d-Matrix Acquires GigaIO for Rack Systems
Etched
The strategic point is that d-Matrix is no longer just selling accelerator cards, it is assembling the whole inference machine around them. By buying GigaIO’s data center assets, d-Matrix brought rack design, PCIe fabric, and systems talent in house, which moves it much closer to Etched’s product boundary of shipping a tightly integrated rack instead of a loose component that others must package into a deployable system.
-
The deal added SuperNODE and FabreX, GigaIO’s rack scale building blocks. Those are the pieces that let many accelerators act like one larger inference system, which is what turns a chip vendor into a seller of complete deployed capacity.
-
d-Matrix had already been moving this way before the acquisition. It announced SquadRack in October 2025 as a rack scale inference system built with Broadcom and Arista, then used the acquisition in April 2026 to internalize more of that stack.
-
That narrows the gap with Etched in how customers buy the product. Instead of purchasing silicon and then relying on an OEM or integrator to wire networking, memory fabric, cooling, and software together, buyers can increasingly get a pre assembled inference rack from one vendor.
The next phase is competition over who can deliver the cleanest rack level outcome, not just the fastest chip. As more inference startups absorb interconnect, software, and system packaging, the market will reward vendors that can ship working racks quickly, fit into real data centers, and turn tokens into revenue with less integration work.