China AI DataTokenHeart News ↗
← Back to dashboardDATA DETAIL / TRAINING

Training compute growth

Maximum disclosed training compute per year on a log10(FLOP) scale. Missing values remain absent; the current year is incomplete.

Training compute growth

Training compute growth2012-01-01: 17.78log10(FLOP); 2013-01-01: 18.42log10(FLOP); 2014-01-01: 20.47log10(FLOP); 2015-01-01: 20.58log10(FLOP); 2016-01-01: 21.82log10(FLOP); 2017-01-01: 20.93log10(FLOP); 2018-01-01: 21.94log10(FLOP); 2019-01-01: 23.03log10(FLOP); 2020-01-01: 23.50log10(FLOP); 2021-01-01: 24.31log10(FLOP); 2022-01-01: 24.44log10(FLOP); 2023-01-01: 25.70log10(FLOP); 2024-01-01: 25.58log10(FLOP); 2025-01-01: 26.70log10(FLOP); 2026-01-01: 27.00log10(FLOP)013.5272012-01-01: 17.78 log10(FLOP)2013-01-01: 18.42 log10(FLOP)2014-01-01: 20.47 log10(FLOP)2015-01-01: 20.58 log10(FLOP)2016-01-01: 21.82 log10(FLOP)2017-01-01: 20.93 log10(FLOP)2018-01-01: 21.94 log10(FLOP)2019-01-01: 23.03 log10(FLOP)2020-01-01: 23.50 log10(FLOP)2021-01-01: 24.31 log10(FLOP)2022-01-01: 24.44 log10(FLOP)2023-01-01: 25.70 log10(FLOP)2024-01-01: 25.58 log10(FLOP)2025-01-01: 26.70 log10(FLOP)2026-01-01: 27.00 log10(FLOP)01/0101/01 UTC

Total 468 · Page 17/19 · Observed 9/25/2026, 3:30:45 AM

NameGroupReported valueDateNotesSource
AlphaX-1Facebook AI Research,Brown University889,344,000,000,000,000 FLOP2019-10-02UnreleasedSource ↗
WeNet (Penn Treebank)Amazon730,000,001,000,000,000 FLOP2019-04-08UnreleasedSource ↗
R-FCNTsinghua University,Microsoft Research719,357,760,000,000,000 FLOP2016-06-21—Source ↗
QRNNSalesforce Research688,664,720,000,000,000 FLOP2018-02-01UnreleasedSource ↗
(ensemble): AWD-LSTM-DOC (fin) × 5 (WT2)NTT Communication Science Laboratories,Tohoku University666,000,000,000,000,000 FLOP2018-08-30Open weights (unrestricted)Source ↗
UDSMProtFraunhofer Heinrich Hertz Institute637,000,000,000,000,000 FLOP2019-09-04Open weights (unrestricted)Source ↗
Unsupervised High-level Feature LearnerGoogle600,000,000,000,000,000 FLOP2012-07-12—Source ↗
4 layer QRNN (h=2500)Salesforce Research591,588,150,000,000,000 FLOP2018-03-22UnreleasedSource ↗
DeepLocTechnical University of Denmark,University of Copenhagen578,102,400,000,000,000 FLOP2017-07-07Hosted access (no API)Source ↗
CT-MoS (WT2)Google,National Tsing Hua University540,000,000,000,000,000 FLOP2020-12-25UnreleasedSource ↗
Visualizing CNNsNew York University (NYU)532,000,000,000,000,000 FLOP2013-11-12—Source ↗
GANsUniversity of Montreal / Université de Montréal518,400,000,000,000,000 FLOP2014-06-10—Source ↗
DCNNUniversity of Maryland,Rutgers University480,986,910,000,000,000 FLOP2015-08-07UnreleasedSource ↗
DNN EM segmentationIDSIA,SUPSI478,000,000,000,000,000 FLOP2012-12-03UnreleasedSource ↗
AlexNetUniversity of Toronto470,000,000,000,000,000 FLOP2012-09-30—Source ↗
GL-LWGC-AWD-MoS-LSTM + dynamic evaluation (WT2)Ben-Gurion University of the Negev456,000,000,000,000,000 FLOP2017-08-29UnreleasedSource ↗
Big-Little Net (speech)IBM429,004,800,000,000,000 FLOP2018-07-10Open weights (unrestricted)Source ↗
AWD-LSTM-DRILL + dynamic evaluation† (WT2)IDIAP408,000,000,000,000,000 FLOP2019-05-14Open weights (unrestricted)Source ↗
AWD-LSTM + MoS + Partial ShuffledUniversity of Texas at Austin315,000,000,000,000,000 FLOP2019-06-10Open weights (non-commercial)Source ↗
DistBelief SpeechGoogle311,400,000,000,000,000 FLOP2012-12-03—Source ↗
Fraternal dropout + AWD-LSTM 3-layer (WT2)Jagiellonian University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),University of Montreal / Université de Montréal306,000,000,000,000,000 FLOP2017-10-31UnreleasedSource ↗
AWD-LSTM - 3-layer LSTM (tied) + continuous cache pointer (WT2)Salesforce Research297,000,000,000,000,000 FLOP2017-08-07UnreleasedSource ↗
Dropout (ImageNet)University of Toronto273,196,800,000,000,000 FLOP2012-06-03UnreleasedSource ↗
Big-Little NetIBM246,048,000,000,000,000 FLOP2018-07-10Open weights (unrestricted)Source ↗
Hierarchical Scene Labeling (Stanford Background)New York University (NYU)237,746,880,000,000,000 FLOP2013-08-01UnreleasedSource ↗

Data: Epoch AI · CC BY · View source · Automatically synced; not an exhaustive global sample