AI Hardware & Chips Companies

NVIDIA

Santa Clara, United States

NVIDIA is the world leader in AI computing hardware and software. Creator of CUDA, cuDNN, and the dominant GPU platform for AI training and inference.

AMD

Santa Clara, United States

AMD designs high-performance CPUs and GPUs, and through its Instinct accelerators and ROCm software is a major supplier of hardware for AI training and inference, competing with NVIDIA in the data-center market.

Intel

Santa Clara, United States

Intel designs and manufactures CPUs and AI accelerators, including its Gaudi line for deep-learning training and inference and AI capabilities integrated across its Core and Xeon processors.

Cambricon

Beijing, China

Cambricon is a Chinese fabless semiconductor company specializing in the design and development of Neural Processing Units (NPUs) for both cloud and edge computing applications. Their core product is a series of AI chips intended to accelerate machine learning workloads in servers, smart terminals, and robotics. Cambricon targets the growing demand for on-device AI processing and integrated end-to-cloud AI solutions within the Chinese market and beyond.

Vicor

Andover, United States

Vicor designs and manufactures modular power systems, specializing in high-density power conversion for demanding applications. Their core technology focuses on optimized 48V power delivery networks, crucial for efficient operation of AI data centers and GPU computing infrastructure. Vicor targets companies building high-performance computing systems, offering solutions that improve power efficiency and scalability in areas like AI, eMobility, and high-performance computing.

Horizon Robotics

Beijing, China

Horizon Robotics is a China-based AI hardware company specializing in edge AI processing units. They design and manufacture automotive-grade AI chips – the Journey series – specifically for advanced driver-assistance systems (ADAS) and autonomous driving capabilities. Targeting automotive OEMs and Tier 1 suppliers, Horizon Robotics provides a full-stack solution enabling on-device AI processing for improved safety and efficiency in vehicles.

Crusoe

San Francisco, United States

Crusoe builds renewable-powered, scalable AI infrastructure and cloud compute solutions. Their core offering centers around a proprietary inference engine, leveraging “MemoryAlloy” technology, designed to optimize performance for large-context AI workloads with low latency. Targeting enterprises deploying demanding AI applications, Crusoe provides both cloud-based services (Crusoe Cloud, Serverless Fine-Tuning, Managed Inference) and dedicated hardware to accelerate AI development and deployment.

SambaNova Systems

Palo Alto, United States

SambaNova Systems develops a full-stack AI platform, including DataScale processors (RDUs) and the Samba-1 model suite, designed to accelerate AI inference and fine-tuning. The company offers both cloud-based (SambaCloud) and on-premise (SambaStack) deployment options, targeting enterprises and governments with demanding data security and performance requirements. SambaNova positions itself as a high-performance, energy-efficient alternative to GPU-based AI infrastructure, particularly for large language models and sovereign AI initiatives.

Sila Nanotechnologies

Alameda, United States

Sila Nanotechnologies develops and manufactures silicon anode materials that significantly increase the energy density and performance of lithium-ion batteries. Their core product, Titan Silicon, is a drop-in replacement for traditional graphite anodes, delivering up to 20% more energy density and faster charging capabilities. Sila targets battery manufacturers and device companies across electric vehicle, consumer electronics, and industrial sectors, offering both materials supply and battery engineering services to optimize cell performance.

Black Sesame Technologies is a Chinese fabless semiconductor company specializing in high-performance, low-power AI chips for automotive applications. Their core product is the Huashan series of Systems-on-Chips (SoCs), which leverage advanced image processing and adaptive light control sensing technology to enhance autonomous driving capabilities. Black Sesame targets automotive OEMs and Tier 1 suppliers requiring customized and comprehensive AI-powered vision solutions for ADAS and autonomous vehicles.

Cerebras Systems

Sunnyvale, United States

Cerebras Systems develops AI hardware, specifically the Wafer Scale Engine, a large-scale chip designed to accelerate deep learning workloads. Unlike traditional GPUs, Cerebras’ technology aims to significantly reduce the time and cost associated with training complex AI models. Their target market is organizations requiring high-performance computing for demanding AI applications, such as large language models and scientific computing.

Graphcore

Bristol, United Kingdom

Graphcore designs Intelligence Processing Units (IPUs), processors with a novel architecture built for machine-intelligence workloads. The UK company was acquired by SoftBank Group in July 2024 and continues to operate under the Graphcore name as a wholly owned subsidiary.

DriveNets

Ra'anana, Israel

DriveNets delivers high-performance Ethernet-based network infrastructure solutions optimized for demanding AI workloads and large-scale cloud deployments. Their core technology is a disaggregated networking operating system enabling service providers and cloud builders to create scalable and open routing fabrics. DriveNets specifically targets organizations requiring high-bandwidth, low-latency networking to support AI training and inference, as well as next-generation service provider networks.

Scaleway

Paris, France

Scaleway is a European cloud infrastructure provider specializing in GPU-accelerated computing for AI and machine learning workloads. Their core offering is the Scaleway AI Supercomputer – currently featuring the Nabu-2023 and expanding to 127 DGX nodes – providing scalable resources for model training and deployment. Scaleway differentiates itself by offering a sovereign European cloud solution with predictable pricing, robust data security, and 24/7 support, targeting organizations prioritizing data residency and cost transparency.

Lightmatter

Boston, United States

Lightmatter develops photonic integrated circuits and co-packaged optics (CPO) designed to accelerate AI workloads. Their core innovation lies in the Passage family of products – including the Passage M1000 3D Photonic Superchip and L200 optical engines – which utilize a vertically integrated electronic-photonic architecture to deliver industry-leading bandwidth density and efficiency for data movement between GPUs, TPUs, and data center switches. Targeting large-scale AI model training and inference, Lightmatter recently demonstrated a world-first 16-wavelength bidirectional optical DWDM link on a single fiber, achieving an 8x increase in bidirectional fiber bandwidth density and positioning them as a leader in next-generation AI infrastructure.

Groq

Mountain View, United States

Groq designs the Language Processing Unit (LPU), a novel processor architecture specifically engineered for high-performance, low-latency language model inference. Pioneered in 2016, the LPU fundamentally differs from traditional GPUs by utilizing a deterministic, software-defined dataflow architecture to maximize throughput and predictability for demanding AI workloads. Groq offers its LPU hardware alongside GroqCloud, a fully managed inference platform, serving customers requiring real-time responses from large language models in applications like generative AI and natural language processing.

Celestial AI

Santa Clara, United States

Celestial AI develops Photonic Fabric, an optical interconnect technology that uses light to move data between chips and memory, targeting the bandwidth and energy bottlenecks of large-scale AI computing.

Tenstorrent

Toronto, Canada

Tenstorrent designs high-performance AI processors and ML infrastructure, currently offering the Grayskull™ chip based on a novel, scalable architecture. Their key innovation lies in the combination of this architecture with an open-source, MLIR-based compiler stack, TT-Forge™, enabling optimized workload deployment and broad software compatibility. Tenstorrent is targeting AI inference and training applications, and recently released TT-Forge™ into public beta to foster community development and accelerate model support for their hardware.

SiMa.ai

San Jose, United States

SiMa.ai is a fabless chip company delivering a software-centric machine-learning system-on-chip (MLSoC) for low-power edge devices, targeting the 5-25W range for computer vision and multimodal generative AI. Founded by former Groq COO Krishna Rangasayee, it has raised $270M.

Blaize

El Dorado Hills, United States

Blaize develops edge AI computing platforms and silicon utilizing their Graph Streaming Processor (GSP) architecture. Their platforms, including Blaize Pathfinder and Xplorer, offer a code-free software suite to simplify and accelerate AI application deployment from data center to edge devices. Blaize targets industries seeking to implement scalable, energy-efficient AI solutions for applications like computer vision, overcoming the limitations of traditional hardware and software approaches.

Rebellions

Seoul, South Korea

Rebellions is a South Korean AI hardware company specializing in high-performance accelerator cards, servers, and rack-scale solutions for data centers and edge computing. Their core product, the REBEL-Quad, utilizes a chiplet architecture and HBM3E memory to deliver industry-leading performance per watt for demanding AI workloads like those built on PyTorch and vLLM. Rebellions targets organizations requiring scalable and efficient AI infrastructure, offering a full-stack hardware and software solution.

Mythic

Austin, United States

Mythic is a US-based AI hardware company developing the Mythic AMP™ analog processor for edge AI applications. Their technology stores AI parameters directly within the processor, eliminating memory bottlenecks and delivering significantly improved power efficiency and performance compared to traditional digital architectures. Mythic targets industries like robotics, defense, and security, enabling advanced, real-time inference at the edge with reduced energy consumption.

d-Matrix

Santa Clara, United States

d-Matrix develops in-memory computing chips—specifically their “Corsair” platform—designed to accelerate Generative AI inference workloads. Their technology integrates memory and compute to deliver ultra-low latency and high throughput, addressing the growing energy consumption and cost challenges of AI deployment. d-Matrix targets enterprises and data centers seeking to efficiently scale Generative AI applications without compromising performance or sustainability.

Untether AI

Toronto, Canada

Untether AI develops specialized hardware accelerating AI inference through an innovative at-memory compute architecture. Their core product, the tsunAImi® accelerator card, significantly improves performance and energy efficiency for demanding AI workloads by integrating computation directly within memory. Targeting applications ranging from assisted driving to smart cities, Untether AI enables cost-effective deployment of AI at the edge and in data centers by supporting standard AI frameworks like TensorFlow and PyTorch.

Enfabrica

San Jose, United States

Enfabrica develops high-performance, low-latency Dataflow Architecture networking chips designed to accelerate AI workloads within data center infrastructure. Their innovative approach bypasses traditional von Neumann bottlenecks, enabling significantly faster data transfer and improved energy efficiency for demanding applications like large language models and generative AI. Recognized as a leading AI hardware innovator – featured in TechRadarPro’s “10 Hottest AI Hardware Companies to Follow in 2025” and a member of The Futuriom 50 – Enfabrica recently surpassed $100M in funding (TechCrunch, 2024), signaling strong investor confidence in their technology.

Etched

Mountain View, United States

Etched is a US-based AI hardware company developing custom silicon focused on accelerating inference workloads. Their core product is a novel, high-performance AI chip architecture designed to deliver leading performance per watt for demanding AI applications. Founded in 2022, Etched targets organizations requiring substantial on-premise AI inference capabilities, positioning themselves for the future of large-scale AI deployment.

EdgeCortix

Tokyo, Japan

EdgeCortix is a Tokyo-based fabless semiconductor company designing energy-efficient AI processors for the edge with a software-first architecture. Its SAKURA accelerators deliver high AI throughput at sub-10-watt power and have received substantial NEDO R&D backing.

Recogni

San Jose, United States

Recogni, now operating as Tensordyne, develops AI inference systems—including custom silicon and optimized math—designed to significantly reduce the computational cost and energy consumption of large AI models. Their technology targets data centers requiring high-throughput, low-latency AI processing for generative AI and other demanding applications. Tensordyne differentiates itself through fundamental mathematical innovation in AI, enabling greater density and efficiency in AI infrastructure.

DEEPX

Seoul, South Korea

DEEPX is a South Korean fabless semiconductor company developing low-power AI chips for on-device inference in robotics, cameras, and consumer electronics.

Kneron

San Diego, United States

Kneron designs and manufactures neural processing units (NPUs) optimized for edge AI applications. Their SoCs enable on-device AI inference, reducing latency and enhancing data privacy compared to cloud-based solutions. Kneron targets smart home, automotive, and industrial IoT markets requiring low-power, high-performance AI capabilities directly within their devices.

Syntiant

Irvine, United States

Syntiant develops neural decision processors that enable on-device AI processing for ultra-low-power applications. Their core technology utilizes at-memory compute to deliver significantly improved efficiency and throughput compared to traditional microcontrollers when running deep learning models. Syntiant targets the mobile, earbud, and IoT markets, providing scalable hardware solutions for applications like voice control, always-on audio recognition, and vibration analysis directly on the edge device.

Sapeon

Seoul, South Korea

Sapeon develops neural processing units (NPUs) specifically designed to accelerate AI inference workloads in data centers. Their technology focuses on delivering high performance and energy efficiency for applications like image recognition and natural language processing. Sapeon targets cloud service providers and enterprises seeking to optimize the cost and performance of their AI-powered services.

Digua Robotics

Beijing, China

Digua Robotics (D-Robotics), spun out of Hong Kong-listed Horizon Robotics in 2024, provides development infrastructure and computing chips for consumer and industrial robotics, supporting China's push into embodied AI. It raised a $100M Series A.

Origin AI

San Francisco, United States

Origin AI develops WiFi-based motion sensing technology for indoor spatial understanding and home monitoring. Their core product utilizes radio frequency (RF) signals to detect and analyze movement within a home without requiring cameras or additional hardware. This technology targets the home security and smart home automation markets, offering privacy-focused activity detection and potential for advanced features like fall detection and occupancy-based energy savings.

Lambda Labs

San Francisco, United States

Lambda Labs is a US-based provider of GPU cloud infrastructure and hardware specifically designed for demanding deep learning workflows. They offer on-demand access to high-performance GPUs, including the NVIDIA B200 and H100, through cloud instances, private cloud deployments, and dedicated workstations. Lambda Labs targets AI researchers and organizations requiring scalable and cost-effective solutions for model training and inference.

AIStorm

San Jose, United States

AIStorm develops charge-domain computing chips and IP that deliver AI, memory, and digital processing directly in silicon—up to 100× more efficient.

Limitless

San Francisco, United States

Limitless develops AI-powered, edge-computing hardware and software solutions focused on delivering personalized AI assistants directly on devices. Their core technology centers around a custom silicon chip designed for on-device large language model (LLM) inference, enabling private and responsive AI experiences without cloud connectivity. Targeting enterprise and automotive sectors, Limitless recently demonstrated a fully functional, on-device generative AI assistant integrated into a vehicle’s infotainment system.

GEMESYS

Bochum, Germany

GEMESYS, a spin-off from Ruhr-University Bochum, develops analog memristor-based AI chips that use ReRAM crossbar arrays for fully analog neural-network processing, enabling on-device training and inference for power-constrained edge hardware. It raised an 8.6M-euro pre-seed in 2024.

Huawei AI

Shenzhen, China

Huawei develops the Ascend series of AI chips – including the Ascend 910 and NPU-based modules – designed to accelerate machine learning workloads for edge and cloud deployments. Their AI capabilities are demonstrated through the Pangu large language model series, which includes models like Pangu-α and Pangu-Weather, showcasing advancements in natural language processing and weather forecasting accuracy. Primarily serving the telecommunications, manufacturing, and financial services sectors, Huawei AI has achieved notable deployments of its solutions in smart city initiatives and industrial automation across China and internationally.

Ambient Scientific

Santa Clara, United States

Ambient Scientific is a fabless semiconductor company making ultra-low-power edge-AI processors. Its GPX10 chip, built on the analog/digital DigAn architecture, runs always-on AI on battery-powered devices and entered volume production in 2024. Founded in 2017 by Gajendra (GP) Singh.

Axelera AI

Eindhoven, Netherlands

Axelera AI builds the Metis AI processing platform for edge inference (computer vision, robotics, industrial), delivering up to ~214 INT8 TOPS at high efficiency. Headquartered in Eindhoven, founded 2021; raised a $68M Series B in June 2024, bringing total funding to about $120M.

Blumind

Toronto, Canada

Blumind designs ultra-low-power analog AI chips for always-on edge applications (wearables, IoT, medical devices), claiming 100–1,000x energy savings versus conventional AI silicon, starting with its BM110 and BM210 processors. Founded in 2020 in Toronto by Niraj Mathur and John Gosson; raised a C$20M Series A in 2025.

Supermicro

San Jose, United States

Supermicro designs and manufactures server and storage hardware specifically optimized for demanding artificial intelligence and GPU-accelerated workloads. Their product line focuses on high-performance, modular systems engineered to maximize GPU density and efficiency within data center environments. This positions Supermicro as a key infrastructure provider for organizations deploying AI applications in areas like machine learning, deep learning, and high-performance computing.

Synopsys

Sunnyvale, United States

Synopsys develops Electronic Design Automation (EDA) tools and Semiconductor IP used in the creation of complex integrated circuits. Their solutions increasingly leverage AI, particularly through collaborations with companies like NVIDIA, to accelerate chip design, verification, and heterogeneous integration – crucial for next-generation AI hardware. Synopsys targets companies developing advanced semiconductors for diverse markets including automotive, high-performance computing, and mobile applications, enabling faster time-to-market and improved silicon success rates.

Wolf Speed

Durham, United States

Wolfspeed is a U.S.-based semiconductor company specializing in silicon carbide (SiC) power devices and materials. They manufacture SiC MOSFETs and modules, including 200mm wafers, designed to improve the efficiency and performance of power electronics systems. Wolfspeed targets industries requiring high-performance power solutions, such as electric vehicles, renewable energy, and industrial power supplies, with a focus on enabling scalable and durable designs.

Habana Labs

Caesarea, Israel

Habana Labs, an Intel company, designs high-performance AI processors specifically for deep learning workloads. Their primary product is the Gaudi® series of accelerators, engineered to efficiently handle both training and inference for large-scale generative AI models. Habana Labs targets data center and cloud service providers requiring optimized compute for demanding AI applications like autonomous vehicles and advanced AI deployments.

Qualcomm AI

San Diego, United States

Qualcomm AI develops specialized semiconductor solutions that integrate AI processing directly onto mobile devices and edge computing infrastructure. Their core technology centers on AI engines embedded within Snapdragon platforms and dedicated edge AI compute, enabling on-device machine learning capabilities. This positions Qualcomm as a key enabler for applications requiring low-latency, power-efficient AI processing in areas like smartphones, automotive, and IoT devices.

BrainChip

Sydney, Australia

BrainChip develops and manufactures Akida, a neuromorphic computing processor IP designed for edge AI applications. This technology mimics the human brain to achieve ultra-low power consumption and efficient on-device processing of sensor data, eliminating reliance on cloud connectivity. BrainChip targets developers of edge AI solutions in markets like automotive, industrial automation, and smart devices where low latency and energy efficiency are critical.

Rain AI

San Francisco, United States

Rain AI develops novel AI hardware, specifically neuromorphic computing chips, designed to significantly reduce energy consumption for AI workloads. Their technology focuses on event-driven processing, mimicking biological neural networks to achieve greater efficiency than traditional architectures. The company targets data center operators and AI infrastructure providers seeking to lower operational costs and improve sustainability.

Esperanto Technologies

Mountain View, United States

Esperanto Technologies developed high-performance, low-power AI and HPC accelerators based on a massively parallel architecture of over a thousand custom RISC-V cores. Their ET-SoC-1 chip utilizes both high-performance, out-of-order cores and energy-efficient, in-order cores to deliver superior compute efficiency for demanding workloads like generative AI and HPC. While the IP has been acquired by Nekko.ai, Esperanto focused on providing a lower total cost of ownership through RISC-V based solutions for data center applications.

Sesame AI

San Francisco, United States

Sesame AI develops specialized AI hardware and software focused on creating highly realistic and responsive voice-based conversational AI. Their core product is a novel AI chip architecture designed to efficiently run large language models for natural language processing, specifically prioritizing low-latency voice interaction. Currently in research preview, Sesame AI targets the emerging market for on-device, lifelike AI assistants and aims to advance the field of human-computer interaction.

Friend

New York, United States

Friend is developing a socially intelligent AI companion embodied in a robotic device. Their core technology centers around a large multimodal model (LMM) designed for natural and empathetic conversational interaction, coupled with advanced robotics for physical presence. Targeting consumers seeking companionship and assistance, Friend recently secured $15 million in seed funding led by Andreessen Horowitz to further develop this hardware and AI platform.

Positron

Reno, United States

Positron develops dedicated AI inference hardware designed to accelerate large language model (LLM) deployment. Their core product is a production-ready appliance powered by their custom Asimov chips, capable of supporting models up to 500 billion parameters with over 8TB of memory. Targeting organizations requiring high-performance, scalable generative AI infrastructure, Positron aims to optimize total cost of ownership for LLM inference.

MatX

Mountain View, United States

MatX develops high-performance AI accelerator chips specifically designed for large language model (LLM) workloads. Their core product, the “Onechip,” prioritizes both high throughput and low latency, offering performance advantages for both training and inference across diverse LLM applications including long-context support. The company has secured funding from prominent venture capital firms and is targeting leading AI “frontier labs” as its primary customer base.

Rivos

Santa Clara, United States

Rivos is developing novel, low-power AI inference chips based on a massively parallel, dataflow architecture. Their core technology aims to deliver significantly improved performance-per-watt compared to existing GPU and CPU-based solutions for AI workloads. While currently in stealth mode, Rivos is targeting edge computing applications requiring high efficiency, such as autonomous vehicles and mobile devices.

FuriosaAI

Seoul, South Korea

FuriosaAI develops neural processing units (NPUs) specializing in accelerating AI inference and training workloads within data centers. Their core product, the Renegade series of NPUs, is designed for high performance and efficiency across applications like computer vision, generative AI, and large language models. Notably, FuriosaAI aims to deliver a leading-edge NPU solution by 2026, targeting cloud and enterprise customers requiring substantial AI compute capabilities.

Moreh

Seoul, South Korea

Moreh develops full-stack inference software optimizing Large Language Model (LLM) deployment across diverse hardware accelerators. Their core technology focuses on custom kernels, routing, and auto-scaling to maximize performance on AMD GPUs, Tenstorrent chips, and heterogeneous GPU clusters. Founded in 2020, Moreh targets organizations requiring high-performance, efficient LLM inference at scale, particularly those leveraging non-NVIDIA hardware.

Ayar Labs

San Jose, United States

Ayar Labs develops in-chip optical I/O solutions to address bandwidth limitations in data-intensive computing. Their core technology utilizes silicon photonics to create high-speed, low-latency data transfer between chips and systems. Targeting high-performance computing, data centers, and eventually edge AI applications, Ayar Labs recently secured significant funding and demonstrated successful chiplet integration with Cerebras Systems’ Wafer Scale Engine 2.

ANORIA makes an AI-powered wearable bracelet that reads emotional state in real time, correlating ~150 audio and biometric signals via a neuroscience framework to surface a 'Flow Score' (energy, mood, focus) and coaching to help users train their emotional intelligence. Founded 2026 in San Francisco by Michael Belhassen (ex-Apple hardware designer); Y Combinator-backed.

Photoneo

Bratislava, Slovakia

Photoneo is a Slovak 3D machine-vision and robot-intelligence company known for its patented Parallel Structured Light technology and the MotionCam-3D camera, which captures 3D scenes in motion without blur. Its AI-powered vision-guided robotics solutions are used in automotive, logistics and manufacturing by customers such as General Motors and Volkswagen. It was founded by Jan Zizka and Tomas Kovacovsky and acquired by Zebra Technologies in 2025.