Nvidia and Qualcomm Release New AI Strategies, Industry Enters Era of Agentic AI and Physical AI

Overview

This Citi report systematically reviews core technology releases and strategic moves from Nvidia GTC and Qualcomm Computex keynotes, analyzing their transmission impact on global AI infrastructure evolution and the China tech industry chain. The report points out that the industry is undergoing a key paradigm shift from foundation Large Language Models (LLM) to Agentic AI and Physical AI; technology iteration will reshape the supply chain landscape and accelerate the landing process of AI in edge scenarios.

Core views

Nvidia completes strategic upgrade: Vera Rubin platform (based on 3nm process, integrating seven dedicated chips, adopting HBM4 memory and 45°C liquid cooling technology) has entered full commercial production, confirming supply chain (Foxconn, Wistron, Pegatron) will start capacity ramp-up in H2 2026. Company positioning transforms from GPU supplier to full-stack AI infrastructure provider, launching Vera CPU designed for 'Agentic AI' (88-core monolithic architecture, 1.2TB/s LPDDR5 memory bandwidth, SQL task performance 3x improvement over x86, real-time data stream processing 6x faster), and jointly launches RTX Spark laptop platform with Microsoft (1 petaflop compute power, 128GB unified memory), extending AI capabilities to edge devices. Lenovo as a core ecosystem partner is expected to benefit from the volume release of new GPU server platforms and high-ASP Windows laptops. Qualcomm accelerates all-scenario layout: Releases Dragonfly data center exclusive product line, aiming to connect consumer devices and backend infrastructure compute continuum; Snapdragon platform is becoming the native execution layer for third-party AI agents (such as Claude Desktop, Gemini), and extends edge silicon capabilities to high-growth areas such as AI-defined vehicles (digital cockpit agent + physical perception system), robots (hierarchical computing architecture), and 6G wireless (AI-centric network architecture, supporting digital twin construction). Impact shows structural differentiation: Nvidia supply chain companies and Lenovo receive positive catalysts; but Vera CPU and Spark platform adopt SOCAMM/LPDDR5X memory architecture replacing traditional RDIMM, if Nvidia CPU market share in servers increases, may erode DIMM chip demand in the long term, posing challenges to memory interface chip manufacturers such as Montage Technology. The report also observes that Chinese software and ICT enterprises are actively migrating to agent provider and tokenized business models, the technology routes of the two giants jointly suggest the adoption pace of AI edge applications in China may exceed expectations.

Analysis framework

The report adopts an event-driven analysis framework, closely tracking technical details and strategic statements from key releases by global tech giants. It first extracts core parameters of new products (such as Vera Rubin mass production status, Vera CPU architecture innovation, Dragonfly positioning), then deduces impacts on upstream and downstream links (manufacturing, components, ecosystem partners) through supply chain transmission logic, and combines industry technology cycle stage judgment (paradigm leap from LLM to Agentic AI) to evaluate mid-to-long-term trends and regional opportunities.

Methodology notes

• Industry/Industrial Analysis FrameworkUpstream-Midstream-Downstream Supply Chain Transmission

  Supply Chain Transmission Analysis

  By analyzing the impact path of upstream technology releases (such as Nvidia's new platform) on midstream manufacturing and downstream application links, judge the benefit or damage logic for related companies. This report deduces the positive impact on supply chain companies such as Foxconn and Lenovo, and the potential impact on Montage Technology.

• Industry/Industrial Analysis FrameworkProduct life cycle

  Technology Paradigm Evolution Analysis

  Divides industry development into different technology stages (such as from foundation LLM to Agentic AI), predicts subsequent market opportunities and risks by identifying the current stage and inflection points. This report points out that the industry has entered a new stage of Agentic AI and Physical AI, implying that China AI edge applications may accelerate landing.

Asset mapping & comparison

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

• Foxconn

  Benefit: Nvidia Vera Rubin platform mass production drives supply chain demand in H2 2026

• Wistron

  Benefit: Nvidia Vera Rubin platform mass production drives supply chain demand in H2 2026

• Pegatron

  Benefit: Nvidia Vera Rubin platform mass production drives supply chain demand in H2 2026

• Lenovo Group

  Benefit: As an Nvidia ecosystem partner, will benefit from new GPU server platforms and high-ASP Nvidia Windows laptops

• Montage Technology

  Potential Damage: Nvidia Vera CPU and Spark platform adopt SOCAMM/LPDDR5X memory architecture, may erode traditional DIMM chip market demand

Key data

• Vera Rubin Platform Process3nmIntegrates seven dedicated chips, adopts HBM4 memory architecture
• Vera CPU Memory Bandwidth1.2TB/sIndustry's first enterprise-grade CPU adopting LPDDR5
• RTX Spark Platform Compute Power1 petaflopLocal AI compute performance, supports complex multimodal autonomous agents
• Vera CPU Performance Comparison3x faster for SQL tasks, 6x faster for real-time data streamsCompared to x86 processor benchmark results

Impact & implications

The report believes that the strategic moves of Nvidia and Qualcomm will accelerate global AI infrastructure iteration and increase the depth of participation of the China tech industry chain in links such as AI edge devices and data center solutions; technology architecture changes (such as memory solution innovation) may reshape the competitive landscape in segmented fields, and related enterprises need to adjust technology routes in time to cope with the paradigm shift.

Risks

• Nvidia adopting LPDDR5X memory architecture may erode traditional DIMM chip market demand in the long term, posing a long-term challenge to memory interface chip manufacturers such as Montage Technology

What to watch

• Qualcomm Dragonfly product details (planned to be announced on June 24, 2026)
• Actual volume release progress of Nvidia Vera Rubin platform supply chain in H2 2026
• Market share changes of Nvidia CPU in the server processor market and its impact on memory interface chip demand