← Back to all notes
R

Roel Huis in ‘t Veld

Director, Kriya Materials · Week of October 26, 2026

Co-packaged optics went from slideware to shipping in one OCP week

Three things happened at the Open Compute summit last week that tell me co-packaged optics is no longer a roadmap item. Intel said it is the only foundry with high-volume indium phosphide lasers on 300 millimetre silicon photonics wafers, where everybody else is stuck on 150. Nvidia is shipping Spectrum-X switches with the optics on the package, built on TSMC's COUPE process. Broadcom showed a third-generation CPO switch at 102.4 terabits with 200 gigabits per lane. Add Innolight starting 800G CPO modules for 2027 and Huawei's near-package Ascend node in September, and the question has moved from whether to how many. My view is that this is good news for the optical materials chain and a trap for anyone in it who thinks volume means tonnes. Start with the number TrendForce buried at the bottom: aligning the fibre to the chip now takes seconds, but fully testing one photonic chip takes over a hundred. That is the bottleneck, and it is why they call 2027 the year of test integration and 2028 the year volume readiness gets judged. Anyone promising CPO materials revenue in 2026 is selling samples. Now the materials. Every one of these packages has a fibre array, a coupler, a polymer or glass waveguide bridge, and somewhere in there a low-index cladding that keeps light in a core a few microns wide. The index contrast is the whole design. A cladding at 1.2 or 1.3 against a core at 1.5 gives you bend radii and mode confinement that a 1.45 cladding simply cannot. That is a coating formulation problem, and it is one of the few places in a data centre where a European nanoparticle company has something to say. The catch is quantity. A CPO switch uses grams of cladding, not litres. A hundred thousand switches a year is still a few hundred kilograms of material for the whole industry. The economics only work if the qualification is yours, the specification is yours, and the price reflects what the material does rather than what it weighs. I have watched suppliers chase display volumes with coatings priced like paint. This market punishes that harder. The other signal came from Karl Guttag's report on the microLED conference in Eindhoven: company after company now pitches optical interconnects and CPO instead of AR glasses. Read that as the microLED-for-AR money finding a market that will pay this decade. Eindhoven's own Photon Bridge raised ten million for interconnect lasers the same week. When the hardware people pivot, the materials people a year behind them should notice. So where does a small coatings maker fit? In the cladding, the index-matched fills and the anti-reflection on the coupling faces, qualified with one packaging house, not sold to the industry. Everywhere else, the fab owns the layer. CPO is real. It is just not big, and the two are not the same thing.