South Korea funds AI chip interconnect project targeting 2025 launch
Panmnesia secured government backing to develop switch chips and link controllers for connecting AI accelerators, with products scheduled to ship in the second half of 2025.
South Korea's Ministry of Science and ICT and the Institute of Information & Communications Technology Planning & Evaluation awarded the project as part of the K-Cloud technology development initiative. The funding recognizes Panmnesia's work on open-standard interconnect technologies for AI infrastructure.
What Panmnesia will build
The company plans to develop two core components: link controllers that enable high-speed communication between AI accelerators, and switches that route data between multiple accelerators.
Both will support UALink and Ethernet. UALink is an open standard designed to connect AI accelerators from different vendors at high speeds, positioning it as an alternative to proprietary technologies like NVLink. The standard is being developed under the Open Compute Project with participation from AMD, Amazon, Intel, Meta, and Microsoft.
Panmnesia is also developing optimized topologies to ensure data flows correctly between controllers, switches, and other devices. The company will validate the system at rack-scale deployments before shipping.
Market context for government technology planning
Open-standard interconnects matter for government AI infrastructure decisions. They reduce vendor lock-in and allow agencies to source components from multiple manufacturers, lowering long-term costs and increasing flexibility.
Panmnesia previously secured a government project related to Compute Express Link (CXL), another open interconnect standard. The company's participation in both UALink and CXL development positions it across multiple emerging standards.
Switch chips supporting both protocols are scheduled to begin shipments in the second half of 2025.
Your membership also unlocks:
