d-Matrix
d-Matrix is a fabless semiconductor company that designs Digital In-Memory Compute (DIMC) AI inference accelerators — most notably the Corsair PCIe Gen5 platform and SquadRack rack-scale solution — sold to hyperscalers, neoclouds, and frontier AI labs for low-latency generative AI inference.
- Company typePrivate
- Founded2019
- HeadquartersSanta Clara, United States
- Headcount251–500
- GTM typeB2B
- OfferingHardware or Manufacturing
What d-Matrix does
d-Matrix is a Silicon Valley-based fabless semiconductor company founded in 2019 that designs data-center AI inference accelerators built around a proprietary Digital In-Memory Compute (DIMC) architecture. Its flagship product, Corsair, is a chiplet-based PCIe Gen5 accelerator using on-chip SRAM and off-chip LP-DDR5 capacity memory, claimed to deliver up to 150 TB/s of aggregate memory bandwidth (vs. ~4–8 TB/s for HBM-based GPUs) at approximately 0.3 pJ/bit, targeting sub-2 ms/token latency for large language model inference. The platform is rounded out by JetStream (a Transparent NIC/I/O accelerator for accelerator-to-accelerator communication over standard Ethernet), Aviator (an open software stack built on PyTorch, MLIR, and Triton), and SquadRack (a rack-scale reference design co-developed with Arista, Broadcom, and Supermicro). Roadmap products include the Pavehawk 3D-stacked DRAM test chip and the forthcoming Raptor accelerator, which integrate 3DIMC for an additional 10x memory bandwidth and energy efficiency step versus HBM4-based designs.
The company monetizes primarily through hardware sales of Corsair accelerators (single-server through full-rack configurations) to hyperscalers, neoclouds, and frontier AI labs under a quote-based, sales-led enterprise model with no publicly disclosed pricing. Distribution is multi-channel: direct enterprise sales (with approximately 90% of customers in the U.S. and the remainder in the Middle East and Southeast Asia), OEM/embedded integration via SquadRack, and cloud-channel deployment with Gimlet Labs. Manufacturing is outsourced to TSMC on the N6 process with Alchip providing ASIC/chiplet integration. Following a $275M Series C at a $2B valuation in November 2025 and the April 2026 acquisition of GigaIO's data center business, Corsair entered full production in June 2026, with the company operating across six global innovation hubs and approximately 101–250 employees.
d-Matrix firmographics
Firmographics- Name
- d-Matrix
- Legal name
- d-Matrix Corporation
- Website
- https://www.d-matrix.ai
- Company type
- Private
- Founded year
- 2019
- Operating status
- Operating
- Headcount range
- 251–500 employees
- Short description
- d-Matrix is a fabless semiconductor company that designs Digital In-Memory Compute (DIMC) AI inference accelerators — most notably the Corsair PCIe Gen5 platform and SquadRack rack-scale solution — sold to hyperscalers, neoclouds, and frontier AI labs for low-latency generative AI inference.
- Ownership category
- akta.pro rank
d-Matrix industry classification
Industry- Product category
- AI Inference Accelerators
- NAICS
- Computer and Peripheral Equipment Manufacturing (3341), Computer Systems Design and Related Services (54151), Software Publishers (5132)
- SIC
- Computer Peripheral Equipment, Nec (3577), Services-Computer Integrated Systems Design (7373), Printed Circuit Boards (3672)
- akta.pro primary industry
- AI Accelerators (GPUs/TPUs/NPUs/ASICs) (HDAAAAAA)
- akta.pro secondary industries
- GPU-Accelerated & AI Training/Inference Servers (HDACABAG), Graphics & Accelerator Cards (GPUs, compute accelerators) (HDACAGAA)
Keywords
Where d-Matrix is headquartered
LocationHeadquarters
- HQ city
- Santa Clara
- HQ country
- United States
- HQ region
- North America
Offices6 records
Markets served
d-Matrix business model
Business model- GTM type
- B2B
- Offering type
- Hardware or Manufacturing
- Cost components
- Technology or R&D, Supply Chain, Personnel, Operations, Infrastructure, Marketing or Sales
Revenue model
- Corsair Accelerator Hardware Sales: Primary revenue stream from selling Corsair C8 inference accelerators as PCIe cards, dual-card configurations, servers, and rack-scale solutions. Air-cooled, standard PCIe form factor for deployment in existing data centers. Full production began in summer 2026 with volume shipping to hyperscalers, neoclouds, and frontier AI labs.
- JetStream I/O Accelerator Sales: Revenue from JetStream Transparent NIC products bundled with or sold alongside Corsair accelerators to enable rack-scale inference deployments.
- SquadRack Rack-Scale Solution Sales: Revenue from complete rack-scale AI inference solutions integrating Corsair, JetStream, Aviator software, and partner components (Arista, Broadcom, Supermicro). Configurable as Single-Server, Dual-Server, Half-Rack, or Full-Rack deployments.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Other | — | Custom quote-based pricing for qualified customers |
Go-to-market motion3 records
Distribution channels3 records
Marketing channels7 records
d-Matrix product offering
Product offeringCore offering
d-Matrix designs, manufactures, and sells the Corsair family of digital in-memory compute (DIMC) AI inference accelerators as PCIe cards and rack-scale systems, paired with the JetStream I/O accelerator (Transparent NIC) and the Aviator software stack. The company targets data center generative AI inference workloads for hyperscalers, neoclouds, and frontier AI labs, claiming up to 10x faster inference and substantially lower latency and energy consumption than GPU-only deployments.
Product overview
d-Matrix offers a unified AI inference platform consisting of multiple integrated products rather than a single product. The core product, Corsair, is the digital in-memory compute inference accelerator, which pairs with JetStream (the I/O accelerator/Transparent NIC for accelerator-to-accelerator communications) and runs on the Aviator software stack. Together, these compose SquadRack, the rack-scale reference design for AI inference at datacenter scale, built with ecosystem partners Arista, Broadcom, and Supermicro. Looking forward, the platform is evolving with Pavehawk (3DIMC test silicon) and the forthcoming Raptor next-generation accelerator, while the GigaIO acquisition added SuperNODE rack-scale platform and FabreX PCIe memory fabric as part of the end-to-end solution.
Differentiator
Problem solved
Functional benefit
Brands
- Corsair: Digital In-Memory Compute (DIMC) AI inference accelerator platform for data centers.
- JetStream
- Aviator
- SquadRack
- 3DIMC
- Raptor
- Pavehawk
- Nighthawk
- Jayhawk
Products and services
- Corsair AI Inference Accelerator d-Matrix's flagship AI inference compute platform and the world's first digital in-memory compute (DIMC) inference accelerator. A standard PCIe FHFL dual-card solution offering up to 4800 TFLOPs MXINT8 dense and 19200 TFLOPs MXINT4 dense per pair, with 4 GB Performance Memory at 300 TB/s and up to 512 GB Capacity Memory. Designed for data center generative AI inference workloads, with claimed 20x better power efficiency, 20x lower latency, and 10x lower TCO versus H100 GPUs.
- JetStream I/O Accelerator d-Matrix's purpose-built I/O accelerator (Transparent NIC) for AI inference. Industry-first PCIe Gen5 NIC with up to 400 GB/s accelerator-to-accelerator connectivity built on IEEE 802.3 Ethernet protocol. Enables decoupled device-initiated data transfer without host-level round trips, powering low-latency multi-node scale-up for AI inference.
- SquadRack Rack-Scale Reference Design Industry's first rack-scale reference design purpose-built for AI inference using a disaggregated standards-based approach. Integrates Corsair accelerators, JetStream networking, and Aviator software with Arista (Ethernet switches), Broadcom (PCIe/Ethernet silicon), and Supermicro (servers). Configurable as Single-Server, Dual-Server, Half-Rack, or Full Rack deployments supporting models up to 100B parameters in performance mode and 1T+ frontier models in capacity mode.
- SuperNODE Rack-Scale Platform Rack-scale platform architecture acquired from GigaIO's data center business in April 2026. Combined with Corsair, a single SuperNODE supports 64 or more Corsair cards, eliminating complex multi-node configurations for AI inference deployment.
Quantifiable outcome
- 10x faster AI inference than standalone Nvidia GPU
- +9 more outcomes
Companies that use d-Matrix
Customer profileSegments5 records
Ideal customer profiles4 records
d-Matrix technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
Integration7 records
AI capability11 records
Feature9 records
d-Matrix partnerships and signals
Strategic signalPartnerships
Ten partnerships are on record, tiered core and flagship.
- Infineon TechnologiescoreInfineon and d-Matrix are collaborating to combine Corsair AI inference accelerators with Infineon's OptiMOS dual-phase power modules for high-density AI data center environments. Partnership addresses power delivery efficiency as a critical bottleneck for scaling AI infrastructure.
- GigaIOflagshipd-Matrix acquired GigaIO's data center business, including the SuperNODE platform and patented PCIe Gen 5-based FabreX AI fabric. The acquisition brought a team of systems engineers based in Carlsbad, California, and established d-Matrix's sixth innovation hub in Southern California. Builds on prior 2025 collaboration integrating d-Matrix Corsair into GigaIO SuperNODE. GigaIO continues independently focused on edge computing with its Gryf product.
- Gimlet Labsflagshipd-Matrix and Gimlet Labs partnered to integrate Corsair accelerators alongside traditional GPUs into Gimlet Cloud's hybrid architecture, delivering up to 10x improvements in latency and throughput per watt compared to GPU-only deployments. Targets frontier AI labs with availability in 2H 2026.
- Alchip Technologiesflagshipd-Matrix and Alchip are collaborating to develop the world's first 3D DRAM-based datacenter inference accelerator (d-Matrix Raptor), featuring d-Matrix 3DIMC technology. Alchip provides ASIC design expertise and chiplet integration for manufacturing. The collaboration enabled the Pavehawk test chip validated in d-Matrix's labs.
- Andes Technologyflagshipd-Matrix and Andes Technology are collaborating to develop a next-generation AI inference accelerator called Raptor, integrating AndesCore AX46MPV CPU with d-Matrix's 3D In-Memory Compute (3DIMC) technology for high-performance, energy-efficient AI inference at scale.
- HTECcoreHTEC and d-Matrix announced a strategic partnership for AI inference hardware development.
- TSMCflagshipTSMC manufactures d-Matrix Corsair on its established N6 process technology, providing reliable, high-volume manufacturing capability that enables d-Matrix to meet customer commitments. The partnership supports d-Matrix's supply chain predictability using organic substrates rather than HBM-based CoWoS packaging.
- Arista NetworkscoreArista provides cloud networking fabric and high-performance switches for the SquadRack reference design. JetStream's accelerator-to-accelerator communication over standard Ethernet pairs with Arista's switches, enabling efficient AI inference scale-out without proprietary networking fabrics. Quote from Vijay Vusirikala (Distinguished Lead, AI Systems and Networks, Arista Networks).
- BroadcomcoreBroadcom provides high-performance PCIe and Ethernet connectivity silicon for the SquadRack reference design. d-Matrix leverages Broadcom's standards-based networking ecosystem for AI inference. Quote from Jas Tremblay (VP and GM, Data Center Solutions Group, Broadcom).
- SupermicrocoreSupermicro integrates d-Matrix Corsair into high-performance end-to-end liquid- and air-cooled rack-level systems for the SquadRack reference design. Supermicro provides proven rack-level integration expertise. Quotes from Vik Malyala (President & Managing Director, EMEA and SVP Technology & AI, Supermicro).
Scale indicators8 records
Recent moves7 records
Expansion highlights7 records
d-Matrix competitors and assessment
Company assessmentBroad incumbents
- AMD: AMD's MI300/MI325/MI350 accelerators and ROCm software stack compete in the same datacenter AI inference segment. As a broader incumbent, AMD is the most credible general-purpose GPU alternative for customers evaluating d-Matrix.
- Nvidia: Nvidia is the dominant incumbent in AI compute with H100/H200/Blackwell GPUs and the CUDA software ecosystem. d-Matrix explicitly benchmarks against H100 and positions DIMC as a 10x performance-per-watt alternative for inference workloads at hyperscalers and neoclouds.
- Intel (Habana Labs): Intel's Habana Gaudi and Gaudi 2/3 accelerators target training and inference for hyperscalers and enterprises. Like d-Matrix, they pursue cost-efficient alternatives to Nvidia for generative AI workloads at hyperscale customers.
Direct peers
- Groq: Groq builds purpose-built LPU inference accelerators targeting ultra-low latency generative AI inference in datacenters. It is the closest direct competitor to d-Matrix's Corsair, both competing on token-throughput economics versus Nvidia GPUs for hyperscaler and frontier lab deployments.
- SambaNova Systems: SambaNova builds RDU (Reconfigurable Dataflow Unit) chips focused on enterprise and hyperscale generative AI inference. It targets similar customers (large enterprises, sovereign deployments) and is at a comparable stage as a venture-backed AI inference silicon vendor.
- Cerebras Systems: Cerebras develops wafer-scale AI accelerators (WSE) for both training and inference, with a recently public profile (NASDAQ: CBRS) as a comparable AI-chip-as-platform vendor. It competes for frontier AI lab and hyperscaler budgets on the same inference-economics pitch d-Matrix makes.
Emerging players
- Etched: Etched builds transformer-specific ASICs (Sohu chip) targeting LLM inference. It overlaps with d-Matrix on the inference-only thesis but is more narrowly focused on transformer workloads rather than the broader DIMC chiplet platform d-Matrix offers.
- Positron AI: Positron develops memory-centric inference accelerators (Atlas and Titan hardware) targeting generative AI inference economics. It is an emerging direct competitor in the inference-ASIC segment, pursuing similar performance-per-watt advantages versus GPUs.
- Tenstorrent: Tenstorrent develops RISC-V-based AI accelerators for training and inference at datacenter and edge. It overlaps with d-Matrix's enterprise and hyperscaler pitch but uses a different underlying architecture (RISC-V/Grayskull/Wormhole) versus d-Matrix's DIMC approach.
Regional players
- Rebellions: Rebellions is a Korean AI chip startup (REBEL and ATOM chips) with strong ties to SK Hynix and Samsung — the same Korean strategic investors backing d-Matrix. It is a comparable inference-focused competitor primarily serving Asia-Pacific and Korean hyperscaler/enterprise customers.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Customer concentration
d-Matrix social profiles
Digital presenced-Matrix financial estimates
Financial estimateRevenue estimate
Valuation estimate
d-Matrix leadership team
Management profileNumber of profiles
Profiles15 records
d-Matrix subsidiaries and ownership
Company hierarchySubsidiaries1 record
d-Matrix funding detail
Funding detailFunding overview
Funding rounds4 records
Investors27 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
d-Matrix M&A and investment
M&A and investmentM&A2 records
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about d-Matrix
What does d-Matrix do?
d-Matrix designs, manufactures, and sells the Corsair family of digital in-memory compute (DIMC) AI inference accelerators as PCIe cards and rack-scale systems, paired with the JetStream I/O accelerator (Transparent NIC) and the Aviator software stack. The company targets data center generative AI inference workloads for hyperscalers, neoclouds, and frontier AI labs, claiming up to 10x faster inference and substantially lower latency and energy consumption than GPU-only deployments.
Is d-Matrix a public or private company?
d-Matrix is a private company. It is classified as venture growth investor backed and is currently operating.
When was d-Matrix founded?
d-Matrix was founded in 2019. It employs 251 to 500 people.
Where is d-Matrix based?
d-Matrix is headquartered in Santa Clara, United States, in the North America region.
How does d-Matrix make money?
Three revenue lines are on record. Corsair Accelerator Hardware Sales are the primary driver. The others are jetStream I/O Accelerator Sales and squadRack Rack-Scale Solution Sales.
Who are d-Matrix's main competitors?
Broad incumbents on record are AMD, Nvidia and Intel (Habana Labs). Direct peers are Groq, SambaNova Systems and Cerebras Systems. Emerging players are Etched, Positron AI and Tenstorrent. Rebellions is listed as a regional player.
Does d-Matrix have an API?
No public API is recorded for d-Matrix.
What industry is d-Matrix in?
d-Matrix's product category is AI Inference Accelerators. Its primary akta.pro industry code is HDAAAAAA, AI Accelerators (GPUs/TPUs/NPUs/ASICs), with a secondary code of HDACABAG, GPU-Accelerated & AI Training/Inference Servers. Its NAICS code is 3341 and its SIC code is 3577.