilling-process challenge. PTFE material is relatively soft, and conventional mechanical drilling easily causes hole-wall tearing or burrs; although laser drilling is feasible, the yield in large-batch production is still not ideal.
Third, high-voltage tolerance. After removing the glass-fiber cloth, the long-term reliability data of the material under high-voltage conditions is not yet sufficient, and an additional verification cycle is needed.
These engineering problems are precisely the reason NVIDIA tends to adopt an M10-and-PTFE hybrid solution — glass-free M10 is responsible for reducing signal loss, while the PTFE layer that retains glass cloth undertakes the mechanical-support function. A pure-PTFE solution does not yet possess mass-production feasibility under existing equipment and process conditions. Industry research reports judge that the M10/PTFE hybrid solution is most likely to be used in the NVL576 switch tray of Rubin Ultra, i.e., the extreme interconnection scenario of an 8-rack cluster. Meanwhile, the mainstream design of the single-rack NVL72 will still mainly use the M8+CCL solution in the next two to three years.
The Real Beneficiary Landscape of Each Link in the Industry Chain
After the tide of concept hype recedes, what is truly worth watching are those companies that can turn this complex manufacturing into stable mass production. First look at TUC (Taiwan Union Technology) EMC. Its market share in M8 material exceeds 70%, and in M9 exceeds 90%. PTFE’s potential impact on it is less than 3% of its total 2027 revenue — basically negligible. The real logic is: NVIDIA is pushing the upgrade on NVL72 from M8+ to M9 and then to M10, and EMC is still the most certain beneficiary on this main line.
Now look at the domestic industry chain. China’s total PTFE capacity is 208,100 tons per year, accounting for 67% of the global total, but the import dependence for high-end electronic-grade and semiconductor-grade products is still very high. In June 2026 the entire industry raised prices uniformly by 5%, marking the establishment of a bottom reversal. The PTFE value per single rack rose from USD 3,000 to 4,000 in the M9 era to USD 12,000 to 16,000 for Rubin Ultra, a growth of 3 to 5 times. The PTFE CCL market size is expected to reach RMB 8 billion (approx. USD 1.18 billion) in 2027. But investors need to keep their eyes sharp — ordinary PTFE resin has long been in overcapacity domestically, with price wars fought to the point of bloodied heads. Any company that only talks about total PTFE capacity without talking about electronic-grade certification progress is fishing in muddy waters. Whether it can produce high-purity electronic-grade PTFE resin, whether it has entered the M10 certification process, and how its yield performs — these are far more important than pure concept.
At the upstream resin end, Juhua Co., Haohua Technology, and Dongyue Group have formed a three-strong tripartite standoff. Haohua Technology’s PTFE capacity is 51,000 tons per year, and its 5,000-ton-per-year 5N-grade ultra-high-purity electronic-grade PTFE dedicated line had already achieved full-capacity operation in the first quarter of 2026. Juhua Co.’s PFA semiconductor-grade product has already achieved mass production. In the midstream precision-processing link, Wote Co. (Walton) has laid out the PTFE-precision-parts and semiconductor-cleaning-equipment supply chain through acquisition, with the smallest market capitalization and the highest odds. But one must also soberly see that major Taiwanese manufacturers have already entered the field — TUC, Taiflex, and Unimicron have entered into a tripartite strategic cooperation, formally initiating a project to develop PTFE hybrid-lamination material to NVIDIA specifications. NVIDIA has a high-voltage line that no single supplier’s share should exceed 50%. Even if Shengyi Technology sits firmly as the first supplier, there are still Taiflex and TUC chasing closely behind. At the upstream resin end, Haohua leads, but both Dongyue and Juhua are accelerating their pursuit.
Local Upgrade and Structural Opportunity
PTFE entering NVIDIA’s supply chain is, in essence, a pragmatic adjustment that NVIDIA makes to its material solution under the engineering goal of racing toward a 337Gbps signal rate. What it solves is the loss problem of the orthogonal backplane’s high-speed signal transmission layer, and it does not affect all the functional layers of the entire board. The hybrid-lamination solution, the threefold cost gap, and the post-2027 mass-production timetable — these three constraints together determine the scope of PTFE’s influence.
And at present, the biggest risk lies not in whether the technology is feasible, but in whether the understanding of the technology route is accurate. Equating “verifying PTFE in the orthogonal backplane’s high-speed signal layer” with “NVIDIA abandoning electronic cloth across the entire platform” is replacing industrial facts with trading sentiment. The real opportunities are concentrated on two types of enterprises: first, the traditional copper-clad-laminate leaders that continue to realize performance on the M8-to-M10 upgrade main line; and second, the fluorochemical enterprises with industrialization capabilities in the high-end electronic-grade PTFE field. PTFE entering the supply chain is a real industrial development, but its scope of influence, timing, and beneficiary entities all need to be reassessed within the framework of engineering constraints and the mass-production pace.
[Disclaimer]: The above content reflects analysis of publicly available information, expert insights, and BCC research. It does not constitute investment advice. BCC is not responsible for any losses resulting from reliance on the views expressed herein. Investors should exercise caution.