DRAM supply becomes the new core ranking factor for AI compute semiconductors, with NVDA and AMD leading and QCOM ranking last

Overview

This report updates Citigroup's ranking of U.S. AI compute semiconductor stocks. The key change is that DRAM/HBM memory supply allocation has been elevated to the top weighting and is considered alongside factors such as accelerators, CPUs, and sales exposure to hyperscale cloud customers. The report favors companies with strong memory supply partners and greater supply visibility.

Core views

AI compute demand remains strong and supply-constrained, with memory shortages being the largest bottleneck limiting compute supply. NVDA leads the mega-cap group with strong HBM partnerships, while AMD leads the large-cap group with supply support from Samsung and Micron; QCOM is constrained by LPDDR memory allocation availability and remains at the bottom of the ranking despite launching its HBC solution.

Analysis framework

The report uses a relative ranking framework to compare compute semiconductor companies across memory supply allocation, accelerator capabilities, CPU capabilities, and sales exposure to hyperscale cloud customers, and re-ranks them in the context of post-1Q earnings.

Methodology notes

• Relative stock rankingCitigroup compute semiconductor updated rankings

  Higher weighting on memory supply allocation

  The report argues that DRAM/HBM supply shortages are the biggest constraint on AI compute supply, and therefore places memory supply allocation ahead of factors such as accelerators, CPUs, and cloud customer sales exposure.

• Fundamental ratingCiti Research 12-month investment rating framework

  Buy/Neutral/Sell based on expected total return and risk

  The disclosed appendix states that Citi's equity ratings include Buy, Neutral, and Sell, with target prices typically based on a 12-month horizon; the main focus of this report is relative ranking rather than target price changes.

Asset mapping & comparison

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

• NVIDIA CORP (NVDA.O)

  Leading mega-cap AI compute semiconductor name

  Strengths

  Strong HBM partnerships, high memory supply visibility, and ranked first in the mega-cap group in Citigroup's updated rankings.

  Weaknesses

  The report does not elaborate on company-specific weaknesses; the main external constraint remains industry-wide HBM/DRAM supply tightness.

  Comparison

  Ranks above AVGO in the mega-cap group.

  Risks

  HBM supply deliveries falling short of expectations, a slowdown in cloud AI demand, or easing GPU supply-demand tightness could weaken its lead.

• ADVANCED MICRO DEVICES INC (AMD.O)

  Leading large-cap AI compute semiconductor name

  Strengths

  Receives supply support from both Samsung and Micron, ranking first in the large-cap group.

  Weaknesses

  Still dependent on allocation from key memory suppliers and faces strong competition from NVDA and others.

  Comparison

  Ranks above CBRS in the large-cap group.

  Risks

  Memory supply execution, AI accelerator competition, and the pace of cloud customer adoption are the main uncertainties.

• QUALCOMM INC (QCOM.O)

  Lower-ranked compute semiconductor name

  Strengths

  Has launched a high-bandwidth compute HBC solution and uses LPDDR memory stacking.

  Weaknesses

  The report believes its memory allocation is limited through 2029, placing it at the bottom of the updated rankings.

  Comparison

  At a clear disadvantage relative to NVDA, AMD, and other compute semiconductor companies.

  Risks

  If LPDDR supply allocation does not improve, commercialization of HBC products and revenue contribution may be constrained.

• Broadcom Inc (AVGO.O)

  Important mega-cap AI compute semiconductor comparison name

  Strengths

  The report assumes about an 85% memory fulfillment rate for its TPU, with strong exposure to cloud custom chips.

  Weaknesses

  Still ranks behind NVDA in Citigroup's ranking.

  Comparison

  Ranks behind NVDA in the mega-cap group.

  Risks

  TPU memory fulfillment rates, concentration of cloud customer demand, and the pace of custom chip deliveries are key risks.

• Micron Technology Inc (MU.O)

  Key beneficiary in the DRAM/HBM supply chain

  Strengths

  The report says it has signed agreements through 2030 with multiple strategic customers, highlighting the scarcity of memory supply and its bargaining position.

  Weaknesses

  The report does not provide detailed company-level weaknesses.

  Comparison

  Within the ranking framework, MU represents the increased importance of memory supply capability itself.

  Risks

  DRAM/HBM cycle volatility, capacity expansion timing, and execution of customer agreements are the main risks.

• Cerebras Systems Inc (CBRS.O)

  Second-ranked comparison name in the large-cap group

  Strengths

  Uses on-chip SRAM rather than HBM, following a different path from traditional HBM-dependent solutions.

  Weaknesses

  Its cloud customer sales exposure and ability to scale supply still need to be validated.

  Comparison

  Ranks behind AMD in the large-cap group.

  Risks

  Commercialization of its technology path, customer adoption, and competition with the GPU/HBM ecosystem are the main uncertainties.

Key data

• AWS EC2 GPU instance price changeUp 20%The report treats this as a recent signal that AI compute demand remains in short supply.
• Term of Micron strategic customer agreementsThrough 2030The report believes these agreements include customers such as MSFT/Dell and show that DRAM supply allocation is locked in over the long term.
• Mega-cap group rankingNVDA first, followed by AVGONVDA leads due to strong HBM partnerships; the report assumes about an 85% memory fulfillment rate for AVGO TPU.
• Large-cap group rankingAMD first, CBRS secondAMD benefits from Samsung/Micron supply support; CBRS uses a different on-chip SRAM solution rather than HBM.
• QCOM rankingLastDespite launching its HBC solution, the report says its memory allocation is limited through 2029.

Impact & implications

The investment implication is that competition in AI compute semiconductors is expanding from pure chip performance to control over the memory supply chain. Companies with long-term HBM/DRAM partnerships and greater supply visibility may achieve higher relative valuations and better order fulfillment capability; companies with insufficient memory allocation may lag in revenue realization and competitive ranking even if they launch new products.

Risks

• Persistent DRAM/HBM shortages may continue to constrain AI compute supply.
• Long-term strategic customer agreements lock in memory capacity, potentially compressing allocation space for late entrants or weaker-supply vendors.
• If cloud AI demand cools or GPU instance prices decline, the report's short-supply thesis may weaken.
• QCOM's limited memory allocation through 2029 may affect HBC product realization.
• Citigroup disclosed investment banking, non-investment banking, or other commercial relationships with multiple covered companies, so investors should interpret the report cautiously in light of conflict-of-interest disclosures.

What to watch

• AWS EC2 GPU instance prices and other AI compute supply-demand signals.
• DRAM/HBM capacity allocation and long-term customer agreements from suppliers such as Micron and Samsung.
• NVDA's HBM partnerships and supply execution.
• AMD's memory supply assurance from Samsung/Micron.
• Improvement in QCOM's memory allocation around and after 2029 and progress in HBC commercialization.
• AVGO TPU memory fulfillment rates and changes in hyperscale cloud customer orders.