Nvidia is buying GPU cluster interconnector and server memory offload appliance developer Enfabrica in a deal valued at more than $900 million. Multiple sources, such as CNBC and Reuters, are reporting this deal, with Nvidia acquiring Enfabrica, its team, and its intellectual property in a cash and stock deal.
Enfabrica was founded in 2019 – officially incorporating in 2020 – by CEO Rochan Sankar and chief development officer Shrijeet Mukherjee to develop a massively distributed, Ethernet-based, server fabric interconnect ASIC and associated software.
Its Accelerated Compute Fabric Switch (ACF-S) chip provides a multi-terabit and elastic compute fabric to interconnect hyperscalers’ scaled-out and hyper-distributed servers running AI and machine learning applications, and combine PCIe and CXL technology with RDMA-NIC functions. PCIe and CXL use memory semantics whereas Ethernet and RDMA links use a networking protocol, and Enfabrica bridges the two domains.
Enfabrica has raised a total of $290 million in VC funding – a $50 million A-round in 2022 with a $50 million valuation, $125 million B-round in 2023 with a $250 million valuation, and a $115 million C-round in 2024, with its valuation now around $600 million, at least according to Pitchbook. Nvidia invested in the B-round.
The EMFASYS (Elastic Memory Fabric System) chassis, built around the ACF-S chip, was announced by Enfabrica in July, providing elastic memory bandwidth and capacity to interconnect more than 100,000 GPUs and their limited high-bandwidth memory (HBM) capacity.
NVLink is Nvidia’s high-bandwidth, GPU-to-GPU interconnect networking technology, used within GPU servers and racks. Enfabrica’s IP can add a rack-scale disaggregated memory fabric. We could envisage a future ACF-S development adding NVLink Fusion IO chipsets to further speed system interconnectivity.
Nvidia has just invested $5 billion to buy a stake in Intel and $700 million in Nscale, a datacenter startup building a UK Stargate center with OpenAI and Nvidia. It bought Israel-based Run:ai and its GPU orchestration software for $700 million in 2024. Nvidia is also involved with Kioxia in developing a 100 million IOPS AI SSD with CXL features.
