Testing, testing, testing.🔥 On Oct 19 $CXMT approved a proposal to put RMB 5.0… · @carrioresearch · TwKit

@carrioresearch

Testing, testing, testing.🔥 On Oct 19 $CXMT approved a proposal to put RMB 5.0B of IPO over-raise money into phase II of its back-end memory wafer test base in Hefei, "to ensure test capacity does not become a bottleneck on the company's output." This is a DRAM maker that just guided nine-month revenue to RMB 265 to 270B, up from RMB 32.1B a year ago. It finished adding wafer capacity and went straight to building its own test plant. Test is now growing faster than the chips it tests. The size of it SEMI's mid-year forecast (July) has test equipment up 31.0% to $15.3B in 2026, against assembly and packaging equipment up 9.6% to $6.7B. It sees test at $20.8B by 2028. Advantest's own tester market numbers: $6.0B in 2024, $9.0B in 2025, $13.0 to 14.5B in 2026. The 2026 figure went up about 19% between its April and July updates. Teradyne CEO Greg Smith in July: "test is now outpacing fab equipment." Three things are stacking up: 1. HBM gets tested several times per stack. DRAM wafers and the base die before stacking, the stacked wafer after, then known good stacks after dicing. Going from 8 to 12 to 16 high, test time keeps stretching. 2. AI accelerator final test is heavier. Advantest said in its July Q&A that accelerators need more test intensity and longer test times, and 3D packaging keeps adding complexity. 3. Optics moving into the package adds new test insertions. Material from Taiwan's National Science and Technology Council in early October puts test at 20 to 30% of the cost of an 800G or 1.6T optical module. Testers (ATE) Advantest (6857 JP): last quarter revenue was JPY 367.5B, up 39.3%. It raised full-year revenue guidance from JPY 1.42T to 1.714T and operating profit from JPY 627.5B to 846.0B. SoC tester share is 66%, target above 70%. Customer forecasts have stretched from 6 months to 18; management's words were "Six months is not enough." The capacity plan of 5,000 units by fiscal year end got pulled forward, and they are now talking about 15,000-plus. $TER: Q2 revenue $1.329B, up 104%, with semiconductor test at $1.122B. Memory test was $212M, 3.5x the $61M a year ago, and memory book-to-bill was above 2. Management called HBM base die test the bright spot in memory. Q3 guide is $1.2 to 1.3B. It launched Magnum E2 for LPDDR6, DDR6 and GDDR7 in late September, and added burn-in to Titan HP on Oct 6. China: Changchuan (SZSE: 300604) grew H1 revenue 59.7% to RMB 3.46B and net profit 125.7% to RMB 964M. Jingzhida (SSE: 688627) signed two tester contracts this year worth RMB 2.89B combined, customer name withheld under a disclosure exemption. Probe cards This is the layer I like most. Every chip needs its own custom card, cards wear out and get rebought, and they sit right on wafer test capacity. $FORM: Q2 revenue $258.2M, up 31.9%. DRAM probe cards were $85.0M, up 49%, and the 10-Q says HBM was roughly 75% of the year-on-year increase. Q3 guide is $270M plus or minus $10M. On Oct 6 Teradyne agreed to buy 2,234,887 new FormFactor shares for $300M, about $134.2 a share, alongside a technology tie-up covering HPC, advanced packaging, HBM and silicon photonics. When the tester leader writes a $300M check for the probe card leader, that tells you where the probe card sits in this chain. Technoprobe (Italy): H1 revenue EUR 464.1M, up 42.4%. It raised 2026 guidance to EUR 1.05 to 1.10B, EBITDA margin to 46 to 48%, and capex from EUR 250M to 350M. MJC (6871 JP): H1 revenue up 48.6%, operating profit up 105.5%. Full-year revenue forecast JPY 103.8B, versus 70.2B last year. Taiwan September revenue: MPI (6223 TT) +82.2%, CHPT (6510 TT) +55.9%, WinWay (6515 TT) +87.7%. Korean outlet TheElec reported in early September that FormFactor and MPI are the only two vendors qualified so far for TSMC's CPO probe stations. Burn-in, handlers, sockets $AEHR: FY27 revenue guide is $130 to 150M, 2.6 to 3.0x FY26, with effective backlog around $100M. On Oct 7 it booked another roughly $6M Sonoma order from a top hyperscaler for that customer's second AI processor, ramping in 2027. Chroma (2360 TT) September revenue NTD 6.86B, up 206.5%. Hon Precision (7769 TT) September up 100%. $COHU: Q3 revenue guide $170M. Its AI compute pipeline was raised to about $850M. Test houses and OSATs KYEC (2449 TT) September revenue NTD 4.07B, up 24.5%, up 34.1% year to date. ASE (3711 TT) September NTD 88.13B, up 45.5%. Test houses are growing slower than tool and probe card makers, and that's how it should look. They get paid on volume; the pricing power sits upstream with the testers and the cards. CXMT building its own test plant fits the same pattern: once a big fab's output gets large enough, it starts pulling test back in-house. Optical test (CPO and silicon photonics) This is the next leg. Teradyne's July number for the CPO test market is $300 to 700M by 2028. FormFactor said 2026 CPO revenue should land at the high end of its $10 to 20M range, and its Systems segment grew 44.1% in Q2. Advantest and FormFactor rolled out a joint CPO test cell in August. The hard part right now is concrete. A full check of a single PIC takes over 100 seconds, and a lot of fiber alignment is still done by hand. Whoever gets that step to production volume collects the rent. Where I come out I'm bullish on the test layer. My ranking: probe cards, then testers, then burn-in, then test houses. $FORM is the cleanest HBM plus CPO test name, and Teradyne's $134 entry price works as an industry buyer's price anchor. $TER grew memory test 3.5x in a year, and I treat its CPO business as an option. When DRAM makers expand, testers and probe cards get ordered first, so these two layers run ahead of memory shipments. $TER $FORM $AEHR $COHU

曝光 538 · 评论 0 · 点赞 4 · 书签 4