When Will Co-Packaged Optics Actually Achieve Mass Production?

Co-packaged optics (CPO) technology is seen as a potential disruptor to traditional optical modules, but the path to its mass production is far longer than market expectations.

Driven by the wave of cloud AI infrastructure upgrades, the commercialization process of CPO technology continues to attract significant market attention. NVIDIA is currently the most aggressive company in advancing CPO mass production, having pressured its suppliers since 2025 to establish production capacity.

However, according to supply chain information, the actual mass production scale of CPO-related products in 2026 remains very limited, and widespread adoption still requires considerable time.

The core bottleneck restricting CPO mass production is yield rate. For CPO to achieve scaled deployment in AI data centers, production yields must reach sufficiently high levels for its cost-effectiveness to surpass existing solutions. This issue is now widely recognized across the industry chain as the primary obstacle.

Yield bottlenecks constrain scaling, with mass production progress falling short of expectations. Although cloud AI manufacturers hope to introduce silicon photonics (SiPh) technology to simultaneously improve cost efficiency and computational performance limits, there remains a significant gap between supply chain reality and market expectations.

From a supply chain perspective, the actual production volume of CPO-related products in 2026 is extremely limited, remaining quite distant from broad implementation. The high difficulty in production and verification processes leaves substantial room for yield improvement. Only when yields reach adequate levels can CPO's cost-effectiveness truly exceed current solutions. This yield challenge is universally identified by industry participants as the biggest bottleneck to规模化 implementation.

CPO supply chain participants note that without the urgent demand from cloud AI upgrades driving massive investment in CPO technology, the timeline for related technological breakthroughs might have been significantly delayed. From this perspective, NVIDIA's proactive push has generated substantial positive ripple effects throughout the ecosystem.

NVIDIA demonstrates the most aggressive stance, while Broadcom and Marvell Technology maintain relatively conservative approaches. Among major chip manufacturers, NVIDIA stands as the most forceful player advancing CPO deployment. Since 2025, the company has continuously urged suppliers to prepare mass production capabilities while emphasizing to customers the significant computational advantages CPO brings to its system architecture, creating considerable pressure on suppliers. Supply chain sources anticipate that NVIDIA's mass production timeline will outpace other manufacturers.

Broadcom has deep expertise in the CPO field and has begun initial shipments to customers. Marvell Technology and MediaTek have also made strategic moves in this technology. However, supply chain information indicates that Broadcom and Marvell Technology are relatively less aggressive in promoting their respective CPO solutions.

CPO-related manufacturers emphasize that regardless of which company achieves the breakthrough first, market demand will continue to grow, benefiting both chip manufacturers and ecosystem partners. At the current stage, the sole factor limiting industry growth remains the production capacity and yield ceiling on the supply side.

Furthermore, from the demand perspective, the continuous expansion of cloud AI infrastructure provides clear and strong market drivers for CPO technology. The silicon photonics ecosystem eagerly awaits substantial revenue contributions, while cloud AI manufacturers have incentives to adopt SiPh technology to break through existing computational power and cost constraints.

However, the critical 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 production ramp-up and yield improvement will directly determine CPO technology's timeline for moving from laboratory research to large-scale data center deployment.