D-Matrix’s focus is on AI inference workloads that require fast responses and smooth user interactions, including chatbots, voice agents and AI coding assistants. Although Nvidia remains the dominant name in AI chips, startups such as D-Matrix are attempting to carve out a niche in the rapidly growing market.
Founders behind D-Matrix
D-Matrix was co-founded in 2019 by Sid Sheth, who currently serves as the company’s chief executive officer, and Sudeep Bhoja, its chief technology officer.
Sid Sheth is a veteran semiconductor executive and serial entrepreneur. Before co-founding D-Matrix, he led the Networking Business Unit at Inphi Corporation, where he helped build the business into generating more than $1 billion in revenue.
D-Matrix co-founder and CTO Sudeep Bhoja serves as the chief architect and leads its work on generative AI inference technologies. Before joining D-Matrix, Bhoja served as CTO of the Datacenter Business Unit at Inphi.
Earlier in his career, he held leadership roles within Broadcom’s Infrastructure and Networking Group.
Sheth believes the AI infrastructure sector represents a massive business opportunity and says there is room for more companies to grow alongside established players such as Nvidia.
What makes D-Matrix different?
At the centre of D-Matrix’s strategy is its flagship AI inference chip, Corsair. According to D-Matrix, the chip is built for AI workloads, including chatbots and voice-based applications, with a focus on low-latency performance and energy efficiency.
Unlike conventional GPUs that depend heavily on DRAM memory, Corsair uses an SRAM-based architecture that places memory and compute functions closer together on the chip. According to D-Matrix, this enables faster processing for certain AI workloads while reducing energy requirements.
The company claims that when paired with Nvidia’s Blackwell GPUs, Corsair can significantly improve inference speeds while lowering operating costs and power consumption.
According to the company, research conducted by Gimlet Labs showed that Corsair, when deployed alongside Blackwell GPUs, can run AI inference workloads up to 10 times faster, at one-third the cost and with as much as five times greater energy efficiency than a standalone GPU.
“We’re not running into a chokepoint around DRAM with our product because our product doesn’t really rely on DRAM to be successful,” Sheth was quoted as saying by CNBC.
Backed by Microsoft
D-Matrix has attracted backing from several investors, including Microsoft‘s venture arm M12. Since its launch, the company has raised roughly $500 million and is currently valued at around $2 billion.
The startup has already begun shipping its chips to selected customers, including hyperscalers, neoclouds and frontier AI labs.
According to the company, most of its early customers are based in the United States, while others are located in regions such as the Middle East and Southeast Asia.
Challenges ahead
Despite its ambitions, D-Matrix faces significant hurdles. Rick Bahr, adjunct professor of electrical engineering at Stanford University, said that while SRAM-based chips can deliver impressive speed and efficiency, they may struggle to handle the massive reasoning models.
SRAM-based chips can deliver very fast AI inference because data moves across shorter distances within the chip. However, the architecture may struggle to support the trillions of parameters used in today’s largest AI models from leaders like OpenAI and Anthropic.
“That number of parameters just simply can’t be put onto an SRAM-based design. That’s the big challenge,” Bahr said.
Building an AI hardware ecosystem
Beyond individual chips, D-Matrix is working with companies including Broadcom, Arista Networks and Super Micro Computer to develop larger rack-scale systems called SquadRack for deploying chips in AI data centres.
Its current Corsair chip is manufactured by Taiwan Semiconductor Manufacturing Company (TSMC) using a 6-nanometre node, while a next-generation product called Raptor is expected to launch next year on TSMC 4 nanometer.
“Building a computing solution for AI inference is going to be the grand prize,” Sheth added.
