d-Matrix Brand Guidelines: The Santa Clara Inference Chip Maker Behind Corsair

Two interconnect engineers and a bet on inference.

d-Matrix was founded in 2019 by Sid Sheth and Sudeep Bhoja, who remain its chief executive and chief technology officer, according to the leadership page on the company’s website. eeNews Europe reported in June 2023 that both had worked at Broadcom before joining Inphi, the data-centre interconnect chipmaker later bought by Marvell Technology, and that both had backgrounds in power-efficient compute and interconnect for data centres. The company is based in Santa Clara, California.

The founding bet was on inference, the work of running trained AI models, rather than on training them. Sheth told eeNews Europe in 2023 that when the company started, much of the industry’s attention was on training, while the founders expected AI to be “all about inference” within five years; they chose not to pursue high-performance computing, analytics or graphics. The Next Platform reported in September 2026 that the company had concentrated on inference since its founding.

The early years brought changes of direction. In a June 2024 essay on the company blog, Sheth wrote that the first design used an analog in-memory compute approach before the team switched to a digital one, built from SRAM memory cells combined with adders in a custom circuit fabric. A proof-of-concept chip called Nighthawk was shown to investors in 2021 at the company’s Cupertino labs, he wrote, and a later test chip was called Jayhawk. The Series A round, reported in April 2022, included Playground Global, Microsoft’s M12 venture fund and the memory maker SK Hynix, eeNews Europe said; a Series B followed in 2023, according to Sheth’s essay.

From one card to whole racks.

The first commercial product is Corsair. eeNews Europe wrote in June 2023 that reports in April 2022 had expected it in the second half of 2023. Sheth told the publication that Corsair was being “augmented” to support transformer and generative AI models, and that the redesign had delayed development; he then expected 6-nanometre silicon back from TSMC in the first half of 2024. He described the chiplet as about 60 percent SRAM, with most of the rest given to interconnect, so that more of a model can sit on a single card.

d-Matrix unveiled Corsair on 19 November 2024 at the SC24 supercomputing conference in Atlanta. eeNews Europe’s report the same day called it the company’s first AI processor, built from Nighthawk and Jayhawk II tiles made in a 6-nanometre process, with each Nighthawk tile containing four neural cores and a RISC-V CPU. It uses what the company calls digital in-memory compute and supports block floating-point number formats. Corsair comes as a full-height, full-length PCIe Gen5 card, with pairs of cards linked by DMX Bridge cards, and runs the company’s Aviator software stack.

The range has since grown from a card into whole systems. A November 2025 funding announcement named JetStream network interface cards alongside Corsair and Aviator, and a reference rack called SquadRack, developed with Arista, Broadcom and Supermicro. On 12 October 2026, at the OCP Global Summit in San Jose, the company announced Squadron, multi-rack clusters that it said were already running in its own Arizona data centre for research and development and were shipping to priority customers. Sheth described the move as the company’s “evolution from a specialized chip company to a data center compute infrastructure business.”

How AI chip companies present themselves.

AI chip companies sell something most of their customers will never see. The processor sits inside a server, the server sits in a rack, and the rack sits in a data hall far from the people who use the models running on it. So the industry’s visual language has grown up around making that hidden hardware visible: a whole silicon wafer, as wide as a dinner plate, held up on a keynote stage; a single chip pinched between finger and thumb for the cameras; a die shot recoloured in neon false colour so that its blocks of memory and compute read like a city map at night.

The second set of images is the data centre. Press photographs tend to show long rows of racks in dark halls, lit by the glow of status lights and cold aisles, with cabling arranged as neatly as a diagram. These pictures stand in for scale, which is what buyers of AI compute are being asked to imagine, and they appear again and again in launch announcements, partner news and investor material.

Keynote slides do much of the remaining work. Model names, token counts and benchmark charts appear beside the chip’s own product name, and the companies’ identities are usually kept spare to let those numbers and images lead: black or white grounds, a single accent colour, a sans-serif wordmark and very little ornament. Several firms have adopted coined product categories, names for their processors that set them apart from the graphics processor.

The shop window is the technical conference. Hot Chips, NeurIPS and the AI Infra Summit are among the events where architectures are presented, demonstrations run and booths compete for engineers’ attention. This guide looks at how each company in the group presents itself, and keeps to what can be checked about who founded it, who owns it and what it makes.

Private, venture-backed and now working with Nvidia.

d-Matrix is privately held and backed by venture investors. Its Series C round, closed in November 2025, was co-led by BullhoundCapital, Triatomic Capital and Temasek, with the Qatar Investment Authority and EDBI joining and M12 returning, according to the company’s announcement as published by StorageNewsletter on 14 November 2025. Representatives of Playground Global, Triatomic Capital, M12, Nautilus Venture Partners, Temasek and Bullhound Capital sit on its board, the company’s website says. Beyond its Silicon Valley headquarters, Sheth wrote in 2024, the company has design centres in Bangalore, Sydney, Seattle and Toronto.

On 10 September 2026 the company announced a multi-year collaboration with Nvidia: its next-generation inference processors, starting with one called Raptor, are to be built into Nvidia’s MGX rack reference design using NVLink Fusion, Nvidia’s chip-to-chip link technology. The Next Platform, reporting the same day, described Corsair being paired in a rack with Nvidia Blackwell GPUs, with the GPUs handling the compute-heavy first phase of a request and Corsair the phase in which the model generates its answer.

The name is written with a lower-case “d,” a hyphen and a capital “M,” and the company keeps that spelling across its website and press releases, including at the start of a sentence. Its tagline is “Transforming AI from unsustainable to attainable.” On its About page it describes its team as “industry veterans” and invites “the round pegs in square holes” to join. Its product names are short and single-word: Corsair, JetStream, Squadron and Aviator.

For a design guide, d-Matrix shows how an inference-only chip company keeps one idea, memory and compute side by side, at the centre of its story as it grows from a single card into racks and clusters. Compare Groq, another company building chips for inference, and Nvidia, whose rack design d-Matrix is joining. For this guide we use a deep indigo, a violet and a pale grey with a sans, an unofficial interpretation; no product or logo is reproduced.

Keep d-Matrix’s identity on hand — save it to a board in the Made Good inspiration library.