PDF 原檔：250608_daiwa_ Taiwan Datacentre Hardware_original.pdf
原始內容
3
Taiwan Datacentre Hardware
Rating: Positive
May sales update: revenue strength from Amazon supply chain
Sheng Cheng
(886) 2 8758 6253
sheng.cheng@daiwacm-cathay.com.tw
Allan Wang
(886) 2 8758 6249
allan.wang@daiwacm-cathay.com.tw
Stacy Lin
(886) 2 8758 6252
stacy.lin@daiwacm-cathay.com.tw
Summary: After market hours on 8 June, Wiwynn and GCE released their monthly sales for May; Accton, King Slide, and TUC had already released their May sales last week. Our read-through is that Amazon's supply chain likely outperformed as we saw strong revenue momentum from May sales, especially for King Slide. Based on our supply-chain checks, we believe Wiwynn will start shipping the next generation of Trainium 3 ASIC AI server (or TR3 PDS) from June, which is tracking slightly faster than our original expectations of 3Q26. We believe this to be the key reason for component levels to deliver a strongerthan-expected QTD sales in 2Q26. Looking ahead, we expect the Trn3 ASIC AI server to have c.35-40k units (eg, 64 chips per units) in the following 4 quarters. Based on our current forecast, we believe the rising revenue momentum trend should last at least until the end of this year. As such, we believe it would become key share price catalyst for the AMZN supply-chain players (eg, Accton, GCE, King Slide, TUC, and Wiwynn).
Datacentre hardware sector: May 2026 monthly sales (TWDbn)
| Company | Ticker | Price (TWD) | Rating | May Sales | MoM | YoY | 2Q26 Sales | QoQ | YoY | Achieved% | PER 2026E | (x) 2027E | 3-Yr Range | Release Date | Consensus | Consensus Achieved% |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Datacentre | ||||||||||||||||
| Wiwynn | 6669 TT | 5,275 | Buy | 84.1 | 2% | 18% | 291.3 | 5% | 32% | 57% | 13.7 | 9.9 | 9-23x | 8-Jun | 276.6 | 60% |
| Accton | 2345 TT | 2,420.0 | Buy | 28.6 | 5% | 57% | 82.9 | 18% | 37% | 67% | 36.2 | 26.3 | 14-36x | 5-Jun | 81.1 | 69% |
| GCE | 2368 TT | 1,360.0 | Buy | 8.8 | 20% | 87% | 23.2 | 20% | 67% | 69% | 36.3 | 23.5 | 7-22x | 8-Jun | 23.0 | 70% |
| King Slide | 2059 TT | 6,180.0 | Buy | 3.8 | 46% | 172% | 7.2 | 32% | 70% | 89% | 35.8 | 30.6 | 10-35x | 5-Jun | 6.9 | 93% |
| TUC | 6274 TT | 1,535.0 | Buy | 4.8 | 4% | 129% | 12.6 | 26% | 86% | 75% | 56.5 | 33.8 | 5-25x | 5-Jun | 13.9 | 68% |
Source: Company, Daiwa forecasts. Pricing as of close on 8 June 2026.
· Wiwynn (6669 TT, TWD5,275, Buy [1])
Wiwynn's May revenue was TWD84.1bn (+2% MoM, +18% YoY), leading QTD sales to reach 57% of our 2Q26E revenue (TWD291.3bn; +5% QoQ, +32% YoY) and 60% of the Bloomberg consensus (TWD276.6bn). With April-May revenue typically accounting for 5365% (average 60%) of Wiwynn's 2Q revenue over the past 5 years, we believe the runrate is largely in line with market expectations. We forecast 2Q26 sales to grow only slightly QoQ despite the rising regular server demand as the memory component in the server will not be recognized into Wiwynn's revenue and COGS due to the sharp spike in memory costs. However, we believe the Trn3 ramp to be tracking slightly faster than our original expectations (ie, 2H26) due to the better-than-expected supply-chain progress. Given our current supply chain check, it will start officially ramping from June 2026. We believe this is also the key reason why AMZN supply chain had a strong revenue momentum since 2Q26. We have a Buy (1) rating on Wiwynn, with a 12-month TP of TWD7,100, based on a target PER of 18x applied to our 1-year forward EPS forecast. Key downside risk: worse-than-expected server demand and capex from CSPs.
· Accton (2345 TT, TWD2,420, Buy [1])
Accton's May revenue was TWD28.6bn (+5% MoM, +57% YoY), with QTD sales reaching 67% of our 2Q26 revenue forecast (TWD82.9bn; +18% QoQ, +37% YoY) and 69% of the Bloomberg consensus forecast (TWD81.1bn). While April-May sales have typically accounted for 64-68% (average: 67%) of Accton's 2Q sales over past 5 years, we believe the current progress is in line with market expectations. We remain confident in our original forecast of c.TWD83bn in revenue (ie >15% QoQ) for 2Q26 given the T3 ramp from 2Q26. We believe Accton maintains its revenue momentum as we originally expected with whitebox switches from hyperscalers and the AMZN T3 AI card being key growth drivers. We
8 June 2026
have a Buy (1) rating on Accton, with a 12-month TP of TWD3,222, based on a target PER of 35x applied to our 1-year forward EPS forecast. Key downside risk: worse-thanexpected demand for ASIC AI servers and switches.
· GCE (2368 TT, TWD1,360, Buy [1])
GCE's May revenue was TWD8.8bn (+20% MoM, +87% YoY), once again reaching a record high. QTD sales reached 69% of our 2Q26 revenue forecast (TWD23.2bn; +20% QoQ, +67% YoY) and 70% of the Bloomberg consensus forecast (TWD23.0bn). With AprilMay revenue typically accounting for 66-68% (average: 67%) of GCE's 2Q revenue over the past 5 years, we believe the current progress is slightly ahead of our expectations. Based on our understanding, GCE's monthly output has increased from TWD6.8bn in 1Q26 to TWD7.5bn in May due mainly to additional ramp from Taiwan (+TWD0.2bn), Suzhou (+TWD0.2bn), Chang I (+TWD0.1bn) and Thailand (+TWD0.2bn) plants. Typically, GCE's monthly sales will be 9-10% higher (eg, TWD8.3bn) than its monthly output (TWD7.5bn), and we suspect the strength is due to additional sales from inventory. Having said that, May sales was still slightly higher than TWD8.3bn. We suspect the strength to be mainly driven by solid Trn3 ramp. We have a Buy (1) rating on GCE, with a 12-month TP of TWD1,650, based on a target PER of 35x applied to our 1-year forward EPS forecast. Key downside risk: worse-than-expected demand for AI servers.
· King Slide (2059 TT, TWD6,180, Buy [1])
King Slide's May revenue was TWD3.8bn (+46% MoM, +172% YoY), reaching a record high. QTD sales reached 89% of our 2Q26 revenue forecast (TWD7.2bn; +32% QoQ, +70% YoY) and 93% of the Bloomberg consensus forecast (TWD6.9bn). With April-May revenue typically accounting for 64-68% (average: 67%) of King Slide's 2Q revenue over the past 5 years, we believe the current progress is well ahead of our expectations. We attribute the revenue strength to 1) AMZN T3 rail kit ramp from April, 2) increasing regular server demand, and 3) GB300 demand revised up given VR200 likely would further delay to late of 4Q26 or early-2027. We have a Buy (1) rating on King Slide. Key downside risk: worse-than-expected demand for AI servers.
· TUC (6274 TT, TWD1,535, Buy [1])
TUC's May revenue was TWD4.8bn (+4% MoM, +129% YoY). QTD sales reached 75% of our 2Q26 revenue forecast (TWD12.6bn; +26% QoQ, +86% YoY) and 68% of the Bloomberg consensus forecast (TWD13.9bn). With April-May revenue typically accounting for 65-69% (average: 67%) of TUC's 2Q revenue over the past 5 years, we believe the current progress is ahead of our expectations. Based on our understanding, while GCE's revenue started ramping up for T3 AI server already from 2Q26, TUC's T3 projects should allocate mainly from GCE's Thailand and Taiwan plants. We believe TUC would have more meaningful revenue contribution from 2H26 once GCE's Thailand plant has a more meaningfully contribution (eg, TWD0.6bn now in May vs. TWD1.3bn in 3Q26). Hence, we believe the revenue strength to be from the CCL price hike and product-mix optimization (migration from mid-to-low end to mid-to-high end). We have a Buy (1) rating on TUC, with a 12-month TP of TWD1,680, based on a target PER of 48x applied to our 1-year forward EPS forecast. Key downside risks: 1) weaker-than-expected demand for switches/AI servers; 2) market share loss to CCL peers; 3) raw material shortages.
In the interests of timeliness, this document has not been edited.