$ASX $NVDA $LITE $COHR $AXTI $MRVL Co-packaged optics is already in volume pro… · @carrioresearch · TwKit

@carrioresearch

$ASX $NVDA $LITE $COHR $AXTI $MRVL Co-packaged optics is already in volume production at the world's biggest chip packaging house. That came out on ASE's third-quarter call today, and it moves the CPO timeline that a lot of people have been arguing about. CFO Joseph Tung said it plainly: "We already started mass production in quarter four. And that would continue into 2027." He put CPO revenue at more than half a billion dollars next year, and he said that number three separate times during the Q&A, so it isn't a slip. Asked what role ASE plays, he wouldn't name products or customers, but said ASE is "the most active participant" in the whole CPO effort, working together with its foundry partner and its customers. Asked whether CPO and near-package designs will fight each other, he said the two will coexist, and if the demand is there, "we will be the major player in that" too. CPO sits inside ASE's leading-edge advanced packaging and test business, which it calls LEAP. The rest of what Tung said about LEAP tells you how big the pipe behind CPO is getting: - Full-process advanced packaging, where ASE runs the whole flow itself, should bring in more than $1 billion next year. Tung called that number conservative and set by capacity - LEAP will be reported in four pieces from now on: on-substrate, full process, wafer sort and CPO. CPO growth will be visible every quarter - Computing is now 34% of ASE's packaging and test revenue, up from 25% a year ago - ASE's EMS arm, the part that builds modules and boards, "will continue to move into optical silicon photonics" and power devices - 2026 capex was raised again, from US$10.5 billion to US$11.5 billion, and Tung said next year's will be "even larger." Half of the new billion goes to buildings - Gross margin on packaging and test hit 30.6%. "The so-called structural margin ceiling has now become the base for us," he said, and he expects it to rise every quarter next year The constraint he kept naming wasn't customers or technology. It was floor space. Demand "can be way higher than what we'll be able to deliver," and "the biggest bottleneck today is really still the space." ASE is buying existing plants where it can and says there aren't many left to buy. When an analyst asked about TSMC's own subsidiaries adding on-substrate capacity, Tung said that competition comes "out of necessity," it won't move market share much, and given the scale gap he doesn't expect a price war. Pricing, in his words, is "very friendly" across both advanced and mainstream packaging. Here's why this matters beyond ASE. A CPO switch moves the optical engine into the package, but the light still has to come from somewhere. In NVIDIA's Quantum-X and Spectrum-X photonics switches, the lasers sit in separate external laser source modules, eight lasers to a module, fed to the engines through fiber. A Quantum-X switch uses 18 of those modules. So every CPO engine ASE packages pulls a set of lasers along with it, and the lasers sit in modules that can be swapped out on their own. The lasers on this chain are mostly indium phosphide. $LITE and $COHR build them, and $AXTI makes the InP substrate they grow on. That layer is already tight before CPO has even ramped. Lumentum's CEO said on Oct. 27 that the company is "basically sold out through 2029," and on Wednesday Bloomberg reported that Marvell is in advanced talks to buy Lumentum. AXT reports after the US close on Oct. 29. A year ago CPO was a roadmap slide. This quarter it's running on a packaging line, and the packager is building capacity for more than $500 million of it in 2027 while saying it still can't keep up. I expect that demand to pass straight through to laser orders across 2027, and the whole chain from substrate to laser to package should keep moving up together.

曝光 924 · 评论 2 · 点赞 7 · 书签 2