Intel, NVIDIA, and the Packaging Race for Optical Compute Platforms
The optical compute race is moving from link demos to platform control, with Intel and NVIDIA competing to own the packaging stack that will define future AI infrastructure.
What is this trend?
Intel and NVIDIA are racing to control the packaging stack that turns optical links into scalable compute platforms, because the winner will shape power, bandwidth, and manufacturability for next-gen AI systems.
- Intel is co-packaging VCSELs, photodetectors, and ASICs for rack-scale optical links.
- NVIDIA is pushing optics onto the interposer, shrinking electrical reach and boosting density.
- The fight is shifting from proving optical links to owning the full optical compute platform.
- Manufacturing scale and thermal integration are becoming as important as link speed.
- Near-term deployments stay pluggables-first, but 2027-2029 is the platform inflection window.
What’s the latest?
Intel and Lumentum pushed the next step in optical packaging with a host-system demo that co-packages 1060-nm VCSEL and photodetector arrays with a host ASIC using fan-out wafer-level packaging and mu
How it developed
- Memory, optics, and power bottlenecks tighten control over AI infrastructure expansion
- Rack-Scale Optics Shifts From Packaging to System Control
Go deeper
Curated long-form picks on this trend — podcasts, videos, and analysis, by vantage.
If you sell into this industry
Intel Outlines Optical Interconnect Vision for AI Chips, Built on 300mm Silicon and Integrated Lasers
Analysis featuring Nicolas Dube on Intel’s optical interconnect vision for AI chips and packaging race.
XenoSpectrum · News
Read →NVIDIA Unveils OOI Optical Interconnect Research to Shorten Electrical Paths and Boost AI Bandwidth
Analysis of NVIDIA’s OOI optical interconnect research and DWDM bandwidth gains for optical compute packaging race.
XenoSpectrum · News
Read →Celestica DS6005: 102.4 Tb/s with Co-Packaged Optics
Analysis of a new switch using co-packaged optics to boost AI-cluster bandwidth and latency in the optical compute race.
igor´sLAB · News
Read →