May AI Rack Shipments Dip Slightly MoM; Full-Year Doubling Forecast Unchanged
AI summary card
May AI Rack Shipments Dip Slightly MoM; Full-Year Doubling Forecast Unchanged
Morgan Stanley tracking shows May GB200/300 rack shipments at approx. 7.7k units; despite a 7% MoM decline, the full-year forecast of 70-80k units remains unchanged; Wistron is the top pick, followed by Hon Hai and Quanta.
- Total May GB200/300 rack shipments approx. 7.7k units, down 7% MoM
- Hon Hai May shipments approx. 3.3k units, down 11% MoM but Q2 expected to rise 18%
- Quanta May shipments 1.8-1.9k units; Q2 estimate adjusted down to 6.7k units accordingly
- Wistron May shipments 1.3-1.4k rack equivalents; revenue up 2% MoM
- Maintains 2026 full-year forecast of 70-80k rack shipments, up over 100% YoY
- ODM Preference Order: Wistron > Hon Hai > Quanta (based on target price upside)
- Overweight rating maintained for all three manufacturers
Report interpretation
Overview
This report is Morgan Stanley's monthly shipment tracking update for the Greater China Technology Hardware sector, focusing on May 2026 shipment data for GB200 and GB300 NVL72 AI server racks. Although overall May shipments saw a slight seasonal MoM decline, the firm remains bullish on the strong momentum of downstream rack assembly businesses, maintains an optimistic forecast for full-year 2026 shipments to double YoY, and has updated its investment preference order based on each manufacturer's target price upside potential.
Core views
Industry Volume & Annual Outlook: The report estimates total May 2026 GB200/300 rack shipments from major ODMs at approximately 7,700 units, down 7% MoM. This short-term fluctuation does not alter the strong full-year growth logic; the firm continues to forecast 2026 full-year rack shipments of 70,000 to 80,000 units, representing YoY growth of over 100% compared to approximately 29,000 units in 2025. Note that this data includes Wistron's compute tray (L10) rack equivalents and does not deduct subsequent L11 assembly and testing time, so actual deliveries to end customers may be slightly lower. Hon Hai Precision (2317.TW): As the largest player by market share, Hon Hai's May GB rack shipments were approximately 3,300 units, down about 11% MoM. Despite the single-month pullback, the firm expects its Q2 AI server rack shipments to maintain sequential growth, with quarterly volume likely reaching approx. 10,000 units, up 18% QoQ. Among the three major ODMs, Hon Hai still leads in absolute shipment scale. Quanta Computer (2382.TW): May revenue was NT$311.5 billion, down 8% MoM, primarily due to GB200/300 rack shipments falling from approx. 2,100 units in April to 1,800-1,900 units, while notebook shipments remained flat at 3.5 million units. Given the lower May shipments, the firm slightly adjusted Quanta's Q2 rack shipment estimate to approx. 6,700 units (still up 40% QoQ), with some shipments deferred to H2, but the full-year forecast of approx. 18,700 units remains unchanged. Wistron Corporation (3231.TW): May revenue was NT$290 billion, up 2% MoM, showing relative resilience. Growth was driven by subsidiary Wiwynn's 2% MoM revenue increase, a 6% rise in monitor shipments, and a slight increase in GB200/300 server compute tray shipments to 1,300-1,400 rack equivalents. The firm maintains its Q2 rack shipment estimate of approx. 4,200 units (up 4% QoQ), implying June shipments of approx. 1,500 rack equivalents. Based on target price upside, Wistron is currently listed as the top pick in the ODM sector.
Analysis framework
The report employs high-frequency monthly shipment tracking, obtaining actual or estimated GB200/300 rack shipments for each ODM via supply chain channel checks and cross-verifying them with revenue data. The analytical logic follows a transmission chain of 'Monthly Data → Quarterly Revision → Full-Year Outlook': first confirming whether single-month shipments deviate from expectations, then determining if it is temporary volatility or a trend shift, and subsequently deciding whether to adjust quarterly or full-year forecasts. Additionally, the firm introduces the concept of 'Rack Equivalents' to standardize statistical calibers across different production stages (e.g., L10 compute trays vs. L11 full racks), facilitating horizontal comparison of actual output contributions among manufacturers. Finally, combined with target price upside derived from valuation models, a dynamic investment preference ranking is formed.
Methodology notes
Monthly Shipment Tracking and Revenue Attribution
By decomposing company monthly revenue into shipment volumes and ASP changes for specific product lines such as AI racks, notebooks, and monitors, the drivers of revenue fluctuations are precisely identified. For instance, in this report, Quanta's MoM revenue decline is explicitly attributed to reduced AI rack shipments rather than weakness in the notebook business; this granular breakdown helps distinguish short-term disturbances from long-term trends.
Production Stage Standardization Conversion (Rack Equivalents)
Since different ODMs participate in different stages of AI server production (some only handle L10 compute trays, others handle L11 full system assembly), direct comparison of physical shipments creates bias. The report converts semi-finished goods into 'NVL72 Rack Equivalents' using conversion factors, making output comparable across manufacturers at different points in the value chain—a common technique for handling vertical specialization differences in hardware manufacturing research.
Residual Income Model (RIM)
The report uses the Residual Income Model for base-case valuation of all three ODMs. This method focuses on future excess returns generated by the enterprise (i.e., returns above the cost of equity). Compared to relying solely on P/E ratios, RIM is better suited for assessing the intrinsic value of capital-intensive manufacturing across different growth stages. Key assumptions include mid-term growth rates (7%-13%) and terminal growth rate (3%).
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Wistron (3231.TW)Top pick based on highest target price upside
- Strengths
- Positive MoM revenue growth, stable contribution from Wiwynn subsidiary, high potential for AI server penetration increase
- Weaknesses
- Valuation could be dragged if consumer electronics divestment progresses slower than expected
- Comparison
- Highest target price upside among the three major ODMs, outperforming Hon Hai and Quanta
- Risks
- Consumer electronics divestment below expectations, weak notebook demand, slowing AI server penetration, margin compression due to competition
- Hon Hai Precision (2317.TW)Second choice; largest shipment scale with high certainty of Q2 sequential growth
- Strengths
- Leading AI rack shipment share (~51%), iPhone sales and EV business provide diverse catalysts
- Weaknesses
- Significant MoM decline in May single-month shipments (-11%)
- Comparison
- Shipment scale far exceeds peers, but target price upside ranks second to Wistron
- Risks
- iPhone sales below expectations, low profit contribution from AI business, slow EV progress, geopolitical risks
- Quanta Computer (2382.TW)Third choice; near-term shipment pressure but solid full-year guidance
- Strengths
- Full-year forecast of 18,700 racks unchanged; notebook and Apple Watch demand provide additional support
- Weaknesses
- Significant MoM decline in May rack shipments; Q2 estimate adjusted downward
- Comparison
- Lowest target price upside among the three; near-term momentum weaker than Wistron and Hon Hai
- Risks
- Weak notebook/watch demand, rising labor costs compressing margins, intense price competition in hyperscale data centers, slowing AI server penetration
Key data
- May Total GB200/300 Rack Shipments~7,700 unitsDown 7% MoM; includes Wistron L10 compute tray rack equivalents
- 2026 Full-Year Rack Shipment Forecast70,000-80,000 unitsUp over 100% YoY vs. ~29,000 units in 2025
- Hon Hai May Rack Shipments~3,300 unitsDown ~11% MoM; Q2 expected to reach ~10,000 units (+18% q/q)
- Quanta May Rack Shipments1,800-1,900 unitsLower than April's ~2,100 units; Q2 estimate adjusted down to ~6,700 units
- Wistron May Rack Equivalent Shipments1,300-1,400 unitsUp 2% MoM; Q2 estimate maintained at ~4,200 units
- Quanta May RevenueNT$311.5 bn-8% MoM, +94% YoY
- Wistron May RevenueNT$290 bn+2% MoM, +39% YoY
Impact & implications
The report views 2026 as another strong year for downstream rack assembly businesses; despite normal fluctuations in monthly data, long-term demand for AI infrastructure construction remains unchanged. For investors, the current shipment pace validates that the capacity ramp-up of AI servers—from chips to system integration—is progressing steadily. The competitive landscape among ODMs is also evolving dynamically: Hon Hai occupies half the market凭借scale advantages, but Wistron has become the firm's current top pick due to a superior product mix and potential valuation elasticity; although Quanta faces near-term shipment pressure, its full-year guidance remains unchanged, indicating management confidence in catching up in H2.
Risks
- AI server penetration slower than expected
- Weak demand for traditional notebooks and consumer electronics
- Intensified price competition in hyperscale data center segment pressuring margins
- Rising labor costs eroding profitability
- Geopolitical developments affecting foreign capital allocation and supply chains
- iPhone sales below expectations (specific to Hon Hai)
- Consumer electronics divestment progressing slower than expected (specific to Wistron)
What to watch
- Subsequent monthly actual GB200/300 rack shipments from each ODM
- AI server business revenue mix and margin performance in Q2 earnings
- Whether H2 rack shipments can ramp up as scheduled to fulfill the 70-80k unit full-year forecast
- Specific timeline and transaction progress for Wistron's consumer electronics divestment
- Hon Hai EV business development milestones and new customer introductions