- Capability group
- Semiconductors & Advanced Computing
- Domain
- Critical Technology · Hard Power
- Identifier
- factbase.org/cap/EEFHJ9SJ2Q
- Reference
- High-performance computing Wikipedia
Papers
31.6k
2020–2024
Paper growth
+15%
2020–24 on 2015–19
Patent families
33.2k
2020–2024
Filed abroad
37%
families in more than one office
In the framework
What the capability covers and what it leaves to others, and the indicators it will be measured by.
Bounds
Covers computing capability as capacity; the centres as facilities belong to Strategic Infrastructure's data and compute capability, access to frontier compute reads at the deploy and control layers, and the models trained on it sit with Artificial Intelligence.
Lineage
Relocated in v0.1 from the connectivity group (formerly High performance computing) at the same level, carrying its identifier under the relocation convention, and absorbing the accelerator node (retired) from the AI group: hardware for computation is one capacity, and v0.0 held it in two groups, neither of them this chain.
Indicators
- top-ranked supercomputers heldPending
- installed AI accelerator computePending
- domestic accelerator design and productionPending
Research
Papers classified to the capability, by year and by the nation of their authors' institutions, 2020–2024. Read from the v0.0 research topics High performance computing and AI algorithms and hardware accelerators.
Papers by year
The five leading nations, papers by year
Share of papers, 2020–2024
Of papers with a known author nation; 32% have none and are left out.
- 1United States25.8%6,596 papers
- 2China14.9%3,819 papers
- 3Germany6.0%1,541 papers
- 4India5.1%1,318 papers
- 5United Kingdom4.3%1,105 papers
- 6South Korea3.6%909 papers
- 7Italy3.3%836 papers
- 8Spain3.2%820 papers
- 9Japan2.9%752 papers
- 10France2.7%703 papers
Leading institutions, 2020–2024
By fractional count: each paper divided among its authors' institutions.
- 1National University of Defense Technology246fractional papers
- 2Tsinghua University196fractional papers
- 3Oak Ridge National Laboratory150fractional papers
- 4Peking University129fractional papers
- 5Argonne National Laboratory120fractional papers
- 6Lawrence Berkeley National Laboratory114fractional papers
- 7Chinese Academy of Sciences103fractional papers
- 8University of Science and Technology of China98fractional papers
- 9Lawrence Livermore National Laboratory88fractional papers
- 10Shanghai Jiao Tong University86fractional papers
Patents
Patent families by the nation of origin of their applicants and their earliest priority year, from the CPC and IPC classes the capability is read from. 2025 onwards is incomplete: families are published 18 months after priority.
Families by priority year
The five leading origins, families by year
8,539 applicants filed here in 2015–2024; the five largest held 16% of families (HHI 75 of 10,000: unconcentrated).
Share of families by origin, 2020–2024
- 1China53.5%17,762 families
- 2United States21.4%7,086 families
- 3South Korea10.8%3,588 families
- 4Japan3.7%1,236 families
- 5Taiwan2.2%733 families
- 6United Kingdom2.1%713 families
- 7Germany1.4%453 families
- 8France0.8%250 families
- 9Singapore0.6%188 families
- 10Canada0.5%165 families
Applicants filing most internationally, 2015–2024
Families filed in more than one office, by applicant as the offices name them.
- 1INTEL CORP1,615international families
- 2SAMSUNG ELECTRONICS CO LTD1,593international families
- 3HUAWEI TECH CO LTD942international families
- 4QUALCOMM INC567international families
- 5NVIDIA CORP558international families
- 6MICROSOFT TECHNOLOGY LICENSING LLC473international families
- 7ADVANCED RISC MACH LTD376international families
- 8IBM373international families
- 9GOOGLE LLC371international families
- 10FUJITSU LTD351international families
Classification read
- G06N3/06 Hardware for neural networks, including AI accelerator chips · 89.9k documents
- G06F15/76 Processor architectures, including parallel arrays and systems on chip · 76.8k documents
- G06F15/173 Interconnection networks between processors · 28.9k documents
- G06T1/20 Graphics processor architectures · 53.5k documents
- G06F9/3885 Parallel execution units within processors · 43.1k documents
- G06F9/3877 Coprocessors and accelerators · 13.2k documents
Shares inventions with
Capabilities whose classes the same families also fall in, by overlap (Jaccard).
- Machine Learning & Foundation Models9,937 families · 0.083
- Applied AI & Advanced Analytics4,216 families · 0.019
- Semiconductor Design & EDA478 families · 0.016
- Protective Cybersecurity Technologies1,075 families · 0.011
- Quantum Computing258 families · 0.009
Companies
50 companies in this capability, across 6 regions. The leading 12:
- 1NVIDIADominant designer of AI accelerators (Hopper, Blackwell, Rubin GPUs), NVLink/InfiniBand interconnect and the CUDA software stack that power most AI training clusters and leading supercomputers.Design
- 2TSMCFabricates nearly all leading-edge AI accelerators and HPC processors (NVIDIA, AMD, Google TPU, AWS Trainium) and supplies the CoWoS advanced packaging that limits global AI chip output.Manufacture
- 3AMDDesigns EPYC CPUs and Instinct MI300/MI350 accelerators that power the exascale Frontier and El Capitan systems and a growing share of AI clusters.Design
- 4SK hynixLeading supplier of high-bandwidth memory (HBM3E/HBM4) stacked beside AI accelerators, the main memory constraint on AI compute.Equipment & materials
- 5BroadcomCo-designs custom AI accelerators (Google TPU and other hyperscaler XPUs) and supplies Tomahawk/Jericho Ethernet switching for AI clusters.Design
- 6Alphabet (Google DeepMind)Designs the Tensor Processing Unit (TPU, now Ironwood) family and operates some of the largest AI training fleets in the world through Google Cloud.Design
- 7Samsung ElectronicsMajor HBM and DRAM supplier and leading-edge foundry, and among the top patent filers in computing hardware.Manufacture
- 8HuaweiChina's leading AI accelerator designer (Ascend 910 series, CloudMatrix 384 supernode) and Kunpeng server CPUs, at the centre of China's domestic AI compute build-out.Design
- 9IntelSupplies Xeon CPUs used across most of the world's HPC installations, Gaudi AI accelerators and the Aurora exascale system, and is the top international patent applicant in this field.Design
- 10MicrosoftOperates among the largest AI supercomputers through Azure (including OpenAI's training clusters) and designs the Maia AI accelerator and Cobalt CPU.Operation
- 11Amazon (AWS)The largest cloud operator; it designs Trainium and Inferentia AI chips (Annapurna Labs) and Graviton CPUs and built Project Rainier for Anthropic.Operation
- 12Hewlett Packard EnterpriseThrough HPE Cray, the leading supercomputer integrator; it built the exascale Frontier, Aurora and El Capitan systems and Europe's LUMI and Alps.Integration