Enfabrica

Enfabrica operates within the semiconductor sector, engineering silicon and software to address input/output (I/O) bottlenecks in AI and accelerated computing infrastructure. The company was established in 2019 by Rochan Sankar and Shrijeet Mukherjee, two veterans of the Silicon Valley technology landscape. Sankar, the CEO, previously led the Data Center Ethernet switching business at Broadcom, managing multiple generations of Tomahawk and Trident chips. His background includes roles in product management and chip architecture, holding a B.A.Sc. in Electrical Engineering and an MBA from the Wharton School. Shrijeet Mukherjee, the Chief Development Officer, has a history of working on pioneering technologies, including what would become GPUs at Silicon Graphics and SmartNICs at Nuova/Cisco, and was also part of Google's network infrastructure group and VP of Engineering at Cumulus Networks.

Enfabrica's primary business is designing and supplying foundational fabric technologies for the AI era. The company targets the market for high-performance data center infrastructure, serving clients who build and operate large-scale AI clusters, such as hyper-scale cloud providers and enterprise data centers. Its business model revolves around the sale of its silicon and software solutions. A significant milestone was the successful close of a $125 million Series B funding round in September 2023 and an oversubscribed $115 million Series C in November 2024, with investors including Nvidia, Cisco Investments, Arm, and Spark Capital. The company has also expanded its global footprint by opening an R&D facility in Hyderabad, India, to scale its engineering team.

The company's core product is the Accelerated Compute Fabric (ACF), a new class of chip designed to streamline data movement. Its first iteration, the Accelerated Compute Fabric Switch (ACF-S), is a single silicon die that integrates the functions of Top-of-Rack network switches, server Network Interface Controllers (NICs), and PCIe switches. This device, also referred to as a "SuperNIC," delivers multi-terabit-per-second data movement between GPUs, CPUs, accelerators, memory, and networking hardware. A key feature is its use of standards-based interfaces like 800 Gigabit Ethernet, PCIe Gen5, and Compute Express Link (CXL), which allows it to be integrated into existing data center architectures without altering the software layers above the device drivers. By collapsing multiple network tiers, the ACF-S reduces latency and the number of device hops, aiming to cut the cost of GPU compute for large language model inferencing by an estimated 50 percent. The ACF SuperNIC chip is scheduled for general availability in the first quarter of 2025.

Keywords: AI networking, accelerated computing, data center infrastructure, silicon, software, high-performance computing, GPU interconnect, Accelerated Compute Fabric, ACF, SuperNIC, PCIe, CXL, network interface controller, data movement, I/O bottlenecks, fabric technology, Rochan Sankar, Shrijeet Mukherjee, semiconductor, LLM inference, memory fabric, scale-out networking, compute infrastructure, data center silicon