NPO, Riding the Wind

NPO surges as AI clusters abandon copper. 2026 2027 mass deployment begins while CPO faces delays until 2028.

Recently, a bearish research report on CPO from SemiAnalysis has stirred waves in the optical communications market. The report points out that the large-scale commercial deployment of CPO will be delayed until 2028–2029, with key bottlenecks in packaging yield, integration difficulty, and insufficient cost advantages.

Morgan Stanley subsequently echoed this view, forecasting that global optical engine shipments in 2027 will reach only 6–7 million units—far below the market’s expected 20–30 million units. The firm judges that true explosive growth for CPO may begin in 2028, with 2026–2028 serving as a transition period in which pluggable modules, NPO, and copper interconnects will coexist. NPO is precisely the “middle path” between CPO and traditional solutions.

The cross-fire discussion between these two major institutions may have shaken the market’s earlier fantasies of rapid, near-term CPO adoption. Yet behind the divergence, a more fundamental consensus is accelerating: while the full rollout of CPO may require more time, the demand for high-bandwidth optical interconnects in AI clusters remains undiminished.

Against this industrial backdrop and expectation game, NPO has quietly stepped into the center of the stage.

In NVIDIA’s(NVDA 0.00%↑ ) newly announced Rubin Ultra NVL576 design, the usage of NPO optical engines has nearly doubled. The number of 3.2T optical engines per GPU has increased from approximately 2.25 to about 4.0, representing a 78% rise. Meanwhile, Huawei’s Ascend roadmap unveiled at the 2025 Connect Conference explicitly introduces its self-developed Hi-ONE silicon photonics engine into the Ascend 960 super node, delivering 8 Tb/s per module and achieving full optical interconnect within the super node.