Recently, at the first AI optical computing power industry forum hosted by Xizhi Technology, Tencent Cloud revealed for the first time a new deployment plan for domestic computing power and optical interconnection: the company will deploy domestic computing power on a large scale next year and the following year, and is expected to deploy NPO supernodes in the fourth quarter of 2026, while also calling for the unification of domestic and foreign NPO industry standards. This move has once again drawn the market's attention to NPO technology routes. Currently, large-scale deployment of AI computing power places higher demands on interconnection within data centers. Traditional copper interconnections face bottlenecks in terms of bandwidth density and power consumption, and optical interconnection has become a key path to break through this bottleneck. By integrating the optical engine and computing chip on the same board, NPO technology shortens the optoelectronic conversion distance and removes the most expensive DSP chip in the optical module, thereby greatly reducing power consumption and latency. Compared to CPO, NPO projects are less difficult to adapt, and there is less disruption to the existing industrial chain. Wang Yachen, vice president of Tencent Cloud, said on WAIC: “As the scale of supernodes expands, optical interconnection has become the key to breaking through bandwidth and power consumption bottlenecks. NPO is currently a 'more practical technical route' for domestic GPUs to build supernodes.” Shen Yichen, founder of Xizhi Technology, also judged at the meeting that NPO is currently “the easiest first step to achieve and has the least ecological disturbance to the entire industrial chain”. The core reason is that after the DSP was removed by NPO, although the performance requirements for the optical engine were greatly improved, the signal noise reduction caused by the shortened copper wire distance can partially make up for the loss of DSP. This makes the NPO solution advantageous in terms of power consumption and cost. In fact, from optical engine design to advanced packaging technology, from test equipment to upper level scheduling software, the maturity of each link will directly affect the actual efficiency of the NPO supernode. Wang Yachen also thanked industry partners, including Xizhi Technology, at the conference for supporting the construction of a localized NPO supernode. As an optoelectronic hybrid computing power supplier, Xizhi Technology has a comprehensive layout at both NPO and CPO: in the NPO field, it has formed application scenario cooperation with switching chip and GPU manufacturers, while in the CPO field, it has carried out mutual technical verification with switching chips and GPUs, highlighting the forward-looking nature of the scale-up optical interconnection industry layout. “Currently, what needs to be strengthened most urgently is to build a truly open industrial ecosystem in the huge computing power industry chain, so that outstanding enterprises in various fields such as computing, storage, exchange, software, and applications can maximize their respective advantages and jointly drive the upgrading of China's computing power infrastructure.” Shen Yichen said.

Recently, at the first AI optical computing power industry forum hosted by Xizhi Technology, Tencent Cloud revealed for the first time a new deployment plan for domestic computing power and optical interconnection: the company will deploy domestic computing power on a large scale next year and the following year, and is expected to deploy NPO supernodes in the fourth quarter of 2026, while also calling for the unification of domestic and foreign NPO industry standards. This move has once again drawn the market's attention to NPO technology routes. Currently, large-scale deployment of AI computing power places higher demands on interconnection within data centers. Traditional copper interconnections face bottlenecks in terms of bandwidth density and power consumption, and optical interconnection has become a key path to break through this bottleneck. By integrating the optical engine and computing chip on the same board, NPO technology shortens the optoelectronic conversion distance and removes the most expensive DSP chip in the optical module, thereby greatly reducing power consumption and latency. Compared to CPO, NPO projects are less difficult to adapt, and there is less disruption to the existing industrial chain. Wang Yachen, vice president of Tencent Cloud, said on WAIC: “As the scale of supernodes expands, optical interconnection has become the key to breaking through bandwidth and power consumption bottlenecks. NPO is currently a 'more practical technical route' for domestic GPUs to build supernodes.” Shen Yichen, founder of Xizhi Technology, also judged at the meeting that NPO is currently “the easiest first step to achieve and has the least ecological disturbance to the entire industrial chain”. The core reason is that after the DSP was removed by NPO, although the performance requirements for the optical engine were greatly improved, the signal noise reduction caused by the shortened copper wire distance can partially make up for the loss of DSP. This makes the NPO solution advantageous in terms of power consumption and cost. In fact, from optical engine design to advanced packaging technology, from test equipment to upper level scheduling software, the maturity of each link will directly affect the actual efficiency of the NPO supernode. Wang Yachen also thanked industry partners, including Xizhi Technology, at the conference for supporting the construction of a localized NPO supernode. As an optoelectronic hybrid computing power supplier, Xizhi Technology has a comprehensive layout at both NPO and CPO: in the NPO field, it has formed application scenario cooperation with switching chip and GPU manufacturers, while in the CPO field, it has carried out mutual technical verification with switching chips and GPUs, highlighting the forward-looking nature of the scale-up optical interconnection industry layout. “Currently, what needs to be strengthened most urgently is to build a truly open industrial ecosystem in the huge computing power industry chain, so that outstanding enterprises in various fields such as computing, storage, exchange, software, and applications can maximize their respective advantages and jointly drive the upgrading of China's computing power infrastructure.” Shen Yichen said.