Credo Targets AI's Memory Wall With Open Chiplet Interconnect Standard - Yahoo Finance

As AI workloads grow more demanding, data centers are increasingly confronted with the 'memory wall'—a growing disparity between processor performance and memory access speed. Credo Technology is addr... Read original →

Credo's OmniConnect initiative is fundamentally reshaping AI infrastructure interconnects, leveraging AXI-over-VSR SerDes to tackle the memory wall bottleneck that traditional HBM architectures cannot address. This technology enables memory disaggregation, delivering up to 25x density gains and 5% higher bandwidth than HBM4, accelerating the adoption of chiplet-based designs in data centers. The move strengthens Credo’s position within the Open Compute Project ecosystem, pressuring competitors like NVIDIA and Astera Labs to fast-track proprietary high-speed connectivity solutions. From a supply chain perspective, foundries such as TSMC must advance 3nm and below processes to support these high-frequency signals, while geopolitical tensions involving Taiwan, China, and the U.S. may disrupt global chip delivery timelines. In the short term, the industry will see intensified competition between open and closed ecosystems, with long-term implications for modular, scalable AI computing platforms.