เกิดอะไรขึ้น
Cornelis announced a $205 million funding round and the launch of Active Fabric, an open networking architecture that adds programmable computing capabilities to AI and HPC networks. The company is collaborating with Qualcomm Technologies on rack-scale AI data center designs. The CN5000 platform is currently shipping, while the CN6000 is sampling with customers ahead of expanded availability in Q4 2026.
Cornelis has secured $205 million in financing to support its entry into scale-up networking and the development of its next-generation product roadmap. The funds will also increase production capacity and support customer deployments across commercial, academic, government, and cloud environments.
The company unveiled Active Fabric, an architecture designed to make networking an active part of the compute system rather than just a data transport mechanism. It combines lossless transport, in-fabric acceleration, and programmable compute to perform operations on data while it moves through the system.
Cornelis is collaborating with Qualcomm Technologies to examine how networking architecture can improve utilization in rack-scale AI data centers. The architecture is built around open standards, including UALink and ESUN for scale-up networking and Ultra Ethernet specifications for scale-out infrastructure, to preserve customer choice across different accelerator platforms.
The CN5000 platform is currently shipping, while the CN6000 is sampling with customers. Expanded availability for the CN6000 is expected during the fourth quarter of 2026. Cornelis estimates that in a 100,000-GPU system, roughly half of GPU hours can be spent waiting for data, representing approximately $1.68 billion in wasted annual capacity based on pre-production modeling.
ทำไมมันถึงสำคัญ
This development addresses a critical bottleneck in large-scale AI infrastructure where accelerators often idle while waiting for data synchronization. By moving computation into the network fabric, Cornelis aims to reduce wasted GPU hours and power consumption. The use of open standards like UALink and Ultra Ethernet ensures vendor neutrality, potentially reshaping the economics of AI data centers by improving the return on expensive hardware investments.
As AI clusters scale, accelerators often spend significant time waiting for data or synchronization, leading to wasted power and capacity. Cornelis’ approach aims to mitigate this by offloading collective operations and performing work inside the fabric, thereby improving the utilization of expensive hardware.
The reliance on open standards is significant for the industry as it prevents vendor lock-in, allowing data center operators to mix and match accelerators from different providers. This could lower barriers to entry for building large-scale AI infrastructure and increase competition among hardware vendors.
The collaboration with Qualcomm Technologies signals a shift toward more integrated rack-scale solutions where networking, , and memory are tightly coupled. This alignment with major chipmakers suggests that in-fabric compute may become a standard in next-generation AI data center designs.
กลไกเชิงโต้ตอบ: มันทำงานอย่างไร
สำรวจเทคโนโลยีเบื้องหลังการพัฒนานี้แบบโต้ตอบ
What is the best response when AI Models Explained makes a mistake in production?
จะดูอะไรต่อไป.
Monitor the Q4 2026 availability of the CN6000 platform and independent benchmarks validating the claimed reduction in GPU idle time. Watch for further integration details with Qualcomm’s rack-scale designs and whether other major AI infrastructure vendors adopt similar in-fabric approaches.
The transition from sampling to general availability of the CN6000 platform in Q4 2026 will be a key milestone. Independent verification of the performance claims, particularly regarding the reduction in GPU idle time, will be necessary to confirm the practical impact of the Active Fabric.
The adoption of UALink and Ultra Ethernet standards by other networking and vendors will determine the long-term viability and market share of Cornelis’ open architecture approach. Watch for announcements from other major AI infrastructure players regarding their own in-fabric compute strategies.