Peking University Optical Chip Boosts AI Speed 100-Fold

What Happened

Researchers at Peking University have developed an all-optical interconnect system that links standard electronic chips with specific algorithms, achieving processing speeds up to 100 times faster than conventional electronic approaches, according to a report published July 13, 2026. The system operates using a fraction of the compute power required by current AI hardware architectures.

What the Technology Does

The breakthrough centers on an all-optical interconnect, a method of transferring data using light rather than electrical signals between chips. The Peking University team connected this optical layer to standard electronic chips paired with tailored algorithms, producing a hybrid system that retains compatibility with existing hardware while delivering the speed advantages of optical data transmission. The reported 100-fold speed increase applies to AI inference and processing tasks that depend heavily on data movement between chips, which is a known bottleneck in current AI infrastructure.

The system also uses significantly less power than conventional electronic interconnects, according to the South China Morning Post report. Compute power consumption was described as a fraction of what comparable electronic systems require, though specific wattage figures were not disclosed in available wire reports.

Background

Optical interconnects have been an active area of research globally for several years. The core premise is that photons, which carry data in optical systems, move faster and generate less heat than electrons in copper-based connections. Applying this principle at the chip-to-chip level, rather than only across longer fiber-optic distances, has remained a significant engineering challenge.

Goldman Sachs identified optical networking as a potential trillion-dollar opportunity in AI infrastructure in a recent analysis, underscoring the commercial interest surrounding this class of technology. Several semiconductor and networking companies in the United States, Europe, and Asia have active programs in silicon photonics and optical interconnects, though large-scale deployment in AI data centers remains limited.

The Peking University development comes amid sustained Chinese government investment in domestic semiconductor and AI hardware research, driven in part by U.S. export restrictions on advanced chips including high-end GPUs from Nvidia. Chinese institutions and companies have increasingly directed research efforts toward alternative hardware architectures that reduce dependence on restricted components.

What the Numbers Say

The headline figure from the research is a 100-fold increase in processing speed relative to conventional electronic chip configurations. The system links standard electronic chips, meaning it does not require fully custom optical processors that would be impractical for near-term deployment. Power consumption was characterized as substantially lower than electronic equivalents, though the research report did not specify a precise reduction ratio in available wire coverage.

No peer-reviewed publication details, journal name, or formal release date for the underlying research paper were available in the wire reports at time of publication.

What It Means in Practice

If the performance claims are validated through independent review, the architecture could be relevant to AI data center operators seeking higher throughput without proportional increases in power draw. The use of standard electronic chips as the base layer, augmented by optical interconnects, suggests a pathway to deployment that does not require replacing existing chip fabrication infrastructure entirely.

The research remains at the laboratory stage based on available information. No commercial partnerships, manufacturing agreements, or product timelines were disclosed in the Peking University announcement as reported.

What Comes Next

Independent peer review of the Peking University findings, and subsequent replication efforts by other research institutions, will determine the timeline for any broader scientific or commercial assessment of the technology.