GPU compute remains in short supply, so the softer May SIA data does not change the constructive semiconductor view
AI summary card
GPU compute remains in short supply, so the softer May SIA data does not change the constructive semiconductor view
Morgan Stanley believes rumors about Meta's cloud services are more a sign of stronger GPU external demand than supply in the AI era, rather than excess GPU demand at hyperscale customers; though May SIA data came in softer, AI, storage, and second-half demand visibility continue to support the semiconductor complex.
- GPU compute remains tight, and hyperscaler GPU resale/rental is more likely to reflect excess industry demand rather than excess compute demand.
- NVIDIA benefits more clearly in the secondary compute market due to its de facto industry standard status in GPUs, broader compute versatility, faster deployment, reliability, and asset lifespan.
- May SIA sales grew +16.1% m/m, below Morgan Stanley's +22.0% estimate but above the 10-year average of +6.9%; 3-month year-over-year growth accelerated to +104.2%.
- DRAM and NAND were both below forecasts, but 3-month year-over-year revenue reached fresh highs; the report believes the AI-driven supply-constrained cycle is continuing.
- The report continues to prefer storage exposure to MU and SNDK, as well as related names NVDA, AVGO, CBRS, LRCX, KLAC, MKSI, ON, ADI, and NXP.
Report interpretation
Overview
This report is Morgan Stanley's North America semiconductors weekly note and focuses on the implication of possible Meta cloud services for GPU supply-demand, plus May 2026 SIA semiconductor sales data. The report argues that current market concerns of oversupply at hyperscale customers are overly bearish and that a more reasonable interpretation is broad industry-wide GPU demand still outstripping supply, leading some customers to lease GPUs externally. While May SIA data came in below consensus, especially for broad markets and storage softness, AI remains the strongest segment and AI-linked storage supply constraints with improving second-half demand visibility continue to support a constructive view.
Core views
The key thesis is: first, the GPU compute market remains tight, cloud demand is still strong, and the Meta-related rumor should not be interpreted as industry-wide compute oversupply; second, the GPU resale trend is constructive for NVIDIA share, as NVIDIA is the sector standard with broad usability and ecosystem advantages ahead of ASIC, while AMD is more general than low-volume ASIC but the secondary compute market is structurally smaller; third, May SIA data below expectations does not alter the strong Q2 and improving H2 outlook; fourth, the storage cycle has moved from traditional inventory normalization toward AI-driven structural supply constraints, with DRAM remaining the clearest bottleneck in AI deployment and NAND recovery looking more durable than market expectations.
Analysis framework
The report combines qualitative event interpretation, cloud-demand signal validation, monthly SIA data, product-segment and regional splits, historical seasonality, 3-month year-over-year trends, and company ratings/valuation tables to form conclusions. For GPU, analysis focuses on the supply-demand implications behind secondary GPU leasing behavior and the substitution differences among NVIDIA, AMD, and ASIC. For SIA, analysis focuses on overall sales, regional mix, broad market, storage, and price/shipments versus estimates and historical averages.
Methodology notes
Assess whether industry supply is oversupplied or short by interpreting hyperscaler behavior in leasing GPU capacity externally.
The report argues that if the most value-accretive use of GPUs is to lease them out, this more likely indicates aggregate excess demand at industry level; in addition, NVIDIA's versatility and de facto standard status make it better positioned in the secondary compute market.
Compare May SIA sales data against analyst forecasts, historical seasonality, and year-over-year trends.
The report uses total sales, regional composition, product categories, shipments, and ASP changes to gauge semiconductor cycle strength, differentiating AI demand, end-device recovery, and storage supply constraints.
Interpret the strong price dynamics in DRAM and NAND as supply constraints driven by AI build-out rather than merely inventory rebuilding.
The report highlights customers' attempts to pre-book supply; DRAM and NAND revenue and ASP year-over-year growth reached record highs, supporting the view that the storage earnings window could remain extended.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- NVIDIA CORP (US.NVDA)Core beneficiary
- Strengths
- De facto GPU industry standard, broad compute applicability, fast deployment, reliability, asset life, and strong per-gigawatt-token efficiency; well suited for the secondary compute market.
- Weaknesses
- Some customers' capex contribution may vary seasonally and there is a risk the market misreads hyperscale compute demand.
- Comparison
- Compared with ASIC, NVIDIA GPUs are more easily leased and widely used; compared with AMD, NVIDIA has higher market share and a much larger potential secondary compute market.
- Risks
- Fluctuations in cloud-capex, competitive alternative acceleration, and negative repricing of industry supply-demand balance.
- META PLATFORMS INC (US.META)GPU demand and cloud-services event catalyst
- Strengths
- Potentially entering cloud services could show its GPU resources have external commercialization value; it also remains an early adopter of AMD Helios.
- Weaknesses
- The report believes Meta capex is no longer the primary driver of GPU upside as it was in prior years.
- Comparison
- The report moves Meta from a primary upside driver to one among many potential sources of GPU demand, with demand leadership rotating among customers.
- Risks
- Cloud-service details remain unfinalized, and the market may interpret GPU external leasing as a sign of weaker internal demand.
- ADVANCED MICRO DEVICES INC (US.AMD)GPU competitor and potential early-adopter candidate for Meta
- Strengths
- More versatile than low-volume ASICs; Meta may be an early adopter of AMD Helios, and the warrant structure is equivalent to at least an initial 1GW taking roughly 75% incentive discount.
- Weaknesses
- Overall market share is below NVIDIA, and the secondary compute market may be smaller.
- Comparison
- Higher leaseability than ASIC, but materially below NVIDIA in standard-setting status and ecosystem scale.
- Risks
- Even if Meta adopts AMD Helios, the report does not see a meaningful impact on NVIDIA procurement.
- Micron Technology Inc. (MU) / SanDisk Corporation (SNDK)Storage-tightness beneficiaries
- Strengths
- Strong DRAM and NAND price and revenue trends; AI build-out is making storage a critical bottleneck with limited supply flexibility.
- Weaknesses
- Monthly data were below expectations, and short-term shipment deceleration caused a mild annual forecast downtick.
- Comparison
- The report views DRAM as the clearest bottleneck, and NAND recovery durability as stronger than investors typically expect.
- Risks
- If supply is released faster than expected or demand softens after customers pre-book supplies, pricing sustainability could weaken.
Key data
- May SIA total sales+16.1% m/mBelow Morgan Stanley's +22.0% forecast, but above the 10-year average of +6.9%.
- May SIA 3-month year-over-year growth+104.2%Accelerated from April's +93.9%; one-month year-over-year growth was +118.8%.
- Regional year-over-year growthAmericas +150.7%, APAC +126.3%, China +101.9%, Europe +83.7%, Japan +39.4%Americas had the highest growth.
- DRAM monthly performance+27.7% m/mBelow the +43.0% consensus and the +45.9% five-year average; 3-month year-over-year revenue growth was +304.8%, a new high since 2001.
- NAND monthly performance+40.7% m/mSlightly below the +43.8% consensus but above the +25.6% five-year average; 3-month year-over-year revenue growth was +364.6%, a historical high.
- Yearly forecast revision2026 outlook revised down from +103% to +99%The main factor was shipping deceleration in storage; the report still views growth as meaningful.
- Semiconductor company inventory days114 daysUp 2 days month-over-month, above the historical median of 23 days.
Impact & implications
From an investment perspective, the report continues to prefer exposure to AI compute, storage, and selected high-quality semiconductor supply-chain names. GPU shortages and the leasing trend reinforce NVIDIA's ecosystem advantages; storage tightness supports MU and SNDK; though broad markets were softer in May, the improving recovery breadth benefits simulation, MCU, equipment, and supply-chain companies. A near-term risk is that investors may continue to worry about cloud capex pacing or misread leasing behavior as demand deceleration, but the report argues fundamental evidence remains largely constructive.
Risks
- May SIA data came in below forecast, which could trigger concerns about the slope of semiconductor recovery.
- Investors may misinterpret hyperscaler GPU leasing as weak internal compute demand or industry-wide overcapacity.
- After the monthly storage trend shifted from seasonality to supply-driven conditions, forecasting has become harder.
- The broad market and storage remain vulnerable to near-term volatility, and inventory days are above the historical median.
- Cloud-capex pacing, AMD-ASIC competition, supply-chain cost pass-through, and margin persistence could all affect sector performance.
What to watch
- Whether cloud-demand and GPU procurement guidance in Q2 earnings remains robust.
- Details of Meta's cloud-service plans and the actual impact on NVIDIA and AMD procurement timing.
- DRAM and NAND pricing, shipment trends, customer pre-booking, and the pace of supply release.
- Whether June and subsequent SIA data confirm that the May softness was a temporary fluctuation.
- Subsequent earnings and sector activity from ASML, TSMC, TXN, AMD, INTC, NXP, KLAC, QCOM, and peers.