D-Matrix, a startup backed by Microsoft, has launched a new AI chip called Corsair, which claims to execute inference workloads ten times faster while using five times less energy compared to Nvidia’s standalone GPU. Located near Nvidia’s headquarters, D-Matrix aims to carve out a niche in the competitive AI chip market, joining other emerging players like Cerebras, which recently had a successful IPO. D-Matrix has raised about $500 million to date, valuing the company at approximately $2 billion, and plans to start shipping its advanced chips to high-profile customers this month, leveraging a novel approach that integrates memory and computing efficiently without relying on DRAM, addressing current shortages of that memory type.

Groq: Groq was an AI hardware company that designed deterministic, low-latency tensor streaming processors for inference workloads. In the article, Groq appears as a precedent for the space, noting that Nvidia acquired its assets and introduced a Groq-branded language processing unit, showing how novel AI chip startups can become strategic acquisitions for incumbents.
TSMC: Taiwan Semiconductor Manufacturing Company (TSMC) is the world’s largest dedicated contract chip manufacturer, fabricating advanced semiconductors for global designers including Nvidia and many AI hardware startups. In this piece, TSMC is mentioned as the foundry producing D-Matrix’s Corsair chip on its 6-nanometer process and targeted for the next-generation Raptor chip on a 4-nanometer node, underscoring D-Matrix’s reliance on cutting-edge manufacturing.
Arista: Arista Networks is a networking company specializing in high-performance switches and software for large-scale data centers and cloud environments. In the article, Arista is identified as one of D-Matrix’s hardware partners helping build the SquadRack rack-scale system, which combines Corsair accelerators with data center networking for AI inference deployments.
Micron: Micron Technology is a major memory manufacturer producing DRAM and NAND components widely used in PCs, servers, and AI accelerators. In the story, Micron appears as one of the DRAM suppliers whose products are in short supply for GPU-based systems, a constraint that D-Matrix claims to sidestep by relying primarily on SRAM on its Corsair chips.
Nvidia: Nvidia is a leading semiconductor company and the dominant provider of GPUs and systems for training and inference in modern AI, widely used across cloud providers and frontier AI labs. In this article, it is the incumbent that D-Matrix is challenging, with D-Matrix claiming its Corsair inference chip can outperform standalone Nvidia GPUs for certain smaller workloads, while Nvidia continues to emphasize fully integrated, end-to-end AI systems such as its Blackwell-based platforms.
OpenAI: OpenAI is an AI research and deployment company known for developing large-scale generative models and providing them via APIs and products such as advanced language and multimodal systems. The article references OpenAI as a creator of massive models with trillions of parameters to illustrate the scale that SRAM-based architectures like D-Matrix’s struggle to support compared with other designs.
Samsung: Samsung Electronics is a global technology company and a leading supplier of DRAM and other memory chips used in data centers and AI hardware. The article mentions Samsung among the DRAM vendors facing tight supply, highlighting that D-Matrix’s SRAM-centric design avoids the same DRAM bottlenecks that affect many GPU deployments.
Broadcom: Broadcom is a diversified semiconductor and infrastructure software company that supplies networking, storage, and custom silicon solutions used in large data centers and cloud platforms. In this news item, Broadcom is listed as a partner working with D-Matrix on its SquadRack rack-scale AI inference system, indicating integration with established data center and networking components.
Cerebras: Cerebras Systems is an AI hardware company known for its wafer-scale engine (WSE) chips and systems that use a massive single-wafer processor coupled with external memory to support very large neural network models. In the news, Cerebras is cited as a reference point for novel AI accelerators, having recently gone public, with D-Matrix adopting a different SRAM-centric on-chip memory approach as it seeks its own niche in the AI chip market.
D-Matrix: D-Matrix is a Silicon Valley AI chip startup focused on memory-centric architectures that integrate compute and SRAM to accelerate generative AI inference with low latency and power usage. In this news, it is presented as a Microsoft-backed challenger to Nvidia, launching its Corsair inference chip and positioning itself to carve out a niche in the fast-growing AI accelerator market alongside players like Cerebras and Groq.
SK Hynix: SK Hynix is a major South Korean semiconductor manufacturer specializing in DRAM and NAND flash memory for servers, PCs, and mobile devices. In this news context, SK Hynix is listed with other DRAM suppliers whose constrained output contributes to GPU memory shortages, which D-Matrix argues its architecture largely avoids.
Anthropic: Anthropic is an AI safety and research company that develops advanced language models and assistant products, including the Claude family, with a focus on reliability and aligned behavior. The news cites Anthropic alongside OpenAI to exemplify the frontier-scale models that pose challenges for SRAM-only chip architectures, framing the segment of the market where D-Matrix’s current approach is less competitive.
Sid Sheth: Sid Sheth is the co-founder and CEO of D-Matrix, with a background in semiconductor and AI hardware design. In this article, he is quoted promoting D-Matrix’s Corsair chip, arguing that AI inference is a trillion-dollar opportunity and positioning the company as an independent, stand-alone competitor rather than a near-term acquisition target.
Gimlet Labs: Gimlet Labs is an AI-focused research and benchmarking organization that analyzes hardware and system performance for machine learning workloads. D-Matrix cites Gimlet Labs’ research in the article to support its claim that pairing Corsair with an Nvidia Blackwell GPU can significantly improve inference speed, cost, and energy efficiency over a standalone GPU.
Super Micro: Super Micro (Supermicro) is a server and storage hardware vendor known for building customizable, high-density systems for cloud and AI data centers. The article notes that D-Matrix partnered with Super Micro to create its SquadRack system, packaging Corsair accelerator cards into server racks for plug-and-play AI inference in data centers.
Jensen Huang: Jensen Huang is the co-founder and CEO of Nvidia, widely recognized as a key figure in the AI hardware industry. In this article, he is quoted from a Computex keynote defending Nvidia’s leadership in low-cost inference by emphasizing the company’s vertically integrated, co-designed systems as a counterpoint to specialized accelerator startups like D-Matrix.
Stacy Rasgon: Stacy Rasgon is a senior semiconductor and equity research analyst at Bernstein Research who covers major chipmakers and AI hardware trends. In the news story, he provides external validation of D-Matrix by noting that startups like it often sell their accelerators alongside Nvidia GPUs and that D-Matrix appears to have meaningful customer engagements.

`json
{
“SRAM_vs_DRAM_Trend”: “There is increasing interest in AI architectures that use on-chip SRAM or closely integrated memory to boost efficiency and reduce latency, even though such designs may not support the largest models requiring external memory solutions like DRAM.”,
“AI_Chip_Competition”: “Industry updates reflect a growing number of AI accelerator startups, such as D-Matrix, aiming to provide specialized alternatives to Nvidia’s GPUs in specific AI processing tasks.”,
“Cloud_and_Hyperscaler_Demand”: “Recent industry discussions highlight cloud providers and advanced AI labs exploring diverse hardware setups by integrating GPUs with specialized AI accelerators to optimize for performance and efficiency across different workloads.”
}
`