News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip

AI training and inference

Accelerators for training and running neural networks. 129 chips on EveryChip so far.

AI chips train neural networks and run them, a task called inference. Training a large language model takes thousands of accelerators running for weeks; inference runs every time someone uses the model. Nvidia's data centre GPUs have led since the deep-learning revival of 2012, while Google's TPU, AMD's Instinct, Amazon's Trainium and many start-up chips compete. Phones and laptops now carry small AI engines too.

Training and inference

Training adjusts billions of weights by passing data through the network and back, again and again, so it needs huge amounts of arithmetic and memory bandwidth and fast links between chips. Inference runs the finished model, often at lower precision, and is spread across data centres and devices. Many chip designs specialise in one or the other.

Why Nvidia led

In 2012 a neural network called AlexNet, trained on two Nvidia gaming GPUs, won the ImageNet image recognition contest by a wide margin. Nvidia's CUDA software, available since 2007, made its GPUs the default for researchers. Its A100 (2020), H100 (2022) and Blackwell (2024) chips became the main tools of the AI industry, and demand outran supply in 2023 and 2024.

Famous AI training and inference chips

Further reading

B200 🇺🇸 United StatesB200 AI GPU with 208 billion transistors for trillion-parameter modelsGPU · GPU architecture Nvidia H100 🇺🇸 United StatesNvidia H100 AI and HPC accelerator based on the Hopper architectureGPU · GPU architecture GB200 Grace Blackwell 🇺🇸 United StatesGB200 Grace Blackwell Superchip Nvidia AI superchip linking Grace CPUs and Blackwell GPUsSoC · Arm Nvidia L40S 🇺🇸 United StatesNvidia L40S Data centre GPU for AI and graphics workloadsGPU · GPU architecture · 2023 Nvidia H200 🇺🇸 United StatesNvidia H200 AI GPU with 141 GB of HBM3e memoryGPU · GPU architecture · 2023 GB300 Grace Blackwell 🇺🇸 United StatesGB300 Grace Blackwell Ultra Superchip AI superchip combining Grace CPUs and Blackwell Ultra GPUsSoC · Arm Nvidia Vera CPU 🇺🇸 United StatesNvidia Vera CPU Arm-based data centre CPU for agentic AI workloadsCPU · Arm Rubin CPX 🇺🇸 United StatesRubin CPX AI GPU for massive-context inference with GDDR7 memoryGPU · GPU architecture · 2026 MTIA v1 🇺🇸 United StatesMTIA v1 Meta’s first custom AI inference chip for recommendationsAI accelerator · RISC-V · 2020 Tesla V100 🇺🇸 United StatesTesla V100 Nvidia Volta GPU for AI and HPC data centresGPU · GPU architecture · 2017 Microsoft Cobalt 200 🇺🇸 United StatesMicrosoft Cobalt 200 132-core Arm server CPU for Microsoft Azure cloud workloadsCPU · Arm · 2025 Microsoft Azure Cobalt 🇺🇸 United StatesMicrosoft Azure Cobalt 100 128-core Arm processor for Microsoft cloud serversCPU · Arm · 2024 Maia 200 🇺🇸 United StatesMaia 200 AI inference accelerator on TSMC 3nm processAI accelerator · Tensor design · 2026 Maia 100 🇺🇸 United StatesMaia 100 Microsoft's first in-house AI accelerator chip for Azure cloud.AI accelerator · Other · 2023 Corsair 🇺🇸 United StatesCorsair AI inference accelerator with in-memory computeAI accelerator · Tensor design · 2024 Intel Data Center GPU 🇺🇸 United StatesIntel Data Center GPU Max Data centre GPU for HPC and AI workloadsGPU · GPU architecture Intel Gaudi 3 🇺🇸 United StatesIntel Gaudi 3 AI accelerator chip using Ethernet for scalingAI accelerator · Other · 2024 Inferentia2 🇺🇸 United StatesInferentia2 Second-generation AI inference accelerator from AWSAI accelerator · Other Gaudi 2 🇺🇸 United States and 1 moreGaudi 2 AI training accelerator chip by IntelAI accelerator · Other · 2022 Intel Gaudi 🇺🇸 United StatesIntel Gaudi AI accelerator series from Intel for training and inference workloads.AI accelerator · Other · 2024 Trainium3 🇺🇸 United StatesTrainium3 Fourth-generation AI accelerator chip from AWSAI accelerator · Other · 2025 AMD Instinct MI250X 🇺🇸 United StatesAMD Instinct MI250X Dual-GPU accelerator for supercomputersGPU · GPU architecture Ascend 910C 🇨🇳 ChinaAscend 910C AI accelerator chip by Huawei for data centresAI accelerator · Other ET-SoC-1 🇺🇸 United StatesET-SoC-1 RISC-V system on a chip for AI inferenceAI accelerator · RISC-V Trainium2 🇺🇸 United StatesTrainium2 AWS second-generation AI training accelerator chipAI accelerator · Other · 2024 NVDLA WorldNVDLA Open-source AI accelerator hardware by NvidiaAI accelerator · Other · 2017 🇺🇸 United StatesTrueNorth IBM neuromorphic chip mimicking brain structureAI accelerator · Other · 2014 Field-programmable obj 🇺🇸 United StatesField-programmable object array Programmable logic device bridging FPGA and ASIC gapsFPGA · Other · 2006 🇺🇸 United StatesHopper Nvidia datacentre GPU microarchitecture for AI and high-performance computing.GPU · GPU architecture · 2022 WorldElectrochemical RAM Non-volatile memory for analog deep learning accelerationMemory · Other 🇺🇸 United StatesAI Engine AMD hardware architecture for AI inference and signal processingAI accelerator · VLIW XDNA 🇺🇸 United StatesXDNA AMD microarchitecture for neural processing units in Ryzen AI chipsAI accelerator · VLIW · 2023 🇺🇸 United StatesBlackwell Nvidia GPU microarchitecture for AI and gamingGPU · GPU architecture · 2024

Photo credits on this page

The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

The weekly digest

Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.