d-Matrix
"Low latency AI inference without compromises"
Fabless semiconductor startup whose Corsair platform uses digital in-memory compute to accelerate generative AI inference in datacenters.
d-Matrix is a fabless semiconductor company headquartered in Santa Clara, California, building purpose-built compute hardware and software for generative AI inference in the datacenter. Its flagship Corsair platform is based on a proprietary Digital In-Memory Compute (DIMC) architecture that integrates memory and compute on chiplet-based silicon to overcome the memory-bandwidth bottleneck that limits GPU-based inference. The Corsair Dual Card is a full-height, full-length PCIe card delivering 4,800 TFLOPs MXINT8 and 19,200 TFLOPs MXINT4 dense performance, 4 GB of performance memory at 300 TB/s, up to 512 GB of capacity memory, and a 512 GB/s card-to-card DMX Bridge. The architecture scales from a single card to an eight-card inference server and a 64-card inference rack capable of running models from 100B parameters in performance mode to over one trillion parameters in capacity mode.
The company also offers JetStream, a custom PCIe I/O accelerator (NIC) for accelerator-to-accelerator communication, and SquadRack, a disaggregated rack-scale solution built with partners including Supermicro, Arista and Broadcom. Its Aviator software stack integrates with open frameworks such as PyTorch, Triton and MLIR. d-Matrix targets up to 10x interactive speed, 3x cost-performance and 3x energy efficiency versus comparable GPU systems for large language model inference. Founded in 2019 by Sid Sheth and Sudeep Bhoja, the company raised a $275 million Series C in 2024 and operates additional offices in Toronto, Sydney, Bangalore and Belgrade. In 2026 Corsair entered full production to meet customer demand.
Primary type & automation activities this supplier delivers:
Corsair
PCIe AI inference accelerator using digital in-memory compute for generative AI in the datacenter.
JetStream
Custom PCIe I/O accelerator (NIC) for blazing-fast accelerator-to-accelerator scale-up in AI inference.
Contact d-Matrix
WEBSITE
https://www.d-matrix.aiHEADQUARTERS
United States
Company Facts
Founded
2019
Primary Role
Component Supplier
Company Size
-
Primary Region
North America
Annual Sales
-
Funding Stage
Growth / Late-Stage VC
Funding Total
-
Certifications & Memberships
Introduction to Robotics J. J. Craig — RSF Top 10 Robot Education Textbooks · No. 6
Six of ten benchmark universities teach kinematics from Craig. Frames, transforms and D-H parameters: the language vendor manuals already speak.
Modern Robotics K. M. Lynch & F. C. Park — RSF Top 10 Robot Education Textbooks · No. 8
The book the field agreed to make free: full preprint PDF, video lectures and open code from the authors. Screw theory and kinematics, taught rigorously.
RSF Top 10 Robot Education Textbooks
Robotics has a canon; robot service does not. Ten books scored on adoption, RSF layer coverage, field value and difficulty, plus six specialist picks.
Capex Analysis 2025: AI Hyperscalers vs Humanoid Robots — Financing Risk Deep Dive
Compare how Microsoft, Google, Oracle, Tesla Optimus, Figure AI and UBTECH finance their Capex. Supply chain risk, ROIC analysis and a 3-dimension investment framework included.
Upstart chipmakers keep challenging Nvidia. This time it's Microsoft-backed D-Matrix
D-Matrix, a competitor to Nvidia, has announced that it is commencing full production of a new AI chip, which the company claims is ten times faster than traditional GPUs. This development aims to address the ongoing memory shortage that has been affecting the tech industry. The chip's innovative design allows it to operate efficiently without relying heavily on conventional memory resources. D-Matrix's move comes at a crucial time as demand for advanced AI processing capabilities continues to surge, and the company seeks to carve out a significant market share in the rapidly evolving landscape of artificial intelligence technology.
Integrated Joint Actuators 2026: Technology, Global Suppliers & China's Leap Forward
From rotary joints hitting 530 N·m+ to linear joints with planetary roller screws, integrated actuators are replacing the three-part drivetrain. Full supplier map: Tihu, PhyArc, RobStride, Schaeffler, maxon — and China's path from 1,000-RMB modules to global standard-setter.
State Grid’s $938M Robot Mega-Order Signals China’s Industrial AI Shift
State Grid Corporation of China plans to buy 8,500 robots for $938M — the largest embodied AI procurement in history. What it means for China's robot market.
China vs. The World: A Space Robotics Competitive Analysis
A domain-by-domain competitive benchmark of China versus the international space robotics field — covering on-orbit servicing, lunar ISRU, space station
China's 15th Five-Year Plan and the Space Robotics Imperative
China's 15th Five-Year Plan names aerospace a strategic pillar industry alongside semiconductors and sets hard mission schedules — Chang'e-7, Chang'e-8, crewed lunar landing — that make space robotics one of the most consequential engineering battlegrounds of the late 2020s.
Columbia's Emo: Solving Robot Face Problem
Columbia's Emo robot breakthrough reveals why facial biomimicry is robotics' second hardest challenge. Analysis of tech barriers and 2028-2030 timeline.
Laser Weeding Robotics 2025 Carbon Robotics Global Lead Vs Hg Tech S China Challenge
Laser weeding robots in 2025 are reshaping agriculture, with Carbon Robotics leading globally while HG Tech pushes competitive innovation from China.
