d-Matrix announced the acquisition of the data center business of GigaIO, adding rack-scale systems expertise, PCIe-based fabric technology, and engineering talent to accelerate deployment of low-latency AI inference infrastructure. The deal centers on GigaIO’s SuperNODE platform and FabreX PCIe Gen5-based memory fabric, extending d-Matrix’s capabilities beyond silicon into full system-level architecture for AI inference at scale.
The acquisition builds on a 2025 collaboration that integrated d-Matrix’s Corsair inference accelerators into GigaIO’s rack-scale systems. With the addition of FabreX and SuperNODE, d-Matrix aims to optimize data movement across disaggregated compute environments spanning CPUs, GPUs, and dedicated inference accelerators. The combined portfolio now includes Corsair accelerators, JetStream interconnects, Aviator software, and the SquadRack rack-scale reference architecture developed with ecosystem partners including Broadcom and Arista Networks.
As part of the transaction, d-Matrix also acquires a team of systems engineers based in Carlsbad, California, establishing a new regional R&D hub. GigaIO will continue as an independent company, pivoting toward edge AI infrastructure with its Gryf portable AI supercomputer platform, targeting use cases in defense, energy, and media that require localized, high-performance compute.
- d-Matrix acquires GigaIO’s SuperNODE platform and FabreX PCIe Gen5 memory fabric
- Expands from chip-level inference to rack-scale and system-level AI infrastructure
- Strengthens portfolio spanning Corsair accelerators, JetStream interconnect, Aviator software, and SquadRack architecture
- Adds engineering team in Southern California to support system deployment
- GigaIO shifts focus to edge AI with its Gryf portable compute platform
“Inference is bigger than any one chip. It’s now a systems problem,” said Sid Sheth, CEO of d-Matrix. “This acquisition accelerates our ability to deliver infrastructure built for this new reality, where low latency, efficiency, and scale all matter at once.”
🌐 Analysis
d-Matrix is a Santa Clara, California–based AI semiconductor company founded in 2019 by Sid Sheth and Sudeep Bhoja, both former Intel executives, focused on low-latency, high-efficiency generative AI inference for data centers. The company has developed a digital in-memory compute (DIMC) architecture designed to optimize throughput, latency, and power efficiency for inference workloads, in contrast to GPU architectures primarily tuned for training. Its platform includes Corsair inference accelerators, the JetStream interconnect, and the Aviator software stack, and is expanding toward rack-scale deployments through partnerships with Broadcom and Arista Networks. The company has raised approximately $450 million in total funding, including a $275 million Series C announced in November 2025, with backing from investors including Temasek, Microsoft (via M12), and Qatar Investment Authority.
GigaIO, founded in 2017 and known for its FabreX memory fabric, has focused on composable infrastructure using PCIe as a high-performance interconnect alternative to traditional networking fabrics. Its SuperNODE architecture demonstrated dense accelerator aggregation within a single system image, while FabreX enabled dynamic resource pooling across CPUs, GPUs, and storage. The divestiture allows GigaIO to concentrate on edge deployments with its Gryf platform, reflecting growing demand for localized AI processing where latency, bandwidth constraints, or data sovereignty limit cloud reliance. The transaction highlights a bifurcation in AI infrastructure: centralized, rack-scale inference systems in data centers versus compact, high-performance edge deployments.
🌐 We’re tracking the latest developments in AI infrastructure and data center architecture. Follow our ongoing coverage at: https://convergedigest.com/category/ai-infrastructure/
