GINC Critical Technology
Artificial Intelligence

Machine Learning & Foundation Models

The capacity to train, tune and field machine learning at the frontier, foundation and general-purpose models, and the model families across language, vision and science built on them.

Capability group
Artificial Intelligence
Domain
Critical Technology · Hard Power
Identifier
factbase.org/cap/AH3D4HX51F
Reference
Machine learning Wikipedia

Papers

680k

2020–2024

Paper growth

+268%

2020–24 on 2015–19

Patent families

506.7k

2020–2024

Filed abroad

15%

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 model development as capacity; frontier compute access reads at the control layer through the semiconductor group, and the research workforce through Human Capital, with field position read here.

Lineage

Carried from v0.0 (formerly Machine learning (including neural networks and deep learning)) with the same identifier, absorbing Natural language processing (retired): the field consolidated around general-purpose models, and a taxonomy that held language apart recorded the discipline as it was, not the capability as it is.

Indicators

  • research publication share in machine learningResearch
  • notable frontier models producedPending
  • highly cited AI researchersPending
  • private AI investmentPending

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 Machine learning (including neural networks and deep learning) and Natural language processing (including speech and text recognition and analysis).

Papers by year

The five leading nations, papers by year

Share of papers, 2020–2024

Of papers with a known author nation; 37% have none and are left out.

  1. 1China20.6%106,094 papers
  2. 2United States13.5%69,574 papers
  3. 3India13.0%66,831 papers
  4. 4United Kingdom4.0%20,488 papers
  5. 5Germany2.9%14,867 papers
  6. 6Canada2.3%11,935 papers
  7. 7Indonesia2.2%11,342 papers
  8. 8South Korea2.1%10,794 papers
  9. 9Australia2.0%10,411 papers
  10. 10Japan1.9%10,002 papers

Leading institutions, 2020–2024

By fractional count: each paper divided among its authors' institutions.

  1. 1Vellore Institute of Technology University2,175fractional papers
  2. 2SRM Institute of Science and Technology1,828fractional papers
  3. 3Tsinghua University1,736fractional papers
  4. 4Peking University1,638fractional papers
  5. 5Chitkara University1,482fractional papers
  6. 6Amrita Vishwa Vidyapeetham1,398fractional papers
  7. 7University of Electronic Science and Technology of China1,265fractional papers
  8. 8Shanghai Jiao Tong University1,241fractional papers
  9. 9Zhejiang University1,177fractional papers
  10. 10Chinese Academy of Sciences1,150fractional 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

38,535 applicants filed here in 2015–2024; the five largest held 8% of families (HHI 27 of 10,000: unconcentrated).

Share of families by origin, 2020–2024

  1. 1China75.1%380,541 families
  2. 2South Korea8.9%45,029 families
  3. 3United States8.5%43,322 families
  4. 4Japan2.3%11,519 families
  5. 5Germany1.2%6,177 families
  6. 6Taiwan0.5%2,321 families
  7. 7France0.4%1,954 families
  8. 8Canada0.4%1,912 families
  9. 9India0.3%1,638 families
  10. 10United Kingdom0.3%1,618 families

Applicants filing most internationally, 2015–2024

Families filed in more than one office, by applicant as the offices name them.

  1. 1SAMSUNG ELECTRONICS CO LTD4,627international families
  2. 2HUAWEI TECH CO LTD2,552international families
  3. 3MICROSOFT TECHNOLOGY LICENSING LLC2,223international families
  4. 4GOOGLE LLC1,926international families
  5. 5BOSCH GMBH ROBERT1,913international families
  6. 6FUJITSU LTD1,440international families
  7. 7NEC CORP1,429international families
  8. 8IBM1,344international families
  9. 9QUALCOMM INC1,218international families
  10. 10INTEL CORP1,176international families

Classification read

  • G06N3/02 Neural networks: architectures and learning methods · 2.8m documents
  • G06N20/00 Machine learning · 481.4k documents
  • G06N7/01 Probabilistic graphical models · 80.3k documents

Shares inventions with

Capabilities whose classes the same families also fall in, by overlap (Jaccard).

Companies

50 companies in this capability, across 8 regions. The leading 12:

  1. 1OpenAIDeveloper of the GPT and o-series frontier models and ChatGPT, the most widely used general-purpose foundation model product.Research
  2. 2Alphabet (Google DeepMind)Google DeepMind builds the Gemini frontier model family and scientific models such as AlphaFold, trained on Google's own TPU infrastructure.Research
  3. 3AnthropicFrontier lab that develops the Claude family of large language models, with a focus on safety research.Research
  4. 4Meta PlatformsReleases the open-weight Llama model family and runs FAIR and Meta Superintelligence Labs, plus PyTorch, the dominant ML framework.Research
  5. 5MicrosoftLargest investor in and cloud host for OpenAI, operator of Azure AI supercomputers, and developer of its own Phi and MAI models.Operation
  6. 6NVIDIASupplies most of the GPUs and the CUDA software stack used to train foundation models worldwide, and publishes its own Nemotron models.Equipment & materials
  7. 7DeepSeekDeveloper of the open-weight DeepSeek-V3 and R1 models, which reached near-frontier performance at low training cost; backed by the hedge fund High-Flyer.Research
  8. 8Alibaba GroupBuilds the Qwen model family, one of the most widely used open-weight model families globally, and runs China's largest AI cloud.Research
  9. 9xAIDevelops the Grok frontier models, trained on its Colossus supercomputer in Memphis; merged into SpaceX in 2026.Research
  10. 10ByteDanceIts Seed team builds the Doubao models behind China's most used consumer chatbot, and it is among China's largest buyers of AI compute.Research
  11. 11Amazon (AWS)AWS provides model training and hosting (Bedrock, Trainium chips), is Anthropic's main investor and cloud partner, and builds its own Nova models.Operation
  12. 12Mistral AIEurope's leading frontier model developer, producing open and commercial models (Mistral Large, Magistral, Codestral); ASML is a major investor.Research
All 50 in the Companies table →