報告_Aletheia_致茂2360_20260730

PDF 原檔：報告_Aletheia_致茂2360_20260730_original.pdf

原始內容

Chroma ATE — Buy #Flashnote Aletheia Capital｜Technology Hardware｜Taiwan｜30 July 2026

Angus Lin　+852 9250 7088　angus.lin@aletheia-capital.com

Between the Lines Lies the Upside

• 2Q EPS beat on non-op gains, while input cost-driven GPM softness appears transitory as management expects pricing adjustments to restore 2H margins to structurally high levels. Revenue momentum in 2H is expected to accelerate, driven by AI GPU/CPU SLT for AMD and Google, as well as transceiver and CPO photonics. Contrary to market concern of a delay in CPO adoption, we see upside to Chroma's 2027 CPO revenue, which now could exceed NT$10bn vs the company's prior implied NT$7bn, given that, in addition to OSATs, foundries are now also procuring ins #4o testers to optimize OE yield earlier in the assembly flow. Accordingly, we see upsides to our already Street-high 2027E EPS of NT$82. We are reviewing our model and will provide further updates. We retain Buy with TP NT$4,000.
• Chroma's 2Q26 EPS of NT$12.10 came in ahead of our/consensus estimates of NT$10.52/NT$9.80, driven primarily by higher non-op income from the disposal gain of Adlink shares and staff housing. GPM was a slight disappointment, declining by 2ppts QoQ due to input cost inflation. That, however, appears temporary, given management expressed confidence in its ability to pass through higher material costs via ASP hikes in 2H while maintaining a structurally high GPM profile, pointing to a GPM recovery in 2H26, hence we expect GPM to rebound to roughly the 1Q26 level of 63% or higher.
• 2H revenue momentum is expected to accelerate, driven by the ramp of SLT demand for AMD's MI450 and Google's CPU. We believe the combined revenue exceeds that of Nvidia Blackwell SLT, implying a backend loaded year — in contrast to the historical pattern, where 1H typically represented the peak SLT shipment period due to Chroma's heavy reliance on Nvidia in the past. Today, however, Chroma's high-site-density SLT has broadened well beyond to include AMD AI GPUs, Google CPUs, and networking ICs, while management indicated that it retains an unassailable competitive position in high-power (≥2.5kW) SLT. Beyond SLT, photonics is another key growth engine in 2H, driven by optical transceiver burn-in testers and the commercialization of CPO ins #3 and #4o testers. The company also expects HVDC power testing and ESS demand to provide additional growth tailwinds. Notably, AI server power testing is 60% of ATS revenue in 1H26 vs 50% in 2025. Management emphasized that both ±400V and 800V power architectures utilize the same power testing modules, regardless of the underlying architecture. Our checks suggest that demand for ±400V/800V power testers could be multiple times that of conventional power architecture.
• Chroma will begin recognizing revenue from its CPO testers in 2H26, including its insertion #3 tester ("58721 light engine O/E test") and insertion #4o tester ("587111 CPO switch module test"). This is very positive given that our checks suggest CPO ins #3 tester ASP is well above US$2mn. Management guided CPO revenue to exceed NT$1bn in 2H26 and to increase sevenfold in 2027. The key incremental takeaway from the conference, however, was that foundries are now also procuring the Insertion #4o tester, rather than demand being limited to OSATs. This reflects increasing adoption of optical engine-level yield optimization earlier in the manufacturing flow and, in our view, represents a meaningful upside to the company's initial expectations. We therefore believe the CPO revenue opportunity for Chroma could exceed NT$10bn next year, well above the NT$7bn opportunity previously implied by the company for 2027. In transceiver photonics, Chroma rolled out a new burn-in system ("585604-HS LD") offering 2,688 channels per system. Management described customer demand as "tremendous," reinforcing the view that transceiver photonics revenue will more than double this year.

圖片清單（已驗證 2026-07-31）

ale 2360_001.png（19KB）未達 40KB 門檻，未 Read，預設 logo。頁 2（研究團隊名單頁）與頁 3（免責聲明頁）字型正常、無資料圖表；頁 1（正文分析頁）本身無內嵌圖表，僅文字段落。本份報告無真資料圖可嵌入 lib 頁。