Morgan Stanley: NVDA Remains the Top Choice for AI Chips

Overview

This Morgan Stanley report summarizes key findings from the Taiwan AI Summit and related meetings, maintaining a positive view on the semiconductor industry, especially AI-related chip manufacturers. It confirms that NVIDIA's next-generation Rubin chips are progressing smoothly in production and highlights that memory shortages remain a significant bottleneck for AI infrastructure development. The report also provides updates on business dynamics and competitive landscapes for major companies such as NVIDIA, AMD, Intel, and Broadcom.

Core views

Demand Side: AI compute demand remains robust, driving demand for high-end GPUs and配套 CPUs and memory. NVIDIA confirms its $200 billion CPU revenue target for 2026, which, although surprising, has been validated by multiple supply chain contacts, with approximately $100-$120 billion coming from integrated solutions and $70-$100 billion from Grace or Vera standalone CPU sales. Supply Side: NVIDIA's next-generation Rubin chips are in production as planned, with minor delays in the past few weeks having little impact on silicon production. The performance is strong, and there are no significant challenges compared to the previous generation Grace Blackwell. The lifecycle of the previous Blackwell product has been extended by several months, with higher-than-expected production volumes, reflecting strong market demand. Memory-wise, both DRAM and NAND face record shortages, with DRAM being a more specific bottleneck, expected to persist in the foreseeable future. Multiple supply chain contacts anticipate a DRAM price increase of over 20% in the third quarter. Competitive Landscape: In the CPU sector, AMD is set to gain further market share in the server market due to its competitive pricing and performance. Its 2nm product 'Venice' is expected to begin production in the second half of the year. In the ASIC space, while Google is collaborating with MediaTek to develop low-cost alternatives, Morgan Stanley expects Broadcom to maintain an 80%+ share in Google's TPU business due to its cost advantages in HBM procurement. Valuation Perspective: The report provides valuation methodologies for each company, such as AMD's target price of $360 (based on 37x EPS for FY2027), Broadcom's 28x ModelWare EPS for CY2027, Intel's 42x CY2027 EPS, and NVIDIA's 22x CY27 EPS, reflecting a discount relative to peers.

Analysis framework

This report employs a bottom-up supply chain validation approach, leveraging interactions with various upstream and downstream industry players during the Taiwan AI Summit to cross-validate product progress, supply-demand conditions, and market expectations for key companies. Analysts focus on several critical variables: whether NVIDIA's new product production ramp is proceeding smoothly; the severity of memory shortages and their impact on AI deployment; CPU market competition dynamics and share shifts; and potential competitive threats in the custom ASIC space. Through this primary information, analysts update and confirm their core views and financial forecasts for leading industry companies.

Methodology notes

• Industry/Supply Chain Analysis FrameworkSupply and Demand Framework

  The core contradiction in this industry is the inability of supply to meet the explosive growth in demand for high-end AI chips and配套 memory.

  The report identifies the current semiconductor industry's core driver as AI, with supply-side bottlenecks in advanced process chips and high-bandwidth memory (HBM) leading to sustained shortages and price increases. This framework is crucial for assessing company performance and stock price trends.

• Valuation methodsForward P/E Valuation

  采用远期市盈率（Forward P/E）进行估值以评估高增长的半导体公司。

  The report employs forward P/E valuations for companies like AMD, AVGO, INTC, and NVDA, based on future fiscal years' EPS, reflecting the market's mainstream pricing approach for high-growth tech stocks.

• Industry/Supply Chain Analysis FrameworkSupply Chain Upstream and Downstream Transmission

  通过验证产业链不同环节的信息来交叉确认核心假设。

  The report emphasizes the importance of cross-verifying information across the supply chain to confirm NVIDIA's Rubin chip production status and the reality of memory shortages, effectively filtering market noise for reliable insights.

Asset mapping & comparison

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

• NVIDIA Corporation (NVDA.US)

  Core beneficiary, absolute leader in AI GPUs

  Strengths

  Smooth product cycle (Rubin/Vera), CPU business exceeding expectations, high ecosystem barriers

  Weaknesses

  High market share and margins limit short-term valuation expansion potential

  Comparison

  Leading AMD in data center GPUs, an emerging but rapidly growing force in CPUs

  Risks

  AI end-market demand underperformance, AMD GPU competitiveness revival, cloud customers developing in-house hardware

• Advanced Micro Devices (AMD.O)

  Benefits from Intel's weakness in CPUs, likely to continue gaining market share

  Strengths

  High性价比 server CPUs, strong performance of 2nm new products, potential to catch up in AI GPUs

  Weaknesses

  Significant gap with NVIDIA in GPUs

  Comparison

  Share gainer in x86 CPUs,追赶者 in AI GPUs

  Risks

  Intel successfully counters with competitive server CPUs in 2027, underperformance in data center GPUs

• Broadcom Inc. (AVGO.O)

  Benefits from AI ASIC customization demand, particularly Google TPU orders

  Strengths

  Leading technology in ASICs, long-term customer relationships, cost advantages (e.g., HBM procurement)

  Weaknesses

  Facing competition from客户自研 (Customer Owned Tooling) and low-cost suppliers like MediaTek

  Comparison

  Dominant in the high-end custom ASIC market but must watch for long-term share erosion

  Risks

  Loss of competitiveness in ASICs, losing major customers, underperformance in VMware integration

• Intel Corporation (INTC.O)

  Under pressure in the short term due to CPU shortages but provides option value with its Foundry business

  Strengths

  Progress in Foundry business (e.g., Panther Lake), server CPU shortages present short-term opportunities

  Weaknesses

  持续 losing share to AMD in CPUs, long-term competitiveness questioned

  Comparison

  On the defensive in traditional CPUs, a追赶者 in the foundry sector

  Risks

  Intensified competition from AMD leading to further share loss, unsuccessful development of Foundry business

Key data

• NVIDIA's 2026 CPU Revenue Target$200 billionThe target is deemed reasonable and has been validated by the supply chain.
• AMD Target Price$360Based on 37x EPS for FY2027.
• Expected Q3 Memory Price Increase>20%Anticipated by most supply chain contacts.
• Broadcom TPU Business Expected Share80%+For the next few years, despite competition from MediaTek.

Impact & implications

The report's findings reinforce its positive stance on AI semiconductor leaders. NVIDIA, as the absolute leader, continues to dominate with its successful product cycle. Memory vendors will benefit from ongoing shortages and price increases. AMD is poised to expand its CPU market share. Broadcom's ASIC moat is deeper than market concerns. Overall, the AI infrastructure buildout is still in its early stages, with clear growth prospects for key suppliers.

Risks

• AI终端市场需求不及预期，导致客户大幅削减GPU采购
• AMD在GPU领域重新成为有力竞争者，缩小与英伟达的差距
• 云客户（除谷歌外）成功开发出有竞争力的自研硬件
• 英特尔在2027年推出有竞争力的服务器CPU，成功夺回市场份额

What to watch

• AMD's MI455 product launch in July 2026 and customer feedback
• Memory price negotiation results in Q3 2026
• Actual performance and cost of NVIDIA's Vera Rubin chips
• Progress and impact of MediaTek's low-cost ASIC chips for Google