Jeff Pu
Technology Equity Analyst at GF Securities
Zhen Ding remains a Buy as Nvidia's Rubin platform sharply raises PCB content per GPU. Content rises from about $440 per GPU on GB300 to about $750 on VR200 and roughly $1,000 on VR300, driven by higher-end CCL materials and rising layer counts. If PTFE is adopted in Rubin Ultra-era boards, content could exceed $1,500 per GPU. Zhen Ding has a new position in the VR200 compute tray. x.com
Zhen Ding Technology is gaining market share in Nvidia's VR200 platform while leveraging mSAP and the ABF supercycle; the Buy rating stands, with the target raised to NT$729. AI server PCB content is rising, at $750 per GPU for Nvidia Vera Rubin versus $400 for GB200, and AI server PCB TAM is projected at $13B in 2026 and $29B in 2027. Zhen Ding also benefits from mSAP dominance in optical modules and CPO, with mSAP capacity constrained in 2027 by 800G optical switch demand and yield limits. x.com
G2AIStack's reported component yield and capacity bottlenecks make NVIDIA's Rubin shipment outlook look dire. It says Rubin system expectations were cut from roughly 10,000 to around 5,000, citing GPUs, orthogonal backplanes and power systems; its ODM checks indicate shipments are falling short. x.com
Nvidia remains the core of the AI supply chain: in Morgan Stanley's bill-of-materials teardown of the next-generation Rubin VR200 rack, GPUs are still about half the total cost, as Nvidia sells rack-scale AI infrastructure. The teardown also shows HBM with the largest cost increase, NVLink Switch and other networking chips roughly doubling, and PCB content up 233%, so value in rack-level upgrades is spreading to memory and networking as bottlenecks shift to communication between GPUs. x.com
Nvidia's shares remain pressured because neoclouds have yet to show demand revenue can cover debt service even at higher rates. CoreWeave and Nebius are dragging the shares, though that concern is being priced in and digested. Reuters reports ByteDance, Alibaba and Tencent ordered $35B in Nvidia chips; Elon Musk says SpaceX is on track to bring roughly 200,000 GB300 GPUs online per month through end-2026. Nvidia made a higher low, and the next higher-high target is $256. x.com x.com x.com x.com
Nvidia can raise prices 15% and still leave customers plenty of margin because Vera Rubin delivers 2-30x more throughput per megawatt than Blackwell Ultra, depending on interactivity. A megawatt of Vera Rubin racks should generate at least twice the revenue of Blackwell racks, and much more in some configurations. SpaceXAI also appears to be a new standalone Vera CPU customer; Nvidia says Vera completes tasks up to 1.8x faster than x86 CPUs. x.com x.com
AI server PCB and substrate makers are among the safest, lower-risk, lower-reward ways to play Nvidia's racks. PCB value rose 233% in one NVL72 generation, second only to memory, and Morgan Stanley flagged PCB as the top downstream hardware gainer. Unimicron leads AI server PCB with 18-22% share; Zhen Ding gets nearly 70% of revenue from AI; Victory Giant is the most direct Nvidia PCB play; Elite Material's low-loss CCL passed Nvidia certification. TTM, Tripod, Compeq, WUS and Shennan also gain. … Show more
Nvidia can still reach $20 trillion by 2030, but it may not offer the best AI risk-reward in 2026. Custom silicon is gaining share, the CUDA moat matters less for inference, and Rubin could be delayed. x.com
High-grade PCB and low-loss glass-fiber makers Nitto Boseki, Elite Material and Taiwan Union Technology should keep rising as the obvious workaround to an inevitable memory despec for Nvidia's Rubin Ultra and Feynman. All three have more than tripled in the past year, and some are up over 10x in three years. Switching suppliers is harder here than in memory, which is why a few firms dominate, but it is higher risk and new supply coming online has not been analyzed. x.com
Nvidia's across-the-board beat shows AI capex has not hit the brakes, which keeps upstream memory demand rising and supports SK hynix. Q2 revenue rose 106% to $96.22B and data-center revenue 117% to $89B, and Q3 revenue guidance of $108B also beat expectations; Vera Rubin should make up about 20% of Q3 data-center revenue. Rising memory and other component costs are starting to pressure Nvidia's gross margin, a new point to watch in coming quarters. x.com
Nvidia's Q2 shows AI infrastructure demand still accelerating, making it a company you simply have to own if the AI revolution is as big as believed. Revenue reached $96.2B, up 106% y/y and 18% q/q, with data center revenue at $89.0B, up 117%, and gross margin steady at 75%. Q3 guidance of $108B assumes zero China data-center compute revenue, which may change. Vera Rubin is in full production. The after-hours dip is irrelevant next to these figures. x.com x.com
NVIDIA's near-term revenue is safe from the Rubin delay rumors, which concern Rubin Ultra and its Kyber rack, not base Vera Rubin. Rubin Ultra reportedly dropped from four chiplets to two over manufacturing concerns, and Kyber's PCB midplane is hard to make defect-free. The base Vera Rubin NVL72 runs on the existing Oberon rack and is ramping. Jensen Huang's $1T of Blackwell and Vera Rubin orders through 2027 does not depend on Rubin Ultra, which still targets 2027 via a 2+2 board design. x.com
