May AI Rack Shipments Dip Slightly MoM; Full-Year Doubling Expectation Unchanged

Overview

This research report is Morgan Stanley's monthly tracking update for the Greater China tech hardware sector, focusing on shipment data for GB200 and GB300 NVL72 AI server racks in May 2026. The core conclusion is: despite a single-digit MoM decline in May shipments, full-year AI rack shipments for 2026 will still achieve more than double YoY growth, maintaining high prosperity in the downstream assembly segment. Among the three major ODM manufacturers, the institution's preferred order based on potential upside to target prices is Wistron, Hon Hai, and Quanta, while maintaining an Overweight rating for all three.

Core views

At the industry aggregate level, total estimated shipments of GB200/300 NVL72 equivalent racks in May were approximately 7,700 units, down about 7% MoM. Despite fluctuations in monthly data, the institution maintains its 2026 full-year shipment forecast of 70,000 to 80,000 units, representing a YoY increase of over 100% compared to approximately 29,000 units in 2025. It should be noted that this statistical data includes Wistron's compute tray (L10) converted rack counts, without deducting the time required for subsequent rack assembly and testing; therefore, the actual quantity delivered to end customers may be lower than the aforementioned figures. For Quanta (2382.TW), May revenue was NT$311.5 billion, down 8% MoM and up 94% YoY. The MoM revenue decline was mainly due to reduced GB200/300 rack shipments, with estimated monthly shipments of 1,800 to 1,900 units, slightly below April's approx. 2,100 units; laptop shipments remained flat at 3.5 million units. Given the weak performance in May, total Q2 rack shipments may be slightly revised down to approx. 6,700 units (up 40% MoM), with some shipments delayed to the second half of the year, but the full-year forecast of approx. 18,700 units remains unchanged. Wistron (3231.TW) reported May revenue of NT$29.0 billion, up 2% MoM and 39% YoY. Growth drivers came from three aspects: subsidiary Wiwynn revenue increased 2% MoM to NT$84.1 billion; monitor shipments reached 950k units, up 6% MoM; GB200/300 server compute tray shipments were estimated to be equivalent to 1,300 to 1,400 racks, up slightly 2% MoM. Although notebook and desktop shipments declined 6% and 25% MoM respectively, the institution maintains its Q2 rack shipment estimate of approx. 4,200 units (up 4% MoM), implying June shipments of approx. 1,500 rack equivalents. Hon Hai (2317.TW) saw May GB-series rack shipments decline approx. 11% MoM to 3,300 units. However, looking ahead to Q2, the institution believes its AI server rack shipments will continue to grow MoM, with total Q2 volume expected to reach approx. 10,000 units, up 18% MoM. Among the three ODMs, Hon Hai remains the largest manufacturer by absolute shipment volume, dominating the industry supply share.

Analysis framework

The institution employs a high-frequency monthly shipment tracking method, obtaining rack-level shipment data from various ODM manufacturers through supply chain surveys, using it as a core leading indicator to judge the prosperity of the AI server industry chain. The analysis主线 focuses on "monthly volatility vs. annual trends": single-month MoM declines are viewed as normal rhythm adjustments, while the full-year YoY doubling growth expectation serves as the pricing anchor. At the company comparison level, the institution does not look solely at absolute shipment volumes but combines them with each company's upside to target prices for ranking, incorporating valuation elasticity into the selection criteria. Additionally, the institution specifically distinguishes between "factory statistical口径" and "actual end-user delivery," reminding readers to note the unassembled/testing portion included in Wistron's L10 compute tray converted rack counts, avoiding overestimation of immediately available supply.

Methodology notes

• Industry/Industrial Analysis FrameworkVolume-Price Split

  Using rack shipment volume as the core quantitative indicator for the prosperity of the AI server industry chain

  In the hardware contract manufacturing industry, shipment volume is a purer prosperity signal than revenue, eliminating price interference such as exchange rates and product mix. This report directly maps the true rhythm of downstream AI computing power demand through monthly rack quantity tracking.

• Industry/Industrial Analysis FrameworkUpstream-Midstream-Downstream Transmission

  Distinguishing the statistical difference between L10 compute tray shipments and complete rack deliveries

  AI server production involves multiple stages; L10 tray ex-factory does not equal whole-rack delivery. The report notes that statistics include unassembled/testing portions, helping readers understand the time lag from component shipments to terminal usable capacity, avoiding equating intermediate goods shipments with final demand realization.

• Valuation MethodRIM Residual Income Model

  Using the residual income model for baseline scenario valuation of the three ODMs

  The report discloses the use of the residual income model for Quanta, Wistron, and Hon Hai, with core assumptions including cost of equity, mid-term growth rate, and perpetual growth rate. This method focuses on future excess profitability rather than mere asset value, suitable for valuing capital-intensive contract manufacturers with volatile earnings.

Asset mapping & comparison

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

• Wistron (3231.TW)

  Top Pick: Largest upside to target price; steady growth in AI server compute tray shipments; incremental contribution from subsidiary Wiwynn

  Strengths

  Positive MoM growth in AI rack shipments; product mix improvement drives margin expansion potential; faster-than-expected progress in consumer electronics divestiture could further unlock value

  Weaknesses

  MoM decline in notebook and desktop shipments; slower-than-expected consumer electronics divestiture would drag overall performance

  Comparison

  Highest upside to target price among the three ODMs, superior to Hon Hai and Quanta

  Risks

  Consumer electronics divestiture slower than expected; weak notebook demand; AI server penetration slower than expected; intensified competition leading to margin compression

• Hon Hai (2317.TW)

  Second Choice: Largest absolute AI rack shipment volume; strongest Q2 MoM growth momentum

  Strengths

  Leading industry supply share (>40%); Q2 AI rack shipments expected to grow 18% MoM; iPhone and EV businesses provide diversified growth points

  Weaknesses

  Large MoM decline in May rack shipments (-11%); lower-than-expected profit contribution from AI business would impact overall profitability

  Comparison

  Shipment scale far exceeds peers, but upside to target price is secondary to Wistron

  Risks

  iPhone sales below expectations; low profit contribution from AI business; slow EV progress; geopolitical risks affecting foreign capital allocation

• Quanta (2382.TW)

  Third Choice: May rack shipments declined MoM; Q2 total may be slightly revised down, but full-year forecast unchanged

  Strengths

  Full-year rack shipment forecast maintained at approx. 18,700 units; above-expectations demand for notebooks and Apple Watch could provide additional leverage; accelerated AI server penetration benefits long-term growth

  Weaknesses

  May rack shipments decreased MoM; Q2 total may be revised down to approx. 6,700 units; fierce price competition in the large data center sector

  Comparison

  Smallest upside to target price among the three; short-term shipment rhythm weaker than Hon Hai and Wistron

  Risks

  Weak demand for notebooks and Apple Watch; rising labor costs and below-expectations sales pressuring margins; AI server penetration slower than expected; intense price competition in mega datacenters

Key data

• Total May GB200/300 Rack ShipmentsApprox. 7,700 unitsDown approx. 7% MoM
• 2026 Full-Year Rack Shipment Forecast70k-80k unitsUp over 100% YoY (approx. 29k units in 2025)
• Quanta May Rack Shipments1,800-1,900 unitsBelow April's approx. 2,100 units; Q2 total may be revised down to approx. 6,700 units
• Wistron May Rack Equivalent Shipments1,300-1,400 unitsUp approx. 2% MoM; Q2 maintained at approx. 4,200 units estimate
• Hon Hai May Rack ShipmentsApprox. 3,300 unitsDown approx. 11% MoM; Q2 expected to reach approx. 10k units (+18% q/q)
• Quanta May RevenueNT$311.5BMoM -8%, YoY +94%
• Wistron May RevenueNT$29.0BMoM +2%, YoY +39%

Impact & implications

The report considers 2026 to remain a strong year for the downstream rack assembly segment, with full-year shipment YoY doubling-plus growth expectations providing solid support for the performance of related ODM companies. Although May saw a short-term MoM pullback, this is more of a normal fluctuation in production rhythm and does not change the upward trend for the full year. For investors, all three companies maintain Overweight ratings, but based on differences in upside to target prices, Wistron offers relatively better value at the current juncture, followed by Hon Hai and Quanta. Meanwhile, it should be noted that the unfinished assembly portion included in the statistical口径 means that actual deliverable capacity release may lag behind shipment data.

Risks

• AI server penetration speed slower than expected
• Continued weakness in notebook and traditional PC demand
• Intensified price competition in the large data center sector eroding profits
• Margin contraction due to rising labor costs and below-expectations product sales
• Negative impact of geopolitical developments on foreign capital allocation
• iPhone sales and EV business progress below expectations (Hon Hai specific)
• Consumer electronics divestiture progress slower than expected (Wistron specific)

What to watch

• Monthly actual GB200/300 rack shipments and MoM changes for each ODM
• Whether Q2 total rack shipments meet expectations (Quanta ~6.7K, Wistron ~4.2K, Hon Hai ~10K)
• Assembly and testing progress of Wistron L10 compute trays converting to complete racks
• Whether H2 rack shipments can absorb orders delayed from H1
• Trends in profit margins for each company's AI server business