Morgan Stanley: Strong AI Chip Demand, NVDA Remains Top Pick

Overview

This report summarizes the key takeaways from Morgan Stanley analysts who attended an AI summit and related meetings in Taiwan. The report holds an optimistic view of the semiconductor industry, particularly for AI-related chips (GPU, CPU, ASIC) and memory sectors. It emphasizes NVIDIA's (NVDA) strong product cycle, Broadcom's (AVGO) stable custom chip business, and how persistent memory shortages will support industry prosperity.

Core views

For NVIDIA (NVDA), the report confirms that mass production of its next-generation Rubin chip is proceeding as planned, countering rumors of redesigns or delays. Supply chain verification at multiple points shows strong chip performance and steady ramp-up by year-end. The company's $20 billion annual CPU revenue target is considered reasonable by analysts, with approximately $10-12 billion from "head node" solutions containing CPUs and another $7-10 billion from standalone Grace and Vera CPUs, reflecting strong demand for ARM-based server CPUs. In the memory sector, the report reiterates that shortages will be the main bottleneck for AI infrastructure construction, with no short-term relief. The supply chain generally expects prices to rise by over 20% in Q3. Analysts show a slight preference for DRAM over NAND, viewing it as a more specific bottleneck despite both being in historic shortages. Broadcom's (AVGO) TPU (Tensor Processing Unit) custom-designed for Google is expected to grow strongly through next year. Although Google is collaborating with MediaTek on a low-cost alternative, Morgan Stanley projects AVGO will maintain an 80%+ share for several years, primarily due to its cost advantage in HBM (High Bandwidth Memory) procurement, where spot market prices are significantly higher than its long-term contract prices. In CPU market competition, AMD is gaining share in the x86 server market through its cost-performance advantages, with its 2nm "Venice" products set to begin mass production in the second half of the year. Intel's Panther Lake chip is progressing well on its 18A process with improving yields, but its long-term server product roadmap may partially shift to TSMC.

Analysis framework

Morgan Stanley's analysis approach is primarily bottom-up supply chain verification and customer interviews. By communicating with multiple links in the industry chain (including suppliers and clients), analysts have cross-verified key information about NVIDIA's product progress, CPU market demand, and memory supply-demand conditions. This method aims to penetrate market noise and rumors to obtain firsthand data to support investment judgments. The report particularly emphasizes distinguishing between "stories" (such as NVDA chip delays) and "facts" (supply chain verification). In terms of valuation, the report uses P/E ratios based on expected EPS for the next one to one and a half years for each company, combined with their respective growth potential and competitive landscape in the AI field to determine differentiated pricing.

Methodology notes

• Industry/Industry Analysis FrameworkSupply-demand framework

  The industry is primarily supply-driven

  In semiconductors, especially for advanced processes and memory, capacity expansion cycles are long and capital expenditures are large, making supply often the key factor determining prices and industry prosperity. This report analyzes supply bottlenecks in memory and advanced CPU/GPU sectors to judge industry trends.

• Valuation MethodPE/PEG valuation

  P/E valuation based on expected future earnings

  The report uses P/E ratios based on expected EPS for the next fiscal year (CY2027e or FY2027e) as the core valuation method for AMD, AVGO, INTC, and NVDA, applying different premiums or discounts based on each company's AI growth prospects.

• Competition & Strategy FrameworkMoat / competitive advantage

  Cost moat formed by long-term contracts

  The report points out that Broadcom (AVGO) has obtained a significant cost advantage in HBM procurement through long-term contracts with upstream suppliers, forming an important moat against low-cost competitors like MediaTek because spot market prices are far higher than its contract prices.

Asset mapping & comparison

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

• NVIDIA CORP (NVDA.US)

  Core beneficiary, AI GPU leader

  Strengths

  Strong product cycle (Rubin/Vera), better-than-expected CPU business, high ecosystem barriers

  Weaknesses

  High market share and gross margin limit short-term valuation expansion

  Comparison

  Far ahead of competitors like AMD in the data center GPU field

  Risks

  AI end-market demand lower than expected; cloud customers achieve breakthroughs in self-developed hardware

• Broadcom Inc. (AVGO.O)

  Benefits from custom ASIC (TPU) scaling

  Strengths

  Strong expertise in ASIC, stable relationships with major customers (Google), HBM procurement cost advantage

  Weaknesses

  Facing competition from low-cost suppliers like MediaTek

  Comparison

  Expected to maintain a significant share advantage over MediaTek (80/20) for the next few years

  Risks

  ASIC chips lose competitiveness; losing in the data center networking market to NVIDIA

• Advanced Micro Devices (AMD.O)

  Gaining share in the x86 CPU market

  Strengths

  High cost-performance in server CPUs, upcoming 2nm new product launch, strong supply chain confidence in AI GPU (MI455)

  Weaknesses

  Significant gap with NVIDIA in data center GPU field

  Comparison

  Has more share growth potential than Intel in the server CPU market

  Risks

  Intel's successful server CPU counterattack in 2027; data center GPU performance below expectations

• Intel Corporation (INTC.O)

  Foundry business has option value, but CPU faces competition

  Strengths

  Progress in 18A process, foundry business expected to gain partners

  Weaknesses

  Continuous competitive pressure from AMD in the processor market, ASP under pressure

  Comparison

  Still has a gap with TSMC in advanced process追赶

  Risks

  Foundry business fails to achieve实质性 success, leading to cost structure expansion; intensified competition from AMD

Key data

• NVIDIA Annual CPU Revenue Target$20 billionCompany target considered reasonable by analysts
• NVIDIA Standalone CPU Revenue Expectation$7-10 billionAnalyst estimate for 2026
• Memory Q3 Price Increase Expectation20%+Based on general supply chain expectations
• Broadcom TPU Business Expected Share80%+For the coming years

Impact & implications

The research report suggests that the current semiconductor industry landscape is beneficial to NVIDIA, Broadcom, and memory manufacturers. NVDA, with its leadership in AI computing and strong product cycles, will continue to dominate the market. AVGO's custom ASIC business model has high stickiness and high barriers, effectively resisting competition. Persistent memory shortages will provide related manufacturers with strong pricing power and profitability support. For investors, this highlights the long-term investment value of core AI infrastructure suppliers.

Risks

• AI end-market demand falls short of expectations, leading to significant cuts in GPU purchases by clients
• AMD becomes a strong competitor again in the data center GPU field
• Cloud customers (except Google) successfully develop competitive self-developed hardware
• Broadcom loses competitiveness in ASIC chips or loses in the networking market to NVIDIA
• Intel further loses share in the processor market

What to watch

• Actual production and delivery progress of NVIDIA's Rubin chips
• Memory price negotiation results for Q3 2026
• Customer feedback on AMD MI455 data center GPU and new customer acquisition
• Share changes between Broadcom and MediaTek in Google's TPU project
• Yield improvement of Intel's 18A process and subsequent product roadmap