CITIC Securities: Focus on Computing Power Chain Inflation Main Line, Optimistic on Nvidia GTC Strengthening AI Industry Sustained Growth Confidence

CNBC Finance APP learned that CITIC Securities released a research report stating that NVIDIA’s GTC 2026 conference is approaching. The company’s chip product lineup is expected to expand further. In addition to the full set of six core chips for the Vera Rubin AI platform, more details about Rubin Ultra chips and server cabinets may be disclosed at the conference, bringing innovations in data connectivity, power supply, and other architectural designs. The rollout of new products such as orthogonal backplanes and CPO is expected to become more visible. Focusing on the inflation of computing power chains, with global demand for computing power continuing to exceed expectations, upstream sector prosperity and price increases are likely to persist. This remains the most certain “prosperity growth” theme for current technology sector allocations. The firm is optimistic that NVIDIA’s GTC 2026 will further strengthen market confidence in the sustained growth of the AI industry and the realization of incremental logic.

CITIC Securities’ main points are as follows:

Highlight 1: Rubin platform introduces a new chip combination, demonstrating extreme collaborative design.

At the 2026 International Consumer Electronics Show (CES), NVIDIA announced the Vera Rubin AI platform, which includes six core chips: Rubin GPU, Vera CPU, BlueField-4 DPU, NVLink 6 Switch, ConnectX-9 SuperNIC, Spectrum-6 Ethernet Switch. These encompass all major components within a server cabinet. The chips are upgraded to TSMC’s 3nm process and equipped with HBM4 memory, with comprehensive upgrades in memory capacity and bandwidth. This product lineup enhances the synergy between GPU and CPU, as well as interconnect chips, with a modular design that makes the overall cabinet more cohesive compared to the previous Blackwell generation.

Highlight 2: More details on Rubin Ultra may be disclosed, with expectations for architectural innovations in data connectivity and power supply.

Considering that at CES 2026, NVIDIA confirmed that the Vera Rubin platform has entered mass production, the firm believes that at GTC 2026, NVIDIA may reveal more details about Rubin Ultra chips and server cabinets. Besides the chip itself, which doubles computing performance by integrating four compute dies, two major architectural directions are noteworthy:

1. Data connectivity: Scale-up capacity will significantly increase. The copper backplane solution may be upgraded to a PCB orthogonal backplane (interconnecting compute and switch boards within canisters) plus optical interconnects (interconnecting canisters), forming a two-layer super network architecture. New processes/materials/products such as 78L RPCB, M9 CCL, Q glass electronics, and CPO are expected to be implemented.

2. Power supply system: Power and energy consumption are increasingly limiting the expansion of computing infrastructure. Solutions like 800V high-voltage DC (HVDC) power systems and modular power supplies are expected to be adopted, potentially leading to upgrades in embedded PCB processes and GaN third-generation semiconductors.

Highlight 3: NVIDIA may release a new inference chip, LPU, to strengthen inference product lines.

NVIDIA is expected to elevate AI inference to a system-level infrastructure, with LPU+CPX’s PD separation scheme enhancing inference products.

Regarding LPU: At GTC, NVIDIA is anticipated to launch a new inference chip integrating Groq LPU technology, likely based on a custom architecture designed specifically for large language model (LLM) inference. It will feature a redesigned tensor streaming processor (TSP) using SRAM on-chip storage, greatly increasing data storage and retrieval speeds, suitable for the high memory bandwidth demands of decoding stages.

Regarding CPX: The Rubin CPX launched in 2025 can effectively reduce prefill costs and may adopt GDDR7 or HBM3E as main memory standards. In terms of form factor, SemiAnalysis suggests CPX might shift from being integrated into the Rubin Compute Tray to a standalone server cabinet supporting NVL72 VR200 shipments. Industry chain information indicates that LPU may also be released as a 256-card LPX standalone cabinet.

Highlight 4: Outlook for next-generation Feynman architecture upgrades.

The design trend of NVIDIA’s next-generation Feynman architecture is gaining increasing industry attention, and NVIDIA may showcase related content at GTC 2026. Based on current industry information, TrendForce predicts Feynman will be among the first chips to adopt TSMC’s A16 process, with power delivery possibly utilizing backside power delivery (SPR), freeing up routing space, and potentially introducing 3D stacking technology for integrating Groq’s LPU hardware stack.

Production is expected to start around 2028, with deliveries beginning in 2029. While specific details of the Feynman architecture remain unclear, the firm believes that NVIDIA’s understanding of future AI computing infrastructure upgrade directions will be crucial. In the context of Moore’s Law slowing, how to support continuous AI industry iteration through innovations in computing, storage, and operational capacity, and how the roles of training and inference evolve, as well as the outlook for AI investment return cycles, NVIDIA may provide more inspiration and surprises at GTC.

Risk Factors:

Geopolitical risks, overseas leading computing power product launches falling short of expectations, AI market demand growth underperforming, rising prices of storage and other components, technological change and product iteration risks, policy regulation and data privacy risks, and increased competition in the PCB industry.