NVIDIA’s major optics partners say their co packaged optics programs are progressing as planned, pushing back against recent rumors that broad CPO deployment had slipped by several years.
Lumentum and Coherent both provided updated timelines during recent investor calls. Their comments suggest that early scale out revenue is beginning in the second half of 2026, while larger scale up deployments are expected to follow in 2027 and 2028.
The updates are important for NVIDIA because future AI systems will need faster and more efficient optical links as rack scale architectures grow beyond the practical limits of copper.
CPO timelines remain active
| Company | CPO timeline |
|---|---|
| Coherent scale out revenue | Second half of 2026 |
| Coherent scale up revenue | 2027 |
| Lumentum scale up volume shipments | Second half of 2027 |
| First customer deployments | Expected in 2028 |
| NVIDIA investment in Lumentum | $2 billion |
| NVIDIA investment in Coherent | $2 billion |
| Main use case | Next generation AI data centers |
Lumentum says demand for ultra high power CPO lasers is increasing, while several near packaged optics engagements are also underway.
The company expects scale up volume shipments to begin in the second half of 2027, followed by first customer deployments in 2028.
Coherent provided an earlier revenue milestone, saying scale out CPO revenue tied to rack to rack connectivity should begin during the second half of 2026. Scale up revenue is expected in 2027.
NVIDIA is investing heavily in optics capacity
NVIDIA has made large investments in both suppliers.
The company announced a multiyear strategic partnership with Lumentum earlier this year, including a $2 billion investment intended to support research, manufacturing capacity and expanded US operations.
Coherent also received a $2 billion investment from NVIDIA.
The company recently started work on a new Texas manufacturing facility that will include a six inch indium phosphide production line for volume manufacturing.
Indium phosphide technology is important for optical networking because it enables high speed transmission between chips, servers and data center infrastructure.
Vera Rubin systems will require more optical connectivity
The shift toward CPO becomes more important as NVIDIA’s AI systems grow larger.
The company has outlined Vera Rubin NVL576 systems that connect eight NVLink racks, each containing 72 Rubin Ultra GPUs, into one 576 GPU domain.
At that scale, traditional copper links become less practical across longer distances because of signal loss, power use and bandwidth limitations.
Optical interconnects can move data farther while improving energy efficiency, which makes them increasingly important for large AI factories.
CPO takes that approach further by bringing optical components much closer to the switching silicon.
AMD is moving in the same direction
NVIDIA is not alone in preparing for optical networking.
AMD is also expected to use co packaged optics in future AI servers, including systems based on its Instinct MI500 GPUs.
Those products are expected to compete with NVIDIA’s Rubin Ultra generation in the second half of 2027.
NVIDIA’s later Feynman generation is expected to advance the company’s optical networking strategy again when it arrives in 2028.
The latest comments from Lumentum and Coherent therefore suggest that CPO development has not been pushed far off schedule. Scale out revenue is already expected in late 2026, with larger deployments following through 2027 and 2028 as NVIDIA and other AI chip makers move toward increasingly dense rack scale systems.
Discussion (0)
Be the first to comment.