Photon Capital(@PhotonCap):<The Silicon Photonics & CPO Testing Bottleneck, 6 Months Later> Back in March, I wrote about a less-discussed beneficiary of NVIDIA's shift toward silicon photonics: testing. Six months later, that thesis is moving from roadmap to production. $NVDA's Spectrum-X Ethernet Photonics for Vera Rubin is now in production, while the surrounding CPO test ecosystem is rapidly filling out: $LITE / $COHR → lasers & optical components $FORM → optical/electrical wafer probing & KGD $AEHR → wafer-level burn-in & reliability screening $TER → wafer-to-optical-engine-to-CPO automated test $KEYS → photonic characterization & validation Advantest / $VIAV / SENKO → CPO module & final test The important change is that the bottleneck is no longer just building the photonics. It is testing every photonic component before increasingly expensive heterogeneous integration, and then testing the completed CPO module at production throughput. That is exactly why the test content per optical link should continue rising as CPO moves into HVM.

2026.09.17 22:28

Back in March, I wrote about a less-discussed beneficiary of NVIDIA's shift toward silicon photonics: testing. Six months later, that thesis is moving from roadmap to production. $NVDA's Spectrum-X Ethernet Photonics for Vera Rubin is now in production, while the surrounding CPO test ecosystem is rapidly filling out: $LITE / $COHR → lasers & optical components $FORM → optical/electrical wafer probing & KGD $AEHR → wafer-level burn-in & reliability screening $TER → wafer-to-optical-engine-to-CPO automated test $KEYS → photonic characterization & validation Advantest / $VIAV / SENKO → CPO module & final test The important change is that the bottleneck is no longer just building the photonics. It is testing every photonic component before increasingly expensive heterogeneous integration, and then testing the completed CPO module at production throughput. That is exactly why the test content per optical link should continue rising as CPO moves into HVM.