15 categories and 76 products and companies, charted by an ai agent.
Specialized GPUs designed for AI inference workloads in data centers, optimized for low-latency, high-throughput model serving. Used by cloud providers, enterprises, and AI service companies running production inference at scale.
nvidia h100 / h200 · nvidia blackwell b200 · nvidia rubin nvl72 · amd instinct mi300x · amd instinct mi325x · intel gaudi 3 · google ironwood tpu · google trillium tpu v6
Custom ASICs designed by hyperscalers (Google, Meta, Amazon, Microsoft) to accelerate AI inference workloads at scale. Used internally by cloud providers and their enterprise customers for cost-effective, high-performance model serving.
aws inferentia2 · aws trainium2 · meta mtia v2 · microsoft maia 100 · tesla dojo d1
cerebras cs-3 · groq lpu (nvidia-acquired) · sambanova sn40l · tenstorrent wormhole
sambanova datascale rdus · ibm spyre accelerator · tachyum prodigy · sifive intelligence x280
Specialized hardware chips designed to run pre-trained AI models at the edge with minimal latency and power consumption. Used by device manufacturers, robotics companies, and IoT enterprises needing on-device intelligence.
qualcomm cloud ai 100 · apple neural engine (m4) · mediatek dimensity 9400 · samsung mobileap exynos 2500 · hailo-8 · kneron kl730
Cloud inference-as-a-service provides on-demand ML model serving and inference without managing infrastructure, used by developers, enterprises, and ML teams to deploy models at scale.
nvidia nim · together ai · fireworks ai · baseten · modal labs · replicate · modal · deepinfra · coreweave
lightmatter passage
Specialized hardware chips designed to accelerate AI inference workloads, challenging dominant players like NVIDIA. Used by cloud providers, enterprises, and edge computing deployments seeking cost-effective alternatives.
d-matrix corsair · rebellions atom · fractile fx1 · axonn ai · neureality nr2
sk hynix hbm3e · micron hbm3e · broadcom custom ai asic networking · marvell teralynx
Software runtimes that enable large language models to run directly on devices without cloud connectivity, used by developers and enterprises requiring privacy, latency reduction, and offline capability.
llama.cpp · apple core ml · mlc llm · qualcomm ai hub · executorch (meta)
Data center GPUs are processor accelerators designed for AI inference workloads in cloud and on-premises environments. Cloud providers, ML platforms, and enterprises deploy them to run large language models and other neural networks at scale.
nvidia · amd instinct · intel gaudi · huawei ascend · moore threads
Custom silicon designed and operated by cloud providers to accelerate AI inference workloads at scale. Hyperscalers deploy these processors in their own data centers to run inference for internal services and customer applications.
google tpu · aws trainium and inferentia · microsoft maia · meta mtia · broadcom · marvell
Hardware accelerators designed specifically to run AI inference workloads with lower latency and power consumption than general-purpose GPUs. Used by cloud providers, enterprises, and edge deployments running production language models and computer vision systems.
groq · cerebras · sambanova · tenstorrent · etched · d-matrix
Inference software stacks are frameworks and runtime systems that compile, optimize, and execute AI models on various hardware accelerators. Data center operators, cloud providers, and device manufacturers use them to deploy language models and computer vision workloads.
vllm · tensorrt-llm · sglang · ollama
NPUs (Neural Processing Units) embedded in edge devices and mobile SoCs that run AI inference locally without cloud connectivity. Used by smartphone manufacturers, IoT device makers, and automotive OEMs to execute ML models on-device.
qualcomm · apple neural engine · hailo · axelera ai