When Will CPO, the "Biggest Rival" to Optical Modules, Truly Reach Mass Production?

Co-packaged Optics (CPO) technology is considered a potential disruptor to traditional optical modules, but its path toward large-scale mass production is far longer than market expectations.

According to a Digitimes report, driven by the wave of cloud AI infrastructure upgrades, the commercialization of CPO technology continues to draw intense market attention. NVIDIA is currently the most proactive company pushing for CPO mass production, continuously pressuring suppliers since 2025 to establish mass production capabilities.

However, according to supply chain information, the actual mass production scale of CPO-related products in 2026 will still be very limited, and large-scale rollout will take time.

The core bottleneck constraining CPO mass production is yield. For CPO to achieve large-scale deployment in AI data centers, production yields must reach a high enough level for cost-effectiveness to surpass existing solutions. This issue is now widely recognized by the entire industry as the largest obstacle.

Yield Bottlenecks Constrain Scaling; Mass Production Progress Falls Short of Expectations

Although cloud AI providers hope to introduce Silicon Photonics (SiPh) technology to simultaneously enhance cost efficiency and computational performance limits, a clear gap remains between supply chain realities and market expectations.

From a supply chain perspective, the actual mass production scale of CPO-related products in 2026 will be extremely limited, remaining a significant distance from widespread adoption. The high difficulty of production and verification processes means there is substantial room for yield improvement; only when yields reach sufficient levels can CPO's cost-effectiveness truly outperform current solutions. This yield issue is currently regarded by all industry stakeholders as the primary bottleneck for large-scale implementation.

CPO supply chain participants pointed out that if not for the urgent demand for cloud AI upgrades driving massive investment in CPO technology, the timeline for related technical breakthroughs might have been significantly delayed. In this regard, NVIDIA's proactive push has exerted a notable positive driving effect on the entire ecosystem.

NVIDIA Most Aggressive; Broadcom and Marvell Relatively Conservative

Among major chipmakers, NVIDIA is the most aggressive player pushing for CPO deployment. Since 2025, NVIDIA has consistently urged suppliers to prepare mass production capacity and emphasized to customers the significant computational advantages CPO brings to its system architecture, placing substantial pressure on suppliers. Supply chain sources expect NVIDIA's pace of mass production advancement to exceed that of other manufacturers.

Broadcom has been deeply involved in the CPO field for many years and has begun preliminary shipments to customers. Marvell Technology and MediaTek also have strategic footprints in this technology. However, according to supply chain sources, Broadcom and Marvell have been relatively less aggressive in promoting their respective CPO solutions.

CPO-related firms emphasized that regardless of which company achieves a breakthrough first, market demand will continue to grow, and both chipmakers and ecosystem partners will benefit. At this stage, the only factor limiting industry growth remains the ceiling on supply-side capacity and yields.

Furthermore, from the demand side, the ongoing expansion of cloud AI infrastructure provides a clear and strong market driver for CPO technology. The Silicon Photonics ecosystem is eager to see material revenue contributions, and cloud AI providers are motivated to adopt SiPh technology to break through the dual constraints of current computing power and costs.

However, the key variable determining whether CPO can truly become the "terminator" of optical modules has shifted from the demand side to the supply side. The speed of capacity ramp-up and yield improvement will directly dictate the timeline for CPO technology to transition from the laboratory to large-scale data center deployment.