AI Accelerators

UNISOC

CN · Shanghai, China

Fourth largest mobile processor manufacturer globally with 13% market share. 5G chips for budget/mid-range phones.

Synopsys

US · Sunnyvale, CA

EDA tools and DesignWare IP. AI-driven design with DSO.ai. ANSYS acquisition 2024.

Synthara

CH · Zurich, Switzerland

ComputeRAM in-memory computing for embedded applications. Backed by Hermann Hauser and Pat Gelsinger.

Vybium

AT · Vienna, Austria

RISC-V AI accelerators for European Digital Strategic Autonomy. Competing with Nvidia datacenter GPUs.

Arm Holdings

GB · Cambridge

World's leading semiconductor IP company. ARM architecture used in 99% of smartphones. Cortex-A/R/M CPU cores, Mali GPUs, Ethos NPUs. Powers billions of devices annually.

Imagination Technologies

GB · Kings Langley

UK-based semiconductor IP company. PowerVR GPU and IMG NPU cores for mobile, automotive, and IoT. Powers Apple devices, MediaTek, and automotive systems.

Renesas Electronics

JP · Tokyo

Japanese semiconductor giant. Automotive MCUs, analog, power, AI/ML edge processors. Multiple acquisitions.

Ventana Micro Systems

US · Cupertino, CA

High-performance RISC-V CPU chiplets for data centers. Veyron processor, 192 cores, D2D interfaces.

Ricursive Intelligence

US · Palo Alto, CA

Uses AI to design semiconductor chips (built on AlphaChip). Recursive loop: AI designs better chips → chips train stronger AI. Targets compressing chip design from years to weeks. Founded by Dr. Anna Goldie and Dr. Azalia Mirhoseini (AlphaChip co-creators, ex-Google Brain). $335M raised ($300M Series A at $4B val, Feb 2026, Lightspeed + NVIDIA NVentures).

Xsight Labs

IL · Kiryat Gat, Israel

Programmable SDN Ethernet switches and ARM CPU-based DPUs for AI and hyperscale data centers.

FPGA-based sustainable AI. $23.5M raised (2025).

Axelera AI

NL · Eindhoven, Netherlands

European edge AI chips. Metis AIPU, in-memory computing.

Sakana AI

JP · Tokyo, Japan

Japanese AI unicorn developing nature-inspired AI models. Achieved unicorn status in under a year.

AI agents for chip design. $24M raised, LLM-powered EDA.

GPU orchestration and workload management. Acquired by NVIDIA (2024).

Quadric

US · Burlingame, CA

Edge AI supercomputer chips. Chimera GPNPU architecture.

Vastai Technologies

CN · China

AI acceleration chip developer for cloud and edge computing applications in Chinese market.

RISC-V data center chips. Acquired by Meta, ~$2B valuation.

Salience Labs

GB · Oxford, UK

Oxford/Münster spinout. Ultra high-speed multi-chip processors combining photonics and electronics for AI.

Starcloud (Lumen Orbit)

US · Redmond, WA

Data centers in space for GPU compute. Solar energy and passive radiative cooling. First launch May 2025, partnership with NVIDIA.

Vertical Compute

BE · Louvain-La-Neuve, Belgium

imec spinoff developing vertically integrated memory and compute for AI accelerators with chiplet-based solutions.

GreenWaves Technologies

FR · Grenoble, France

GAP ultra-low power processors. RISC-V based edge AI from France.

Kunlunxin

CN · Beijing, China

Baidu AI chip spinout. AI accelerators for search, cloud AI, and Baidu products. Revenue projected at 3.5B yuan ($500M) in 2025, 8B yuan in 2026. Filed confidentially for Hong Kong Stock Exchange IPO (Jan 2026).

Space and robotics AI chips. Radiation-hardened edge inference.

DeGirum

US · Santa Clara, CA

ORCA edge AI accelerator for computer vision. Hardware-agnostic PySDK. Ex-Marvell founders.

University of Michigan spinout. Memristor + CMOS integration for AI inference.

Celus

DE · Munich, Germany

AI-powered PCB and electronic design automation. Generative design for hardware engineers.

Perceive

US · San Jose, CA

Edge AI inference chips. Ergo processor for ultra-low power AI.

LightOn

FR · Paris, France

Optical processing units (OPUs) for AI and random projections. French photonic computing pioneer.

RoboSense

CN · Shenzhen, China

World's #1 LiDAR company with 1M+ units delivered. Proprietary M-Core SoC chip for LiDAR processing.

Huawei Ascend

CN · Shenzhen, China

Ascend 910B/910C chips. Roadmap to 970 (2028), targeting 4 ZettaFLOPS by 2028.

Merged with Rebellions (2024). X330 AI chip, SK Telecom spinoff.

Broadcom

US · Palo Alto, CA

Custom AI ASICs with optical attach, Tomahawk switches. Major hyperscaler partnerships.

Founded by ex-Nvidia China GM. $1.1B Shanghai IPO planned, DeepSeek/Qwen compatible.

Horizon Robotics

CN · Beijing, China

Journey series autonomous driving chips. 49% China ADAS market share.

Cadence Design Systems

US · San Jose, CA

EDA tools and Tensilica processor IP. AI-driven chip design with Cerebrus.

Speedata

IL · Tel Aviv, Israel

Analytics Processing Unit (APU). Apache Spark acceleration.

Kontron

DE · Augsburg, Germany

Global leader in IoT and embedded computing with rugged SOSA-aligned VPX solutions for defense applications.

Marvell Technology

US · Wilmington, DE

Silicon photonics leader. Acquired Celestial AI ($3.25B, Dec 2025), 1.6T light engines for AI networks.

Google DeepMind TPU

US · Mountain View, CA

Trillium TPU v6 (2024). 4.7x improvement over v5, custom Axion ARM CPUs.

Astera Labs

US · Santa Clara, CA

AI connectivity semiconductors. ~$10B market cap (public), PCIe/CXL retimers and smart cables.

HOUMO.AI

CN · China

AI chip company for autonomous driving applications. Computing solutions for smart vehicles.

Exa Laboratories

SE · Stockholm, Sweden

Reconfigurable dataflow chips (XPUs) for AI training and inference. YC S24, claims 27.6x efficiency vs H100.

Kalray

FR · Grenoble, France

MPPA intelligent processor. Coolidge chip for data centers, storage, and automotive.

SiMa.ai

US · San Jose, CA

MLSoC platform for edge computer vision applications.

Tesla Dojo

US · Austin, TX

D1 chip for FSD training. Exapod supercomputer, custom interconnect.

Cornami

US · Campbell, CA

Fully homomorphic encryption acceleration. Privacy-preserving compute.

CEVA

US · Rockville, MD

DSP and AI processor IP for wireless, audio, vision. SensPro2 NPU, Bluetooth/Wi-Fi IP.

SEMRON

DE · Dresden

German analog AI chip startup using memcapacitor technology. Encoding neural network weights in hardware for extreme energy efficiency in AI inference.

Metalware

US · San Francisco, CA

Y Combinator-backed AI tools for accelerating HDL and embedded C/C++ development from datasheets.

Corintis

EU · Europe

In-chip microfluidic cooling system for data centers. Revolutionary thermal management for AI chips.

Gyrfalcon Technology

US · Milpitas, CA

Lightspeeur AI accelerators. Low-power edge inference chips.

Ubitium

EU · Europe

Universal RISC-V processor handling CPU, GPU, DSP, and FPGA workloads on single chip architecture.

Tencent-backed AI chips. T20/L600 chips, TSMC access, not on Entity List.

Positron AI

US · Reno, NV

FPGA-powered AI inference with memory-optimized architecture. Claims 3.5x better performance per dollar than H100.

Optalysys

GB · Glasgow, UK

Optical computing for genomics and AI. Light-based Fourier transforms.

Analog AI inference chips. Ultra-low power continuous signal processing.

Etched

US · San Francisco, CA

Sohu — Transformer-specific ASIC on TSMC 4nm. Claims 10–20× better performance-per-watt vs H100. $500M raise (2026) at $5B valuation; $1B total funding. Stripes, Peter Thiel, Ribbit Capital investors.

Normal Computing

US · San Francisco, CA

Thermodynamic computing — physics-based ASICs that exploit natural noise for 1000x energy efficiency gains in generative AI inference. CN101 tape-out completed Aug 2025 (world's first thermodynamic chip). $85M+ raised; Samsung Catalyst, Micron Ventures, Eric Schmidt's First Spark Ventures investors.

Leonardo DRS

US · Arlington, VA

Defense technology with 100,000+ mission computers delivered to US Army. AI Processor (AIP) for tactical edge.

Iluvatar CoreX

CN · Shanghai

Chinese AI chip startup developing general-purpose GPUs and AI accelerators. Targeting data center AI training and inference markets with domestic alternatives.

HyperAccel

KR · Seoul, South Korea

LLM Processing Unit with LPDDR5X memory on Samsung 4nm process. Optimized for large language model inference.

Socionext

JP · Yokohama, Japan

Major custom SoC provider (2nd globally) developing ADAS/AD SoCs using TSMC 3nm/5nm automotive processes.

Openchip

ES · Barcelona, Spain

European systems company integrating RISC-V accelerators and SoCs. Leading EU DARE project.

Duality Technologies

US · Newark, NJ

Fully homomorphic encryption (FHE) accelerator chips enabling computation on encrypted data at near-native speed.

Synaptics

US · San Jose, CA

Human interface solutions. Touch controllers, display drivers, fingerprint sensors, AI-powered audio/video ICs.

AI inference optimization. Hardware-software co-design.

Parallel Processing Unit (PPU). Claims 100x CPU performance boost.

VeriSilicon

CN · Shanghai

Chinese semiconductor IP and custom chip company. Vivante GPU IP, ZSP DSP, VIP NPU cores. One-stop silicon platform for fabless companies.

LeddarTech

CA · Quebec City, QC

LeddarEngine SoC for ToF LiDAR and LeddarVision sensor fusion software supporting SAE L2-L5.

d-Matrix

US · Santa Clara, CA

In-memory computing. Corsair platform: 10× GPU performance, 3× lower cost, 3–5× better energy efficiency. Includes Corsair accelerators, JetStream networking ASICs, Aviator software. $275M Series C (Nov 2025, $2B valuation, $450M total). Temasek, QIA, Microsoft M12 investors.

Outerport

US · San Francisco, CA

Hot-swapping system for AI model weights. Swap models on same GPU in ~2 seconds (150x faster). Save 40% on GPU costs.

Semifive

KR · Seongnam, South Korea

Korean custom silicon platform. RISC-V based AI/HPC chips.

Tenstorrent

US · Austin, TX

RISC-V AI chips including Wormhole and Blackhole. $700M Series D (Dec 2024). $800M Series E in talks (Fidelity-led, ~$3.2B valuation). Jim Keller as CEO. Samsung fab; Hyundai automotive AI contracts.

Cerebras Systems

US · Sunnyvale, CA

Wafer-Scale Engine (WSE-3): 900K cores, 4T transistors. OpenAI $20B+ chip supply deal. $1B Series H (Feb 2026) at $23B valuation. Filed Nasdaq IPO April 2026 targeting $35B+ valuation.

NextSilicon

IL · Israel

Reconfigurable/adaptive HPC accelerators with Intelligent Compute Architecture. Better performance-to-power than NVIDIA.

CoreWeave

US · Roseland, NJ

GPU cloud provider. $23B valuation, NVIDIA-powered AI infrastructure.

FuriosaAI

KR · Seoul, South Korea

RNGD chip. $246M total funding, 2.25x performance/watt vs GPUs. Meta acquisition interest.

Lambda Labs

US · San Francisco, CA

GPU cloud for AI. Deep learning workstations, cloud GPU clusters.

AI accelerator startup. $28M Series A (Dec 2025).

Achronix

US · Santa Clara, CA

Speedster7t, VectorPath cards. 200% cost/power advantage vs GPUs for LLMs, eFPGA IP.

Nanya Technology

TW · Taipei, Taiwan

Taiwan's #4 global DRAM manufacturer. Developing customized HBM for edge AI including AI PCs and robotics.

Ceremorphic

US · San Jose, CA

High-performance AI training chips with hierarchical deep learning architecture. Founded by former Intel VP.

AI accelerator startup. Next-gen inference optimization.

NVIDIA

US · Santa Clara, CA

GPU market leader (~88% share). Blackwell H100/H200/B200 deployed at scale. Vera Rubin NVL72 (50 PFLOPS FP4, HBM4, NVLink 6) shipping H2 2026. Acquired Groq assets ($20B, Dec 2025). CUDA ecosystem dominant.

Preferred Networks

JP · Tokyo, Japan

MN-Core AI processors. MN-Core 2 (7nm), L1000 GenAI chip (2026), Green500 #1 supercomputer.

GSI Technology

US · Sunnyvale, CA

Associative Processing Unit (APU) delivering GPU-class AI performance with 98% lower energy via compute-in-SRAM.

Kinara

US · San Jose, CA

Ara-2 processor. NXP partnership - positioned as "CUDA of automotive".

Andes Technology

TW · Hsinchu, Taiwan

RISC-V CPU IP provider and founding RISC-V International Premier member. D23-SE core for ASIL-B/D automotive safety.

SparseOne series. Sparsity-focused, ultra-efficient (32x sparsification).

Esperanto Technologies

US · Mountain View, CA

RISC-V based ET-SoC-1 chip with 1,088 RISC-V cores delivering 100-200 TOPS at <20W for ML inference and HPC.

NEXTDC

AU · Brisbane, Australia

AI-optimized data centers. ASX listed, NVIDIA platform certified, Australia's largest.

Rebellions

KR · Seoul, South Korea

Korean AI chip unicorn. Merged with Sapeon (Dec 2024). Rebel-Quad AI accelerator in production. $850M raised, $2.34B valuation. Arm + Samsung investors. IPO targeted for 2026.

Photonic AI accelerators. $26M seed (2025), Paris-based, energy-efficient.

LeapMind

JP · Tokyo, Japan

Ultra-low-power deep learning on small FPGAs. DeLTA-Lite tools for neural network optimization.

Ambarella

US · Santa Clara, CA

Video AI chips with CV/ISP integration. Edge AI for automotive and security.

Quanta Computer

TW · Taoyuan

World's largest notebook ODM. Major AI server manufacturer for hyperscalers. NVIDIA GPU server partner.

PrimisAI

US · San Jose, CA

Generative AI for chip design with RapidGPT natural language interface for HDL and verification.

AheadComputing

US · Beaverton, OR

64-bit RISC-V application processors to tackle CPU bottlenecks in AI workloads. Founded by former Intel CPU architects.

Qualcomm

US · San Diego, CA

Hexagon DSP in Snapdragon SoCs. AI Engine for on-device ML, billions of devices powered.

BR166 GPU lineup. ~$2B valuation, Hong Kong IPO planned.

Netrasemi

IN · Bangalore, India

Graph stream parallel SoCs. $12.5M Series A.

Liquid AI

US · Boston, MA

Liquid Neural Networks from MIT spinout (2023). Biologically-inspired flexible AI models.

Navitas Semiconductor

IE · Dublin, Ireland

GaNFast gallium nitride power ICs and GeneSiC silicon carbide MOSFETs for AI data centers, EVs, and consumer electronics.

Empower Semiconductor

US · San Jose, CA

High-performance integrated power management for xPUs, enabling full processor speed with IVR technology.

Kneron

US · San Diego, CA

Edge AI for autonomous vehicles. $190M+ raised.

Calterah Semiconductor

CN · Shanghai, China

mmWave radar chip portfolio including 77/79 GHz and 60 GHz SoCs. Ships 16 million units annually.

Cambricon

CN · Beijing, China

MLU 590 (345 TFLOPS FP16). ~$79B market cap, Big Fund II backed, only pure-play A-share AI chip.

Chiplego

CN · Shanghai

Chinese chiplet startup developing modular semiconductor solutions. Creating specialized chiplets for automotive and AI applications amid export restrictions.

VersaLogic

US · Tualatin, OR

40+ years of embedded computers with MIL-STD-202H qualification. Sabertooth AI system for defense applications.

Uhnder

US · Austin, TX

Pioneer of digital 4D imaging radar-on-chip. 16x better resolution than analog radar with DCM technology.

Blueberry Semiconductors

IN · Bengaluru

VLSI, ASIC, and SoC solutions. AI, 5G, and industrial IoT expertise. Indian semiconductor design services.

Huashan A2000 automotive SoC. High-performance ADAS chips.

OpenAI (Custom Chip)

US · San Francisco, CA

Custom chip developed with Broadcom. 3nm TSMC, mass production 2026.

Felafax

US · San Francisco, CA

Open-source AI platform for non-NVIDIA GPUs (TPU, Trainium, AMD, Intel). 30% lower cost. Trained LLaMA3 405B on AMD MI300x.

AMD

US · Santa Clara, CA

Radeon RDNA 4 GPUs (RX 9000 series). Instinct MI350 (CDNA 4, 3nm, 288GB HBM3E) shipping 2025; MI455X (HBM4, 40 PFLOPS FP4) for 2026. Acquired Enosemi (photonics, May 2025). Helios AI rack. ROCm ecosystem.

Altair Engineering

US · Troy, MI

Computational science and AI company. HyperWorks simulation, electromagnetic analysis, AI/ML tools. Being acquired by Siemens (2024).

UPMEM

FR · Grenoble, France

Processing-in-Memory acceleration with hundreds of programmable coprocessors integrated into DRAM. Acquired by Qualcomm.

ECM3532 ultra-low power AI. Continuous learning at the edge, acquired by Renesas.

Circuit Mind

GB · London, UK

AI-driven electronics design software. Automates schematic capture and component selection.

C500/C600 chips, dual-chiplet design. Ex-AMD founders, $545M IPO planned.

Graphcore

GB · Bristol, UK

Intelligence Processing Unit (IPU). Acquired by SoftBank ($600M+, 2024).

Photonic AI supercomputer on a chip. Claims to surpass NVIDIA GPUs in performance/watt.

SambaNova

US · Palo Alto, CA

Reconfigurable Dataflow Units (RDU). Intel acquisition rejected; $350M strategic investment from Intel Capital (Feb 2026) + partnership. SambaNova Cloud expanding on Intel Xeon infrastructure. Enterprise AI inference focus.

Meta (MTIA)

US · Menlo Park, CA

MTIA v2 for training and inference. Built on 5nm process.

Ahead Computing

US · Austin, TX

RISC-V processor IP for AI and HPC. High-performance vector processing cores.

Zettascale

US · San Francisco, CA

Energy-efficient reconfigurable XPUs for AI training and inference. Polymorphic dataflow architecture optimizes per model.

Snowcap Compute

US · Palo Alto, CA

Superconducting compute platform for AI and quantum-classical hybrid using Josephson junctions. Pat Gelsinger is Chairman.

AlphaICs

IN · Bangalore, India

RAP (Real AI Processor) for edge inference. Bangalore-based AI chip startup.

Curtiss-Wright Defense

US · Davidson, NC

SOSA-aligned embedded computing for defense. VPX GPU modules with NVIDIA Blackwell delivering 100+ TFLOPS.

Opteran

GB · Sheffield

Brain-inspired autonomous machine intelligence. Natural Intelligence algorithms from insect brain research.

Blueshift Memory

GB · Cambridge, UK

Near-data processing memory architecture. Reduces data movement for AI and analytics workloads.

Oculii

US · Dayton, OH

AI-powered radar software for automotive. Virtual aperture imaging radar. Acquired by Ambarella.

SiPearl

FR · Maisons-Laffitte, France

Rhea processor for EU supercomputers. €130M raised (2025), Arm/Atos backed, Jupiter supercomputer.

Taalas

CA · Toronto, Canada

Automated flow for implementing deep learning models in silicon. Entire LLM on one chip without external memory. Founded by Tenstorrent veterans.

Crusoe Energy

US · Denver, CO

Sustainable AI compute. Flare gas-powered data centers for AI.

Ouster

US · San Francisco, CA

Digital 3D LiDAR with Chronos automotive-grade chip. Merged with Velodyne 2023. 600+ customers in 50+ countries.

EdgeQ

US · Santa Clara, CA

5G+AI converged chips. Base station and edge infrastructure.

MediaTek

TW · Hsinchu, Taiwan

Dimensity mobile SoCs. Taiwan's largest fabless company, NVIDIA DGX Spark collaboration.

Recogni

US · San Jose, CA

Vision AI processors. Automotive perception, low-latency inference.

Shanghai Tianshu Zhixin

CN · Shanghai, China

AI chip company focused on data center applications. Developing domestic alternatives to Western GPUs.

Optical Processing Unit (OPU) using micron-scale metamaterial modulators — 10,000× smaller than conventional photonic elements. Claims 100× performance and efficiency vs leading GPUs. Production target mid-2028. $118M total ($110M oversubscribed Series A, Jan 2026) — Gates Frontier, Microsoft M12, Aramco Ventures, Bosch Ventures.

Indie Semiconductor

US · Aliso Viejo, CA

Pure-play automotive fabless semiconductor for driver safety and electrification. Surya LiDAR SoC.

Hailo

IL · Tel Aviv, Israel

Hailo-8, Hailo-10H (GenAI at edge, Jul 2025), Hailo-15. Valuation dropped from $1.2B peak to under $500M. Emergency bridge loan Jan 2026. 10% workforce layoff; pivoting to robotics. SPAC IPO planned (MOU Mar 2026).

AWS Trainium/Inferentia

US · Seattle, WA

Trainium2 and Inferentia2 custom AI chips for AWS. UltraCluster 100K chip clusters.

Morphing Machines

IN · Bengaluru, India

REDEFINE reconfigurable processor. IISc Bengaluru spinout, ₹38Cr Series A (2025), FPGA-like flexibility.

Untether AI

CA · Toronto, Canada

At-memory architecture. speedAI240 with 2 PetaOps. Acquired by AMD (2025).

SpacemiT

CN · Hangzhou

RISC-V processor company. K1 octa-core RISC-V SoC for AI edge applications. Founded by ex-Intel/ARM engineers.

LLM training chip startup targeting NVIDIA GPU replacement. $620M raised ($500M Series B, Feb 2026) at ~$3B+ valuation. Jane Street + Situational Awareness led. Marvell, Spark Capital, Stripe investors. Ex-Google TPU engineers.

Korean edge AI. $15.3M raised, automotive vision focus.

Fabric Cryptography

US · Santa Clara, CA

Verifiable Processing Unit (VPU) for zero-knowledge proofs and FHE. Founded by MIT/Stanford dropouts.

Optalysys

GB · Glasgow, UK

Optical computing for FHE and AI. Light-based Fourier transform processors.

Qualcomm

US · San Diego, CA

Mobile chip leader. Snapdragon processors for smartphones, automotive, IoT. 5G modems, WiFi 7, AI/ML acceleration. ARM architecture licensing.

Tachyum

SK · Bratislava, Slovakia

Prodigy universal processor. CPU/GPU/TPU unified architecture from Slovakia.

DEEPX

KR · Seoul, South Korea

Korean AI chip startup. DX-M1 NPU for on-device AI.

HyperCIM

US · San Jose, CA

Compute-in-memory AI accelerators for edge inference. Ultra-low power neural network processing.

Neuchips

TW · Taipei, Taiwan

DLRM and recommendation system specialist. RecAccel chip for deep learning.

SandboxAQ

US · Palo Alto, CA

Google/Alphabet spin-off combining quantum sensors with AI for ultra-precise detection in navigation and medical diagnostics.

VSORA

FR · Paris, France

AI processors for autonomous vehicles. $46M raised, software-defined approach.

Thales Group

FR · Paris, France

European leader in defence electronics. €20.6B revenue. Fly-by-wire systems, mission computers, quantum technologies.

Groq

US · Mountain View, CA

Language Processing Unit (LPU) — 500K tokens/sec inference speed. NVIDIA acquired ~$20B in Groq assets + ~90% of employees (Dec 24, 2025). GroqCloud continues as standalone inference service under new CEO.

ScaleFlux

US · San Jose, CA

Computational storage SoC with integrated data compression. 4x better performance per watt for AI workloads.

Codasip

CZ · Brno

Czech RISC-V processor IP company offering customizable processor cores and design automation tools. Codasip Studio enables custom instruction design for optimized RISC-V implementations.

Elbit Systems

IL · Haifa, Israel

Israel's first computer company (1966). Intelligent avionics, decision support systems, and EW solutions. $5.5B revenue.

Optical fiber computing. All-optical AI processing in fiber.

RISC-V AI chips. Chinese startup, edge AI processors.

Megaport

AU · Brisbane, Australia

Network-as-a-Service for AI infrastructure. ASX listed, global interconnect.

Ferveret

US · San Jose, CA

Two-phase immersion liquid cooling inspired by nuclear reactors. 96% cooling cost savings, 2x chip performance, 300+kW per rack.

Cepton

US · San Jose, CA

LiDAR with proprietary Komodo ASIC for point cloud processing. ISO26262 ASIL-B certified. Acquired by Koito.

Composable AI infrastructure. Unified memory architecture.

Momenta

CN · Beijing, China

Full-stack autonomous driving company developing self-developed ADAS chips. Supplies 60% of city-NOA solutions in China.

Apple Silicon

US · Cupertino, CA

M-series chips (M4, M4 Pro/Max/Ultra). Industry-leading perf/watt for consumer.

Cornelis Networks

US · Wayne, PA

HPC scale-out networking. Omni-Path architecture for AI/HPC. CN5000 chips. NNSA partnership. Supermicro collaboration.

Mobileye

IL · Jerusalem

Autonomous driving chips and ADAS. Intel subsidiary. EyeQ SoC processors. Powers BMW, VW, Ford autonomous systems.

Extropic

US · San Francisco, CA

Thermodynamic AI computing using superconducting processors. Z1 chip with 250K p-bits, claims 10,000x less energy than GPUs.

NeuroBlade

IL · Tel Aviv, Israel

SQL Processing Unit (SPU) for analytics acceleration.

Flexxon

SG · Singapore

World's first AI-embedded cybersecurity SSD (X-PHY) with real-time threat detection at firmware level.

deepsilicon

US · San Francisco, CA

Custom silicon for ternary transformer models. 8x compression through 2-bit weights, 5x less RAM, up to 20x faster inference.

Systel

US · Sugar Land, TX

Rugged mission-critical computing since 1988. 8000+ systems deployed for US military airborne missions.

EdgeCortix

JP · Tokyo, Japan

SAKURA-II and upcoming SAKURA-X (2,000 TOPS for generative/physical AI). $110M+ oversubscribed Series B (Nov 2025, TDK Ventures/Jane Street/CDIB). NEDO ¥3B grant. Edge AI for defense, robotics, automotive.

Enfabrica

US · San Jose, CA

ACF-S SuperNIC. 3.2 Tbps, designed for 500K GPU clusters.

Retym

US · San Jose, CA

Programmable coherent DSP solutions for cloud and AI infrastructure. High-performance signal processing.

Etron Technology

TW · Hsinchu, Taiwan

Fabless IC design pioneering KGDM bare-die memories. JV with Nanya for ultra-high-bandwidth AI memory.

Aitech Defense Systems

IL · Kfar Saba, Israel

Rugged embedded computing since 1983. A230 Vortex AI supercomputer with NVIDIA Jetson AGX Orin (275 TOPS).

Microsoft (Azure Maia)

US · Redmond, WA

Azure Maia 200 (Jan 2026): 3nm TSMC, 216GB HBM3e at 7 TB/s, 272MB on-chip SRAM, native FP4/FP8. Claims 3× Amazon Trainium performance. Deployed in US Central (Iowa) datacenter.

Crystal Group

US · Hiawatha, IA

Employee-owned rugged computing. 20,000+ servers in US Navy operations. IP68-rated AI edge processors.

Oxmiq Labs

US · San Jose, CA

GPU hardware IP and software. OxCore (RISC-V GPU IP), OxPython (run CUDA apps on non-NVIDIA hardware without modification), Capsule (GPU container technology), OxQuilt (chiplet design). Founded by Raja Koduri (ex-Intel, AMD, Apple). $20M seed from MediaTek.

Thunder Compute

US · Atlanta, GA

GPU virtualization over TCP - the VMware for GPUs. 6x efficiency through oversubscription. A100s from $0.66/hr.

SiEngine Technology

CN · Wuhan, China

Geely/Arm China JV developing 7nm autonomous driving chips with 512 TOPS NPU computing power.

Together AI

US · San Francisco, CA

Open AI cloud platform. Custom GPU clusters, model training.

Esperanto Technologies

US · Mountain View, CA

RISC-V AI chips. $115M raised (2025), 7nm process.