Search
Professionals
23-05-11
OpenAI unveiled ChatGPT, a conversational AI on November 30, 2022. Within two months of its launch, ChatGPT has amassed over 100 million users, largely due to its innovative approach of moving away from “search” and embracing “conversation” with AI.
In Korea, the development of hyperscale AI models which serve as the foundation for conversational AI such as ChatGPT, is in full swing. EXAONE of LG boasts multimodality capabilities and can handle 300 billion parameters, allowing for the simultaneous processing of voice and image data. Hyperclova, released by NAVER, is a Korean-specialized AI language model with 204 billion parameters, while Kakao has released KoGPT, a Korean-specialized GPT-3 model, as well as minDALL-E and Karlo, image-generating models, to the open-source community.
The competition for global patents related to hyperscale AI technologies is getting fierce.
According to the Korean Intellectual Property Office, the number of patent applications relating to hyperscale AI filed with IP5 has increased by about 28 times in the past decade (2011 to 2020) with 14,848 applications filed in 2020. The average annual growth rate of applications over the past five years is even higher at 61.3%.
The majority of applications hail from the United States (15,035), China (13,103), Japan (4,906), South Korea (4,785) and Germany (1,138). South Korea (89.7%) and China (79.3%) show the highest average annual growth rates.
(Unit: number of patent applications)
‘11 | ‘12 | ‘13 | ‘14 | ‘15 | ‘16 | ‘17 | ‘18 | ‘19 | ‘20* | Total (Share) | |
US | 276 | 385 | 466 | 592 | 755 | 1,087 | 1,711 | 2,426 | 3,129 | 4,208 | 15,035 (35.6%) |
CN | 31 | 39 | 51 | 82 | 187 | 391 | 829 | 1,929 | 3,630 | 5,934 | 13,103 (31.0%) |
JP | 100 | 87 | 98 | 131 | 186 | 294 | 472 | 822 | 1,256 | 1,460 | 4,906 (11.6%) |
KR | 6 | 23 | 36 | 41 | 115 | 141 | 398 | 809 | 1,304 | 1,912 | 4,785 (11.3%) |
DE | 42 | 21 | 5 | 33 | 60 | 47 | 101 | 207 | 289 | 333 | 1,138 (2.7%) |
Others | 75 | 84 | 90 | 100 | 112 | 233 | 374 | 490 | 703 | 1,001 | 3,262 (7.7%) |
Total | 530 | 639 | 746 | 979 | 1,415 | 2,193 | 3,885 | 6,683 | 10,311 | 14,848 | 42,229 (100%) |
*Patent applications which had not been published by 2020 were excluded
Increase rate over last 10 years (‘11→’20, average annual increase rate) | Increase rate over last 5 years (‘16→’20, average annual increase rate) | |
US | 15.2 times, 35.4% | 3.9 times, 40.3% |
CN | 191.4 times, 79.3% | 15.2 times, 97.4% |
JP | 14.6 times, 34.7% | 5.0 times, 49.3% |
KR | 318.7 times, 89.7% | 13.6 times, 91.9% |
DE | 7.9 times, 25.9% | 7.1 times, 63.1% |
Others | 13.3 times, 33.4% | 4.3 times, 44.0% |
Total | 28.0 times, 44.8% | 6.8 times, 61.3% |
* Patent applications which had not been published by 2020 were excluded
The top applicants applying for hyperscale AI-related patents include global companies such as Samsung (1,213 applications), IBM (928 applications), Google (824 applications), Microsoft (731 applications), and Baidu (572 applications). Korean companies and research institutes such as LG (384 applications), STRADVISION (209 applications), Electronics and Telecommunications Research Institute (ETRI) (157 applications), KAIST (80 applications), Crowdworks (80 applications), NAVER (70 applications) and SK (64 applications) also feature prominently among the top applicants.
(Unit: number of patent applications)
Ranking | Nationality | Applicant Name | Sum | ‘11 | ‘12 | ‘13 | ‘14 | ‘15 | ‘16 | ‘17 | ‘18 | ‘19 | ‘20 |
1 | KR | SAMSUNG | 1,213 | - | 6 | 14 | 20 | 50 | 65 | 175 | 262 | 250 | 371 |
2 | US | IBM | 928 | 9 | 20 | 16 | 21 | 43 | 58 | 91 | 128 | 204 | 338 |
3 | US | 824 | 26 | 17 | 23 | 36 | 33 | 81 | 149 | 105 | 159 | 195 | |
4 | US | Microsoft | 731 | 24 | 20 | 28 | 60 | 79 | 64 | 90 | 101 | 108 | 157 |
5 | CN | Baidu | 572 | - | 1 | - | 1 | 21 | 42 | 45 | 86 | 116 | 260 |
6 | JP | Canon | 569 | 7 | 5 | 13 | 13 | 15 | 12 | 30 | 100 | 187 | 187 |
7 | CN | Tencent | 549 | - | - | - | 2 | 1 | 8 | 22 | 66 | 158 | 292 |
8 | DE | Simens | 484 | 8 | 9 | 2 | 11 | 49 | 22 | 57 | 129 | 94 | 103 |
9 | US | Intel | 478 | 7 | - | - | 5 | 14 | 19 | 60 | 74 | 84 | 215 |
10 | KR | LG | 384 | 1 | 2 | - | 3 | 37 | 9 | 16 | 27 | 196 | 93 |
25 | KR | STRADVISION | 209 | - | - | - | - | - | - | 5 | 20 | 68 | 116 |
36 | KR | ETRI | 157 | - | 1 | 1 | - | 3 | 7 | 20 | 33 | 32 | 60 |
66 | KR | KAIST | 80 | 1 | - | 3 | 2 | 4 | 5 | 12 | 24 | 9 | 20 |
66 | KR | Crowdworks | 80 | - | - | - | - | - | - | 1 | 3 | 5 | 71 |
76 | KR | NAVER | 70 | - | - | - | - | - | 1 | - | 13 | 28 | 28 |
85 | KR | SK | 64 | - | - | - | 1 | 3 | 5 | 10 | 10 | 14 | 21 |
* Patent applications which had not been published by 2020 were excluded
When analyzing the trend of hyperscale AI-related patent applications by technology, the majority (69.3%) are related to data generation, followed by learning models (25.8%), and specialized services (16.4%). Notably, there has been a rapid increase in patent application related to language models, which fall under the learning model technology.
It is projected that by 2024, the global AI market size, including hyperscale AI, will reach approximately USD 554.3 billion. Accordingly, competition to secure patents for preempting related technologies is expected to further intensify.
SHARE
Go to List page