Enfabrica

Develops AI networking chips and fabrics for scalable datacenters (62 chars)

Updated April 2026

Overview

Website: enfabrica.net
Headquarters: Santa Clara, CA
Segment: Networking Silicon

Product overview

Enfabrica develops high-performance networking silicon like the ACF SuperNIC (ACF-S) chip, a 3.2 Tbps 5nm device for interconnecting GPUs and accelerators in AI datacenters with low-latency Ethernet and PCIe/CXL support. Their EMFASYS AI Memory Fabric System enables memory disaggregation, pooling up to 18TB of shared memory for efficient large language model inference. The company focuses on elastic, resilient fabrics to address compute-memory bandwidth imbalances in AI workloads.

Revenue model

Sells AI networking chips, SuperNICs, memory fabric systems, and software stacks to datacenter operators

Moat

Enfabrica's key competitive moat is its proprietary Accelerated Compute Fabric (ACF) silicon technology, particularly the ACF-S SuperNIC chip, which uniquely integrates 3.2Tbps network throughput with 128 PCIe lanes (including CXL support) to solve AI infrastructure I/O bottlenecks, enabling 50% lower compute costs, 50x memory expansion, and scalable connectivity for up to 524,288 accelerators. This first-mover hardware innovation, backed by Nvidia's acquisition and groundbreaking design for memory tiering/offload in hyperscale AI clusters, creates high barriers via technical superiority and integration stickiness over rivals.

Headwinds

Competing against established networking giants like Broadcom and Nvidia in a capital-intensive hardware market where customer adoption cycles are long and switching costs are high.