NVIDIA Vera Rubin Enters Full Production, CPU Performance Surpasses Expectations

Overview

This research report is a rapid commentary by Citigroup on NVIDIA's keynote speech at the opening day of COMPUTEX Taipei. The report highlights that NVIDIA's next-generation Vera Rubin platform has entered full production, and it released a dedicated Vera CPU designed for AI agents, which significantly outperforms traditional x86 architecture. Additionally, updates were also made in network interconnectivity and consumer PC segments. Based on strong earnings expectations, Citigroup maintains its buy rating and $300 target price for NVIDIA.

Core views

Product and Capacity Progress: NVIDIA's Vera Rubin platform has entered full production, eliminating investor concerns about potential delays. This platform integrates GPU, CPU, storage, and networking into a single AI supercomputer, improving agent throughput by 10 times compared to the Blackwell architecture. Meanwhile, the Spectrum-X Ethernet photonic technology incorporating co-packaged optics has also gone into production, offering 5x energy efficiency improvement, enabling the construction of million-GPU AI factories, with CoreWeave, Microsoft, and OCI among the first ecosystem partners. Rising Strategic Importance of CPU: NVIDIA launched a dedicated Vera CPU designed for AI agents. Its performance is 1.8x higher than x86 architecture in common AI agent tools like Python and code analysis; in SQL workloads and data processing, its performance is 3x and 6x that of AMD/Intel x86 competitors respectively. Citigroup emphasizes that as AI agents evolve, CPUs are becoming bottlenecks for GPUs. Historically, NVIDIA sold 2 GPUs per 1 CPU, but now there are independent CPU racks (256 per system), making the CPU to GPU ratio approach 1:1. Anthropic and OpenAI have become early adopters of the Vera CPU. Consumer PC Innovation: NVIDIA partnered with Microsoft to launch the RTX Spark superchip based on ARM architecture, aiming to reshape Windows PCs in the age of AI. This chip combines a Blackwell RTX GPU with a 20-core Grace CPU designed in collaboration with MediaTek, delivering 1 petaflop of AI performance and up to 128GB unified memory, capable of running 120B parameter large language models locally, balancing long battery life and thin design.

Analysis framework

Citigroup analyzes NVIDIA's investment value from three dimensions: product rollout timing, system architecture evolution, and end-market expansion. First, by confirming that core components such as Vera Rubin and Spectrum-X have entered full production, the company's supply chain execution is validated, directly addressing market concerns about delays. Second, the report captures the new demand scenario of AI agents, pointing out that compute bottlenecks are shifting from GPUs to CPUs. Through quantitative comparisons and configuration ratio changes, it demonstrates NVIDIA’s incremental space in the CPU market. Finally, by reviewing early customer endorsements and cross-industry partnerships in consumer PCs, it reinforces the continuous expansion of its ecosystem.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply-demand framework

  Structural Demand Growth Due to System Bottleneck Shift

  The report points out that in the age of AI agents, CPUs are becoming bottlenecks for GPUs. This judgment is based on the common sense that data processing demands surge in system architectures: when GPU compute power increases dramatically, if the CPU responsible for scheduling and preprocessing cannot keep pace, overall efficiency suffers. Therefore, NVIDIA increasing CPU configuration ratios (from 2:1 to 1:1) essentially addresses new supply-side bottlenecks to uncover new hardware demand growth.

• Valuation MethodPE/PEG valuation

  P/E Valuation Based on Historical Averages

  When setting the $300 target price, Citigroup applied a P/E multiple of approximately 28 times and benchmarked it against NVIDIA's average P/E over the past three years. This valuation method assumes that after experiencing high growth, the company's valuation will revert to historical averages, deriving a reasonable market cap through forecasting future profitability.

Asset mapping & comparison

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

• NVIDIA Corp (NVDA)

  Core focus of the report analysis, benefiting from AI compute demand and CPU product expansion

  Strengths

  Absolute leader in AI compute, next-generation platforms produced on schedule, complete software and hardware ecosystem

  Weaknesses

  Risk of demand volatility in gaming and data center markets

  Risks

  Loss of gaming market share, slow adoption of new platforms, impact of cryptocurrency mining on gaming sales

• CoreWeave (CRWV)

  Early ecosystem partner and adopter of NVIDIA's Spectrum-X photonic Ethernet technology

  Strengths

  First to adopt the latest million-GPU AI factory networking technology

  Comparison

  Early adopters alongside Microsoft and OCI

Key data

• Target Price$300Based on a ~28x P/E multiple and CY27E EPS forecast of ~$11.9
• Vera CPU SQL Performance3xPerformance multiplier vs. x86 architecture (AMD/Intel)
• Vera CPU Data Processing Performance6xPerformance multiplier vs. x86 architecture
• Spectrum-X Energy Efficiency Improvement5xvs. networks using traditional optical modules
• Grace CPU ShipmentsNear 2.5MHistorical cumulative shipments

Impact & implications

The report suggests that NVIDIA successfully expanded its product line beyond standalone GPU accelerators into more fundamental general computing by launching dedicated CPUs and significantly increasing their ratio in AI systems, bringing significant incremental revenue to its data center business. At the same time, the mass production of networking technologies and the introduction of ARM-based chips for PCs further consolidate its moat in AI infrastructure and edge AI ecosystems.

Risks

• Intensifying competition in gaming may lead to loss of market share and drag stock performance.
• If adoption of the new platform lags expectations, it could reduce data center and gaming sales.
• Order volatility in automotive and data center markets may increase stock and valuation multiples swings.
• Cryptocurrency mining activities may unpredictably affect gaming GPU sales.