23.9% of US-based AI staff at ten labs who list a bachelor's earned it in mainland China
AI roles at OpenAI, Anthropic, Google DeepMind, xAI, Meta, NVIDIA, Google, Microsoft, Apple, and Amazon, counted by where each person's bachelor's institution is located, on the Metix AI Platform on September 22, 2026.
23.9%
of US-based AI staff at the ten labs whose profile shows a bachelor's degree earned it at an institution in mainland China (4,582 of 19,178).
Who is counted
- Role
- A current job at one of the ten labs with an AI title: the title matches one of 12 terms such as machine learning, research scientist, or member of technical staff.
- Country
- The profile lists the United States as its country.
- Degree
- The profile shows a bachelor's degree.
- Counted when
- That bachelor's institution is located in mainland China, judged against a list of 37 match words (cities, provinces, and distinctive names) and 145 institution names. The list includes mainland campuses of foreign and Hong Kong universities and is built to leave out institutions in Hong Kong, Macau, and Taiwan.
- Never inferred
- Only the institution's location decides: nothing is inferred from names, languages, or nationality.
- What a count is
- A count of profiles visible on the Metix AI Platform, a lower bound rather than a headcount. Shares are ratios within those profiles and can sit above or below the share among all staff.
In brief
- The ten labs together are at 23.9%, at least 1.9 times the level across all US AI roles, which is at most 12.2% by the same list. The labs differ widely, from 11.2% at Anthropic to 36.0% at Meta.
- Among people whose current job started in the last 24 months the share is 21.7%, against 25.8% among people whose current job started earlier. Lab by lab the gap is 2.6 points, and only Meta and Microsoft differ by more than chance; the other labs differ within chance, in both directions.
- Titles with scientist or researcher are at 32.2%, against 17.9% for other AI titles at the same labs. The gap is clear at Amazon, Meta, and Microsoft and absent at Apple, Google DeepMind, and Google (not DeepMind).
- The most common of the twelve named institutions are Tsinghua University, Shanghai Jiao Tong University, University of Science and Technology of China, Zhejiang University, and Peking University. The University of Science and Technology of China comes before Peking University, and its row is a strict count.
01
From 11.2% at Anthropic to 36.0% at Meta; 9 of the ten are above the ceiling for all US AI roles
Share of US-based AI staff with a visible bachelor's whose bachelor's institution is in mainland China. Axis 0% to 50%; the dashed line is the ceiling for all US AI roles, the solid line the ten labs together
Meta 36.0% (1672 of 4643); xAI 26.4% (99 of 375); Amazon 23.9% (1052 of 4393); Google DeepMind 22.2% (165 of 743); NVIDIA 20.2% (243 of 1203); Apple 18.6% (371 of 1995); OpenAI 17.9% (273 of 1523); Microsoft 17.7% (355 of 2011); Google (not DeepMind) 17.0% (279 of 1640); Anthropic 11.2% (73 of 652)
What it shows
Across the US, more than 100,000 people in AI roles show a bachelor's degree, and 12,194 of them earned it in mainland China, so the level across all US AI roles is at most 12.2%. 9 of the ten labs are above that ceiling; Anthropic, at 11.2%, is below it, so the data cannot say whether it is above the actual all-US level.
Method and limits
Someone with current jobs at two of the labs counts once, at the first in the list, so the rows do not overlap and the ten-lab line pools them. Worldwide, not only in the US, the ten labs are at 19.0%.
Source: Metix AI Platform, profiles, 2026-09-22.
Queriespopulation.json · institutions.json
{ "note": "Who is counted. An AI role is a current job (experience.is_current eq true) whose title matches any of ai_title_terms; the lab and the recent-hire window sit in the same has_experience entry, so they describe that one job. US-based uses the profile's location.country: every profile with a mainland-China bachelor's in an AI role lists one (countries.json has no profile without a location), while the job-level location is often missing. The denominator is people with any Bachelor entry; the numerator is those whose Bachelor entry names an institution in institutions.json.", "ai_title_terms": [ "machine learning", "research scientist", "research engineer", "deep learning", "member of technical staff", "applied scientist", "artificial intelligence", "AI engineer", "AI researcher", "LLM", "NLP", "computer vision" ], "research_title_terms": [ "scientist", "researcher" ], "recent_hire_window": "now-24m", "us": { "field": "location.country", "eq": "United States" }, "labs": [ { "id": "openai", "name": "OpenAI", "companies": [ "OpenAI" ] }, { "id": "anthropic", "name": "Anthropic", "companies": [ "Anthropic" ] }, { "id": "google-deepmind", "name": "Google DeepMind", "companies": [ "Google DeepMind", "DeepMind" ] }, { "id": "xai", "name": "xAI", "companies": [ "xAI" ] }, { "id": "meta", "name": "Meta", "companies": [ "Meta" ] }, { "id": "nvidia", "name": "NVIDIA", "companies": [ "NVIDIA" ] }, { "id": "google", "name": "Google, other than Google DeepMind", "companies": [ "Google" ], "note": "The company name Google also matches people whose current job is at Google DeepMind, because the Platform matches company names word by word. Google DeepMind comes earlier in the order, so its staff count there and not here." }, { "id": "microsoft", "name": "Microsoft", "companies": [ "Microsoft" ] }, { "id": "apple", "name": "Apple", "companies": [ "Apple" ] }, { "id": "amazon", "name": "Amazon", "companies": [ "Amazon", "Amazon Web Services (AWS)", "AWS" ], "note": "Amazon Web Services adds 254 profiles with a Bachelor entry to the 6,098 under Amazon." } ], "institutions": [ { "id": "tsinghua", "name": "Tsinghua University", "match": [ "Tsinghua" ], "exact": [ "清华大学" ] }, { "id": "peking", "name": "Peking University", "match": [ "Peking University" ], "exact": [ "北京大学" ] }, { "id": "sjtu", "name": "Shanghai Jiao Tong University", "match": [ "Shanghai Jiao Tong" ], "exact": [ "Shanghai Jiaotong University", "上海交通大学" ] }, { "id": "zju", "name": "Zhejiang University", "exact": [ "Zhejiang University", "浙江大学" ], "exclude": [ "Sci-Tech", "Normal", "Gongshang", "Finance", "Media" ] }, { "id": "ustc", "name": "University of Science and Technology of China", "exact": [ "University of Science and Technology of China", "USTC", "中国科学技术大学" ], "exclude": [ "Electronic" ] }, { "id": "fudan", "name": "Fudan University", "match": [ "Fudan" ], "exact": [ "复旦大学" ] }, { "id": "nju", "name": "Nanjing University", "exact": [ "Nanjing University", "南京大学" ], "exclude": [ "Aeronautics", "Posts", "Information", "Normal", "Tech", "Medical", "Forestry", "Agricultural", "Audit", "Finance", "Science Technology" ] }, { "id": "hust", "name": "Huazhong University of Science and Technology", "match": [ "Huazhong Science Technology" ], "exact": [ "华中科技大学" ] }, { "id": "hit", "name": "Harbin Institute of Technology", "match": [ "Harbin Institute Technology" ], "exact": [ "哈尔滨工业大学" ] }, { "id": "beihang", "name": "Beihang University", "match": [ "Beihang" ], "exact": [ "Beijing University of Aeronautics and Astronautics", "北京航空航天大学" ] }, { "id": "xjtu", "name": "Xi'an Jiaotong University", "exact": [ "Xi'an Jiaotong University", "Xian Jiaotong University", "西安交通大学" ], "exclude": [ "Liverpool" ] }, { "id": "uestc", "name": "University of Electronic Science and Technology of China", "exact": [ "University of Electronic Science and Technology of China", "UESTC", "电子科技大学" ] } ], "institutions_note": "One institution per count. The Platform matches names word by word, so Nanjing University also matches Nanjing University of Aeronautics and Astronautics, and University of Science and Technology of China also matches University of Electronic Science and Technology of China. A row with exclude words leaves out entries that contain them, which makes it a strict count: entries such as Nanjing University, Department of Computer Science and Technology drop out too. The twelve are large mainland universities named in the prompt before any count was run; they are not a ranking of every institution, and others outside the twelve are not counted one by one.", "countries": [ "United States", "China", "Canada", "United Kingdom", "Singapore", "Germany", "Australia", "Switzerland", "France", "Japan", "Hong Kong" ], "lab_order_note": "Rows do not overlap. A lab's row counts people whose current AI-titled job is at that lab and who hold no current job at a lab listed before it, so someone with current jobs at two labs counts once, at the first. The ten-lab row is the sum of the ten rows.", "countries_note": "Values of the profile's location.country. Hong Kong is its own value on the Platform, listed here as a region."}
{ "note": "Institutions located in mainland China, used on education.school.name inside one has_education entry together with education.degree eq Bachelor. A profile counts when one Bachelor entry names an institution on this list and none of the exclude words; nothing is inferred from personal names or languages. The Platform matches both match terms and the names in exact word by word: a name counts when every one of its words appears in the entry, in any order, so a longer name that contains a listed one also counts. Match terms are words that only mainland institutions use in their names (cities, provinces, distinctive names). The exclude words remove entries whose names contain a listed name but belong to an institution elsewhere: National Sun Yat-sen University in Taiwan contains Sun Yat-sen University, and Southeast Missouri State University contains Southeast University. Mainland campuses of foreign and Hong Kong universities count, because the rule is where the institution is. The list is built to leave out institutions in Hong Kong, Macau, and Taiwan. The school country recorded on an entry cannot be queried, so it cannot serve as the rule. The query limit of 64 conditions keeps the match terms to 37; less common institutions are listed by name.", "how_built": "Built from the Bachelor entries of 250 profiles with a current AI title and a current location in China, read in full. The school country recorded on each entry is missing for many mainland institutions (Beihang University, Huazhong University of Science and Technology, and some Tsinghua University entries among them), so each institution's location was decided from its name and reviewed by hand. The list catches 173 of the 173 mainland Bachelor entries in that pool.", "audit": "Two independent samples, read in full. 250 profiles with a current AI title at the ten labs, 25 per lab: the list caught 28 of the 28 mainland Bachelor entries. 200 US-based AI staff with a Bachelor entry at Meta, Amazon, NVIDIA, and Microsoft: the list caught 60 of 63 and matched one institution in Taiwan; the three it missed (Capital Medical University, Southern Medical University, Jimei University) are now listed and the Taiwan match is excluded. On the ten labs' US-based staff the exclude words remove 6 entries and the three added names add 5. Less common institutions still slip through, so shares built on the list are probably slightly low.", "match": [ "Tsinghua", "Peking University", "Fudan", "Tongji", "Nankai", "Beihang", "Renmin", "Huazhong", "Jiaotong", "Jiao Tong", "Xidian", "ShanghaiTech", "SUSTech", "Kunshan", "Ningbo", "Beijing", "Shanghai", "Tianjin", "Chongqing", "Nanjing", "Wuhan", "Harbin", "Hangzhou", "Hefei", "Xiamen", "Dalian", "Qingdao", "Guangzhou", "Shenzhen", "Chengdu", "Zhengzhou", "Jinan", "Zhejiang", "Sichuan", "Shandong", "Jilin", "Hunan" ], "exact": [ "Anhui Agricultural University", "Anhui University", "Capital Medical University", "Capital Normal University", "Capital University of Economics and Business", "Central South University", "Central University of Finance and Economics", "China Agricultural University", "China Pharmaceutical University", "China University of Geosciences", "China University of Mining and Technology", "China University of Petroleum", "Communication University of China", "Donghua University", "Duke Kunshan University", "East China Normal University", "East China University of Science and Technology", "Fujian Normal University", "Fuzhou University", "Guangdong University of Technology", "Guangxi University", "Guizhou University", "Hainan University", "Hebei University of Technology", "Heilongjiang University", "Henan University", "Hohai University", "Hubei University", "Inner Mongolia University", "Jiangnan University", "Jiangsu University", "Jiangxi University of Finance and Economics", "Jilin Agricultural University", "Jimei University", "Kunming University of Science and Technology", "Lanzhou University", "Liaoning University", "Minzu University of China", "Nanchang University", "Nanjing Forestry University", "National University of Defense Technology", "New York University Shanghai", "North China Electric Power University", "Northeast Forestry University", "Northeast Normal University", "Northwest A&F University", "Northwest University", "Northwestern Polytechnical University", "Ocean University of China", "Peking Union Medical College", "Shaanxi Normal University", "Shenyang University of Technology", "Soochow University (CN)", "South China Agricultural University", "South China Normal University", "South China University of Technology", "Southeast University", "Southern Medical University", "Southern University of Science and Technology", "Southwest University", "Sun Yat-Sen University", "Sun Yat-sen University", "Taiyuan University of Technology", "The Chinese University of Hong Kong, Shenzhen", "USTC", "University of Electronic Science and Technology of China", "University of International Business and Economics", "University of Science and Technology of China", "Westlake University", "Xi'an Jiaotong-Liverpool University", "Xinjiang University", "Yunnan University", "上海交通大学", "上海大学", "上海科技大学", "上海财经大学", "东北大学", "东华大学", "东南大学", "中南大学", "中国人民大学", "中国农业大学", "中国海洋大学", "中国科学技术大学", "中央民族大学", "中央财经大学", "中山大学", "佛山大学", "兰州大学", "北京交通大学", "北京大学", "北京师范大学", "北京理工大学", "北京科技大学", "北京航空航天大学", "北京邮电大学", "华东师范大学", "华东理工大学", "华中科技大学", "华南师范大学", "华南理工大学", "南京大学", "南京理工大学", "南京航空航天大学", "南京邮电大学", "南开大学", "南方科技大学", "厦门大学", "合肥工业大学", "吉林大学", "同济大学", "哈尔滨工业大学", "哈尔滨工程大学", "四川大学", "国防科技大学", "复旦大学", "大连理工大学", "天津大学", "对外经济贸易大学", "山东大学", "山东理工大学", "暨南大学", "杭州电子科技大学", "武汉大学", "武汉理工大学", "武汉科技学院", "河海大学", "济南大学", "浙江大学", "深圳大学", "清华大学", "湖南大学", "电子科技大学", "苏州大学", "西交利物浦大学", "西北农林科技大学", "西北工业大学", "西南交通大学", "西安交通大学", "西安电子科技大学", "西湖大学", "郑州大学", "重庆大学", "首都师范大学", "香港中文大学（深圳）" ], "exclude": [ "National Sun Yat-sen", "Taiwan", "Malaysia", "Missouri" ]}
02
People who started their current job in the last 24 months are 4.1 points lower, with the largest gaps at Meta and Microsoft
Share among people whose current job started earlier (open) and in the last 24 months (filled). Violet rows differ by more than groups this size vary by chance; the ink row is the ten labs together
Meta: earlier 39.5%, last 24 months 31.0%; Microsoft: earlier 20.8%, last 24 months 14.8%; NVIDIA: earlier 21.3%, last 24 months 18.7%; Amazon: earlier 24.9%, last 24 months 22.8%; Apple: earlier 19.0%, last 24 months 17.8%; Google (not DeepMind): earlier 16.5%, last 24 months 17.9%; OpenAI: earlier 16.4%, last 24 months 18.6%; Google DeepMind: earlier 20.1%, last 24 months 23.8%; Anthropic: earlier 8.2%, last 24 months 12.1%
What it shows
Across the ten labs the gap is 4.1 points, partly because the labs where the most people started their current job recently have lower shares to begin with; compared lab by lab and weighted by those people, it is 2.6 points. Only Meta and Microsoft differ by more than chance would explain (a two-proportion test treating the visible profiles as a sample, |z| of at least 1.96); at the other labs |z| is at most 1.7. xAI has a group under 100 and is not drawn.
Method and limits
This is the start of the current job, internal moves included, so it describes today's staff by tenure, not a hiring pipeline. The earlier group has had longer to lose people who left, so if people with and without a mainland-China bachelor's leave at different rates, the gap may reflect who stayed as well as who was hired. Profiles with no start date on the current job fall in the earlier group. The comparison does not account for job type either: scientist and researcher titles have higher shares (figure 03), the data does not show whether recent starters hold them as often, and the effect could run in either direction.
Source: Metix AI Platform, profiles, 2026-09-22.
Queriespopulation.json · institutions.json
{ "note": "Who is counted. An AI role is a current job (experience.is_current eq true) whose title matches any of ai_title_terms; the lab and the recent-hire window sit in the same has_experience entry, so they describe that one job. US-based uses the profile's location.country: every profile with a mainland-China bachelor's in an AI role lists one (countries.json has no profile without a location), while the job-level location is often missing. The denominator is people with any Bachelor entry; the numerator is those whose Bachelor entry names an institution in institutions.json.", "ai_title_terms": [ "machine learning", "research scientist", "research engineer", "deep learning", "member of technical staff", "applied scientist", "artificial intelligence", "AI engineer", "AI researcher", "LLM", "NLP", "computer vision" ], "research_title_terms": [ "scientist", "researcher" ], "recent_hire_window": "now-24m", "us": { "field": "location.country", "eq": "United States" }, "labs": [ { "id": "openai", "name": "OpenAI", "companies": [ "OpenAI" ] }, { "id": "anthropic", "name": "Anthropic", "companies": [ "Anthropic" ] }, { "id": "google-deepmind", "name": "Google DeepMind", "companies": [ "Google DeepMind", "DeepMind" ] }, { "id": "xai", "name": "xAI", "companies": [ "xAI" ] }, { "id": "meta", "name": "Meta", "companies": [ "Meta" ] }, { "id": "nvidia", "name": "NVIDIA", "companies": [ "NVIDIA" ] }, { "id": "google", "name": "Google, other than Google DeepMind", "companies": [ "Google" ], "note": "The company name Google also matches people whose current job is at Google DeepMind, because the Platform matches company names word by word. Google DeepMind comes earlier in the order, so its staff count there and not here." }, { "id": "microsoft", "name": "Microsoft", "companies": [ "Microsoft" ] }, { "id": "apple", "name": "Apple", "companies": [ "Apple" ] }, { "id": "amazon", "name": "Amazon", "companies": [ "Amazon", "Amazon Web Services (AWS)", "AWS" ], "note": "Amazon Web Services adds 254 profiles with a Bachelor entry to the 6,098 under Amazon." } ], "institutions": [ { "id": "tsinghua", "name": "Tsinghua University", "match": [ "Tsinghua" ], "exact": [ "清华大学" ] }, { "id": "peking", "name": "Peking University", "match": [ "Peking University" ], "exact": [ "北京大学" ] }, { "id": "sjtu", "name": "Shanghai Jiao Tong University", "match": [ "Shanghai Jiao Tong" ], "exact": [ "Shanghai Jiaotong University", "上海交通大学" ] }, { "id": "zju", "name": "Zhejiang University", "exact": [ "Zhejiang University", "浙江大学" ], "exclude": [ "Sci-Tech", "Normal", "Gongshang", "Finance", "Media" ] }, { "id": "ustc", "name": "University of Science and Technology of China", "exact": [ "University of Science and Technology of China", "USTC", "中国科学技术大学" ], "exclude": [ "Electronic" ] }, { "id": "fudan", "name": "Fudan University", "match": [ "Fudan" ], "exact": [ "复旦大学" ] }, { "id": "nju", "name": "Nanjing University", "exact": [ "Nanjing University", "南京大学" ], "exclude": [ "Aeronautics", "Posts", "Information", "Normal", "Tech", "Medical", "Forestry", "Agricultural", "Audit", "Finance", "Science Technology" ] }, { "id": "hust", "name": "Huazhong University of Science and Technology", "match": [ "Huazhong Science Technology" ], "exact": [ "华中科技大学" ] }, { "id": "hit", "name": "Harbin Institute of Technology", "match": [ "Harbin Institute Technology" ], "exact": [ "哈尔滨工业大学" ] }, { "id": "beihang", "name": "Beihang University", "match": [ "Beihang" ], "exact": [ "Beijing University of Aeronautics and Astronautics", "北京航空航天大学" ] }, { "id": "xjtu", "name": "Xi'an Jiaotong University", "exact": [ "Xi'an Jiaotong University", "Xian Jiaotong University", "西安交通大学" ], "exclude": [ "Liverpool" ] }, { "id": "uestc", "name": "University of Electronic Science and Technology of China", "exact": [ "University of Electronic Science and Technology of China", "UESTC", "电子科技大学" ] } ], "institutions_note": "One institution per count. The Platform matches names word by word, so Nanjing University also matches Nanjing University of Aeronautics and Astronautics, and University of Science and Technology of China also matches University of Electronic Science and Technology of China. A row with exclude words leaves out entries that contain them, which makes it a strict count: entries such as Nanjing University, Department of Computer Science and Technology drop out too. The twelve are large mainland universities named in the prompt before any count was run; they are not a ranking of every institution, and others outside the twelve are not counted one by one.", "countries": [ "United States", "China", "Canada", "United Kingdom", "Singapore", "Germany", "Australia", "Switzerland", "France", "Japan", "Hong Kong" ], "lab_order_note": "Rows do not overlap. A lab's row counts people whose current AI-titled job is at that lab and who hold no current job at a lab listed before it, so someone with current jobs at two labs counts once, at the first. The ten-lab row is the sum of the ten rows.", "countries_note": "Values of the profile's location.country. Hong Kong is its own value on the Platform, listed here as a region."}
{ "note": "Institutions located in mainland China, used on education.school.name inside one has_education entry together with education.degree eq Bachelor. A profile counts when one Bachelor entry names an institution on this list and none of the exclude words; nothing is inferred from personal names or languages. The Platform matches both match terms and the names in exact word by word: a name counts when every one of its words appears in the entry, in any order, so a longer name that contains a listed one also counts. Match terms are words that only mainland institutions use in their names (cities, provinces, distinctive names). The exclude words remove entries whose names contain a listed name but belong to an institution elsewhere: National Sun Yat-sen University in Taiwan contains Sun Yat-sen University, and Southeast Missouri State University contains Southeast University. Mainland campuses of foreign and Hong Kong universities count, because the rule is where the institution is. The list is built to leave out institutions in Hong Kong, Macau, and Taiwan. The school country recorded on an entry cannot be queried, so it cannot serve as the rule. The query limit of 64 conditions keeps the match terms to 37; less common institutions are listed by name.", "how_built": "Built from the Bachelor entries of 250 profiles with a current AI title and a current location in China, read in full. The school country recorded on each entry is missing for many mainland institutions (Beihang University, Huazhong University of Science and Technology, and some Tsinghua University entries among them), so each institution's location was decided from its name and reviewed by hand. The list catches 173 of the 173 mainland Bachelor entries in that pool.", "audit": "Two independent samples, read in full. 250 profiles with a current AI title at the ten labs, 25 per lab: the list caught 28 of the 28 mainland Bachelor entries. 200 US-based AI staff with a Bachelor entry at Meta, Amazon, NVIDIA, and Microsoft: the list caught 60 of 63 and matched one institution in Taiwan; the three it missed (Capital Medical University, Southern Medical University, Jimei University) are now listed and the Taiwan match is excluded. On the ten labs' US-based staff the exclude words remove 6 entries and the three added names add 5. Less common institutions still slip through, so shares built on the list are probably slightly low.", "match": [ "Tsinghua", "Peking University", "Fudan", "Tongji", "Nankai", "Beihang", "Renmin", "Huazhong", "Jiaotong", "Jiao Tong", "Xidian", "ShanghaiTech", "SUSTech", "Kunshan", "Ningbo", "Beijing", "Shanghai", "Tianjin", "Chongqing", "Nanjing", "Wuhan", "Harbin", "Hangzhou", "Hefei", "Xiamen", "Dalian", "Qingdao", "Guangzhou", "Shenzhen", "Chengdu", "Zhengzhou", "Jinan", "Zhejiang", "Sichuan", "Shandong", "Jilin", "Hunan" ], "exact": [ "Anhui Agricultural University", "Anhui University", "Capital Medical University", "Capital Normal University", "Capital University of Economics and Business", "Central South University", "Central University of Finance and Economics", "China Agricultural University", "China Pharmaceutical University", "China University of Geosciences", "China University of Mining and Technology", "China University of Petroleum", "Communication University of China", "Donghua University", "Duke Kunshan University", "East China Normal University", "East China University of Science and Technology", "Fujian Normal University", "Fuzhou University", "Guangdong University of Technology", "Guangxi University", "Guizhou University", "Hainan University", "Hebei University of Technology", "Heilongjiang University", "Henan University", "Hohai University", "Hubei University", "Inner Mongolia University", "Jiangnan University", "Jiangsu University", "Jiangxi University of Finance and Economics", "Jilin Agricultural University", "Jimei University", "Kunming University of Science and Technology", "Lanzhou University", "Liaoning University", "Minzu University of China", "Nanchang University", "Nanjing Forestry University", "National University of Defense Technology", "New York University Shanghai", "North China Electric Power University", "Northeast Forestry University", "Northeast Normal University", "Northwest A&F University", "Northwest University", "Northwestern Polytechnical University", "Ocean University of China", "Peking Union Medical College", "Shaanxi Normal University", "Shenyang University of Technology", "Soochow University (CN)", "South China Agricultural University", "South China Normal University", "South China University of Technology", "Southeast University", "Southern Medical University", "Southern University of Science and Technology", "Southwest University", "Sun Yat-Sen University", "Sun Yat-sen University", "Taiyuan University of Technology", "The Chinese University of Hong Kong, Shenzhen", "USTC", "University of Electronic Science and Technology of China", "University of International Business and Economics", "University of Science and Technology of China", "Westlake University", "Xi'an Jiaotong-Liverpool University", "Xinjiang University", "Yunnan University", "上海交通大学", "上海大学", "上海科技大学", "上海财经大学", "东北大学", "东华大学", "东南大学", "中南大学", "中国人民大学", "中国农业大学", "中国海洋大学", "中国科学技术大学", "中央民族大学", "中央财经大学", "中山大学", "佛山大学", "兰州大学", "北京交通大学", "北京大学", "北京师范大学", "北京理工大学", "北京科技大学", "北京航空航天大学", "北京邮电大学", "华东师范大学", "华东理工大学", "华中科技大学", "华南师范大学", "华南理工大学", "南京大学", "南京理工大学", "南京航空航天大学", "南京邮电大学", "南开大学", "南方科技大学", "厦门大学", "合肥工业大学", "吉林大学", "同济大学", "哈尔滨工业大学", "哈尔滨工程大学", "四川大学", "国防科技大学", "复旦大学", "大连理工大学", "天津大学", "对外经济贸易大学", "山东大学", "山东理工大学", "暨南大学", "杭州电子科技大学", "武汉大学", "武汉理工大学", "武汉科技学院", "河海大学", "济南大学", "浙江大学", "深圳大学", "清华大学", "湖南大学", "电子科技大学", "苏州大学", "西交利物浦大学", "西北农林科技大学", "西北工业大学", "西南交通大学", "西安交通大学", "西安电子科技大学", "西湖大学", "郑州大学", "重庆大学", "首都师范大学", "香港中文大学（深圳）" ], "exclude": [ "National Sun Yat-sen", "Taiwan", "Malaysia", "Missouri" ]}
03
Titles with scientist or researcher are at 32.2%, against 17.9% for other AI titles at the same labs
Titles with scientist or researcher (filled), against the same lab's other AI titles (open). Only labs with at least 100 of each are drawn
Amazon: scientist or researcher 29.9%, other AI titles 7.4%; Meta: scientist or researcher 46.2%, other AI titles 25.6%; Microsoft: scientist or researcher 24.7%, other AI titles 11.4%; NVIDIA: scientist or researcher 24.1%, other AI titles 18.4%; Apple: scientist or researcher 19.9%, other AI titles 18.3%; Google DeepMind: scientist or researcher 22.7%, other AI titles 21.7%; Google (not DeepMind): scientist or researcher 17.6%, other AI titles 16.7%
What it shows
Applied Scientist counts here: at Amazon, 3,229 of the 4,393 AI staff with a bachelor's hold such a title. The gap is far beyond chance at Amazon, Meta, and Microsoft, marginal at NVIDIA (it passes 1.96 but not a correction for the 7 comparisons), and absent at Apple, Google DeepMind, and Google (not DeepMind).
Method and limits
OpenAI, Anthropic, and xAI mostly use titles such as member of technical staff and have fewer than 100 scientist or researcher titles, so they are not drawn. Some of their cells are too small to publish, so there is no ten-lab figure here: the summary row pools the labs drawn.
Source: Metix AI Platform, profiles, 2026-09-22.
Queriespopulation.json · institutions.json
{ "note": "Who is counted. An AI role is a current job (experience.is_current eq true) whose title matches any of ai_title_terms; the lab and the recent-hire window sit in the same has_experience entry, so they describe that one job. US-based uses the profile's location.country: every profile with a mainland-China bachelor's in an AI role lists one (countries.json has no profile without a location), while the job-level location is often missing. The denominator is people with any Bachelor entry; the numerator is those whose Bachelor entry names an institution in institutions.json.", "ai_title_terms": [ "machine learning", "research scientist", "research engineer", "deep learning", "member of technical staff", "applied scientist", "artificial intelligence", "AI engineer", "AI researcher", "LLM", "NLP", "computer vision" ], "research_title_terms": [ "scientist", "researcher" ], "recent_hire_window": "now-24m", "us": { "field": "location.country", "eq": "United States" }, "labs": [ { "id": "openai", "name": "OpenAI", "companies": [ "OpenAI" ] }, { "id": "anthropic", "name": "Anthropic", "companies": [ "Anthropic" ] }, { "id": "google-deepmind", "name": "Google DeepMind", "companies": [ "Google DeepMind", "DeepMind" ] }, { "id": "xai", "name": "xAI", "companies": [ "xAI" ] }, { "id": "meta", "name": "Meta", "companies": [ "Meta" ] }, { "id": "nvidia", "name": "NVIDIA", "companies": [ "NVIDIA" ] }, { "id": "google", "name": "Google, other than Google DeepMind", "companies": [ "Google" ], "note": "The company name Google also matches people whose current job is at Google DeepMind, because the Platform matches company names word by word. Google DeepMind comes earlier in the order, so its staff count there and not here." }, { "id": "microsoft", "name": "Microsoft", "companies": [ "Microsoft" ] }, { "id": "apple", "name": "Apple", "companies": [ "Apple" ] }, { "id": "amazon", "name": "Amazon", "companies": [ "Amazon", "Amazon Web Services (AWS)", "AWS" ], "note": "Amazon Web Services adds 254 profiles with a Bachelor entry to the 6,098 under Amazon." } ], "institutions": [ { "id": "tsinghua", "name": "Tsinghua University", "match": [ "Tsinghua" ], "exact": [ "清华大学" ] }, { "id": "peking", "name": "Peking University", "match": [ "Peking University" ], "exact": [ "北京大学" ] }, { "id": "sjtu", "name": "Shanghai Jiao Tong University", "match": [ "Shanghai Jiao Tong" ], "exact": [ "Shanghai Jiaotong University", "上海交通大学" ] }, { "id": "zju", "name": "Zhejiang University", "exact": [ "Zhejiang University", "浙江大学" ], "exclude": [ "Sci-Tech", "Normal", "Gongshang", "Finance", "Media" ] }, { "id": "ustc", "name": "University of Science and Technology of China", "exact": [ "University of Science and Technology of China", "USTC", "中国科学技术大学" ], "exclude": [ "Electronic" ] }, { "id": "fudan", "name": "Fudan University", "match": [ "Fudan" ], "exact": [ "复旦大学" ] }, { "id": "nju", "name": "Nanjing University", "exact": [ "Nanjing University", "南京大学" ], "exclude": [ "Aeronautics", "Posts", "Information", "Normal", "Tech", "Medical", "Forestry", "Agricultural", "Audit", "Finance", "Science Technology" ] }, { "id": "hust", "name": "Huazhong University of Science and Technology", "match": [ "Huazhong Science Technology" ], "exact": [ "华中科技大学" ] }, { "id": "hit", "name": "Harbin Institute of Technology", "match": [ "Harbin Institute Technology" ], "exact": [ "哈尔滨工业大学" ] }, { "id": "beihang", "name": "Beihang University", "match": [ "Beihang" ], "exact": [ "Beijing University of Aeronautics and Astronautics", "北京航空航天大学" ] }, { "id": "xjtu", "name": "Xi'an Jiaotong University", "exact": [ "Xi'an Jiaotong University", "Xian Jiaotong University", "西安交通大学" ], "exclude": [ "Liverpool" ] }, { "id": "uestc", "name": "University of Electronic Science and Technology of China", "exact": [ "University of Electronic Science and Technology of China", "UESTC", "电子科技大学" ] } ], "institutions_note": "One institution per count. The Platform matches names word by word, so Nanjing University also matches Nanjing University of Aeronautics and Astronautics, and University of Science and Technology of China also matches University of Electronic Science and Technology of China. A row with exclude words leaves out entries that contain them, which makes it a strict count: entries such as Nanjing University, Department of Computer Science and Technology drop out too. The twelve are large mainland universities named in the prompt before any count was run; they are not a ranking of every institution, and others outside the twelve are not counted one by one.", "countries": [ "United States", "China", "Canada", "United Kingdom", "Singapore", "Germany", "Australia", "Switzerland", "France", "Japan", "Hong Kong" ], "lab_order_note": "Rows do not overlap. A lab's row counts people whose current AI-titled job is at that lab and who hold no current job at a lab listed before it, so someone with current jobs at two labs counts once, at the first. The ten-lab row is the sum of the ten rows.", "countries_note": "Values of the profile's location.country. Hong Kong is its own value on the Platform, listed here as a region."}
{ "note": "Institutions located in mainland China, used on education.school.name inside one has_education entry together with education.degree eq Bachelor. A profile counts when one Bachelor entry names an institution on this list and none of the exclude words; nothing is inferred from personal names or languages. The Platform matches both match terms and the names in exact word by word: a name counts when every one of its words appears in the entry, in any order, so a longer name that contains a listed one also counts. Match terms are words that only mainland institutions use in their names (cities, provinces, distinctive names). The exclude words remove entries whose names contain a listed name but belong to an institution elsewhere: National Sun Yat-sen University in Taiwan contains Sun Yat-sen University, and Southeast Missouri State University contains Southeast University. Mainland campuses of foreign and Hong Kong universities count, because the rule is where the institution is. The list is built to leave out institutions in Hong Kong, Macau, and Taiwan. The school country recorded on an entry cannot be queried, so it cannot serve as the rule. The query limit of 64 conditions keeps the match terms to 37; less common institutions are listed by name.", "how_built": "Built from the Bachelor entries of 250 profiles with a current AI title and a current location in China, read in full. The school country recorded on each entry is missing for many mainland institutions (Beihang University, Huazhong University of Science and Technology, and some Tsinghua University entries among them), so each institution's location was decided from its name and reviewed by hand. The list catches 173 of the 173 mainland Bachelor entries in that pool.", "audit": "Two independent samples, read in full. 250 profiles with a current AI title at the ten labs, 25 per lab: the list caught 28 of the 28 mainland Bachelor entries. 200 US-based AI staff with a Bachelor entry at Meta, Amazon, NVIDIA, and Microsoft: the list caught 60 of 63 and matched one institution in Taiwan; the three it missed (Capital Medical University, Southern Medical University, Jimei University) are now listed and the Taiwan match is excluded. On the ten labs' US-based staff the exclude words remove 6 entries and the three added names add 5. Less common institutions still slip through, so shares built on the list are probably slightly low.", "match": [ "Tsinghua", "Peking University", "Fudan", "Tongji", "Nankai", "Beihang", "Renmin", "Huazhong", "Jiaotong", "Jiao Tong", "Xidian", "ShanghaiTech", "SUSTech", "Kunshan", "Ningbo", "Beijing", "Shanghai", "Tianjin", "Chongqing", "Nanjing", "Wuhan", "Harbin", "Hangzhou", "Hefei", "Xiamen", "Dalian", "Qingdao", "Guangzhou", "Shenzhen", "Chengdu", "Zhengzhou", "Jinan", "Zhejiang", "Sichuan", "Shandong", "Jilin", "Hunan" ], "exact": [ "Anhui Agricultural University", "Anhui University", "Capital Medical University", "Capital Normal University", "Capital University of Economics and Business", "Central South University", "Central University of Finance and Economics", "China Agricultural University", "China Pharmaceutical University", "China University of Geosciences", "China University of Mining and Technology", "China University of Petroleum", "Communication University of China", "Donghua University", "Duke Kunshan University", "East China Normal University", "East China University of Science and Technology", "Fujian Normal University", "Fuzhou University", "Guangdong University of Technology", "Guangxi University", "Guizhou University", "Hainan University", "Hebei University of Technology", "Heilongjiang University", "Henan University", "Hohai University", "Hubei University", "Inner Mongolia University", "Jiangnan University", "Jiangsu University", "Jiangxi University of Finance and Economics", "Jilin Agricultural University", "Jimei University", "Kunming University of Science and Technology", "Lanzhou University", "Liaoning University", "Minzu University of China", "Nanchang University", "Nanjing Forestry University", "National University of Defense Technology", "New York University Shanghai", "North China Electric Power University", "Northeast Forestry University", "Northeast Normal University", "Northwest A&F University", "Northwest University", "Northwestern Polytechnical University", "Ocean University of China", "Peking Union Medical College", "Shaanxi Normal University", "Shenyang University of Technology", "Soochow University (CN)", "South China Agricultural University", "South China Normal University", "South China University of Technology", "Southeast University", "Southern Medical University", "Southern University of Science and Technology", "Southwest University", "Sun Yat-Sen University", "Sun Yat-sen University", "Taiyuan University of Technology", "The Chinese University of Hong Kong, Shenzhen", "USTC", "University of Electronic Science and Technology of China", "University of International Business and Economics", "University of Science and Technology of China", "Westlake University", "Xi'an Jiaotong-Liverpool University", "Xinjiang University", "Yunnan University", "上海交通大学", "上海大学", "上海科技大学", "上海财经大学", "东北大学", "东华大学", "东南大学", "中南大学", "中国人民大学", "中国农业大学", "中国海洋大学", "中国科学技术大学", "中央民族大学", "中央财经大学", "中山大学", "佛山大学", "兰州大学", "北京交通大学", "北京大学", "北京师范大学", "北京理工大学", "北京科技大学", "北京航空航天大学", "北京邮电大学", "华东师范大学", "华东理工大学", "华中科技大学", "华南师范大学", "华南理工大学", "南京大学", "南京理工大学", "南京航空航天大学", "南京邮电大学", "南开大学", "南方科技大学", "厦门大学", "合肥工业大学", "吉林大学", "同济大学", "哈尔滨工业大学", "哈尔滨工程大学", "四川大学", "国防科技大学", "复旦大学", "大连理工大学", "天津大学", "对外经济贸易大学", "山东大学", "山东理工大学", "暨南大学", "杭州电子科技大学", "武汉大学", "武汉理工大学", "武汉科技学院", "河海大学", "济南大学", "浙江大学", "深圳大学", "清华大学", "湖南大学", "电子科技大学", "苏州大学", "西交利物浦大学", "西北农林科技大学", "西北工业大学", "西南交通大学", "西安交通大学", "西安电子科技大学", "西湖大学", "郑州大学", "重庆大学", "首都师范大学", "香港中文大学（深圳）" ], "exclude": [ "National Sun Yat-sen", "Taiwan", "Malaysia", "Missouri" ]}
04
The twelve named universities add up to 2,854 entries, and USTC comes before Peking University
US-based AI staff at the ten labs by bachelor's institution, as a share of the 4,582; axis 0% to 40%. The Platform matches names word by word; hatched bars are strict counts that leave out entries whose names also contain another university's words
Tsinghua University 520, Shanghai Jiao Tong University 409, University of Science and Technology of China 385, Zhejiang University 345, Peking University 332, Fudan University 191, Huazhong University of Science and Technology 173, Nanjing University 154, Beihang University 96, Xi'an Jiaotong University 94, Harbin Institute of Technology 92, University of Electronic Science and Technology of China 63
- Tsinghua University11%520
- Shanghai Jiao Tong University9%409
- University of Science and Technology of China8%385
- Zhejiang University8%345
- Peking University7%332
- Fudan University4%191
- Huazhong University of Science and Technology4%173
- Nanjing University3%154
- Beihang University2%96
- Xi'an Jiaotong University2%94
- Harbin Institute of Technology2%92
- University of Electronic Science and Technology of China1%63
- Every other listed institution, at least38%1,728
What it shows
Every bar is a share of the 4,582 at the ten labs. Each institution is counted on its own, and someone with bachelor's entries at two of them appears under both, so the twelve rows add up to 2,854 entries and at least 1,728 people studied at another listed institution. The twelve are large universities named in the prompt before any count ran, not a ranking of every institution.
Method and limits
The Platform matches institution names word by word: a name counts when each of its words appears in the entry, so Nanjing University also matches Nanjing University of Aeronautics and Astronautics, and the University of Science and Technology of China also matches the University of Electronic Science and Technology of China. The hatched rows leave out entries with those words, which makes them strict counts: an entry such as "Nanjing University, Department of Computer Science and Technology" drops out too.
Source: Metix AI Platform, profiles, 2026-09-22.
Queriespopulation.json · institutions.json
{ "note": "Who is counted. An AI role is a current job (experience.is_current eq true) whose title matches any of ai_title_terms; the lab and the recent-hire window sit in the same has_experience entry, so they describe that one job. US-based uses the profile's location.country: every profile with a mainland-China bachelor's in an AI role lists one (countries.json has no profile without a location), while the job-level location is often missing. The denominator is people with any Bachelor entry; the numerator is those whose Bachelor entry names an institution in institutions.json.", "ai_title_terms": [ "machine learning", "research scientist", "research engineer", "deep learning", "member of technical staff", "applied scientist", "artificial intelligence", "AI engineer", "AI researcher", "LLM", "NLP", "computer vision" ], "research_title_terms": [ "scientist", "researcher" ], "recent_hire_window": "now-24m", "us": { "field": "location.country", "eq": "United States" }, "labs": [ { "id": "openai", "name": "OpenAI", "companies": [ "OpenAI" ] }, { "id": "anthropic", "name": "Anthropic", "companies": [ "Anthropic" ] }, { "id": "google-deepmind", "name": "Google DeepMind", "companies": [ "Google DeepMind", "DeepMind" ] }, { "id": "xai", "name": "xAI", "companies": [ "xAI" ] }, { "id": "meta", "name": "Meta", "companies": [ "Meta" ] }, { "id": "nvidia", "name": "NVIDIA", "companies": [ "NVIDIA" ] }, { "id": "google", "name": "Google, other than Google DeepMind", "companies": [ "Google" ], "note": "The company name Google also matches people whose current job is at Google DeepMind, because the Platform matches company names word by word. Google DeepMind comes earlier in the order, so its staff count there and not here." }, { "id": "microsoft", "name": "Microsoft", "companies": [ "Microsoft" ] }, { "id": "apple", "name": "Apple", "companies": [ "Apple" ] }, { "id": "amazon", "name": "Amazon", "companies": [ "Amazon", "Amazon Web Services (AWS)", "AWS" ], "note": "Amazon Web Services adds 254 profiles with a Bachelor entry to the 6,098 under Amazon." } ], "institutions": [ { "id": "tsinghua", "name": "Tsinghua University", "match": [ "Tsinghua" ], "exact": [ "清华大学" ] }, { "id": "peking", "name": "Peking University", "match": [ "Peking University" ], "exact": [ "北京大学" ] }, { "id": "sjtu", "name": "Shanghai Jiao Tong University", "match": [ "Shanghai Jiao Tong" ], "exact": [ "Shanghai Jiaotong University", "上海交通大学" ] }, { "id": "zju", "name": "Zhejiang University", "exact": [ "Zhejiang University", "浙江大学" ], "exclude": [ "Sci-Tech", "Normal", "Gongshang", "Finance", "Media" ] }, { "id": "ustc", "name": "University of Science and Technology of China", "exact": [ "University of Science and Technology of China", "USTC", "中国科学技术大学" ], "exclude": [ "Electronic" ] }, { "id": "fudan", "name": "Fudan University", "match": [ "Fudan" ], "exact": [ "复旦大学" ] }, { "id": "nju", "name": "Nanjing University", "exact": [ "Nanjing University", "南京大学" ], "exclude": [ "Aeronautics", "Posts", "Information", "Normal", "Tech", "Medical", "Forestry", "Agricultural", "Audit", "Finance", "Science Technology" ] }, { "id": "hust", "name": "Huazhong University of Science and Technology", "match": [ "Huazhong Science Technology" ], "exact": [ "华中科技大学" ] }, { "id": "hit", "name": "Harbin Institute of Technology", "match": [ "Harbin Institute Technology" ], "exact": [ "哈尔滨工业大学" ] }, { "id": "beihang", "name": "Beihang University", "match": [ "Beihang" ], "exact": [ "Beijing University of Aeronautics and Astronautics", "北京航空航天大学" ] }, { "id": "xjtu", "name": "Xi'an Jiaotong University", "exact": [ "Xi'an Jiaotong University", "Xian Jiaotong University", "西安交通大学" ], "exclude": [ "Liverpool" ] }, { "id": "uestc", "name": "University of Electronic Science and Technology of China", "exact": [ "University of Electronic Science and Technology of China", "UESTC", "电子科技大学" ] } ], "institutions_note": "One institution per count. The Platform matches names word by word, so Nanjing University also matches Nanjing University of Aeronautics and Astronautics, and University of Science and Technology of China also matches University of Electronic Science and Technology of China. A row with exclude words leaves out entries that contain them, which makes it a strict count: entries such as Nanjing University, Department of Computer Science and Technology drop out too. The twelve are large mainland universities named in the prompt before any count was run; they are not a ranking of every institution, and others outside the twelve are not counted one by one.", "countries": [ "United States", "China", "Canada", "United Kingdom", "Singapore", "Germany", "Australia", "Switzerland", "France", "Japan", "Hong Kong" ], "lab_order_note": "Rows do not overlap. A lab's row counts people whose current AI-titled job is at that lab and who hold no current job at a lab listed before it, so someone with current jobs at two labs counts once, at the first. The ten-lab row is the sum of the ten rows.", "countries_note": "Values of the profile's location.country. Hong Kong is its own value on the Platform, listed here as a region."}
{ "note": "Institutions located in mainland China, used on education.school.name inside one has_education entry together with education.degree eq Bachelor. A profile counts when one Bachelor entry names an institution on this list and none of the exclude words; nothing is inferred from personal names or languages. The Platform matches both match terms and the names in exact word by word: a name counts when every one of its words appears in the entry, in any order, so a longer name that contains a listed one also counts. Match terms are words that only mainland institutions use in their names (cities, provinces, distinctive names). The exclude words remove entries whose names contain a listed name but belong to an institution elsewhere: National Sun Yat-sen University in Taiwan contains Sun Yat-sen University, and Southeast Missouri State University contains Southeast University. Mainland campuses of foreign and Hong Kong universities count, because the rule is where the institution is. The list is built to leave out institutions in Hong Kong, Macau, and Taiwan. The school country recorded on an entry cannot be queried, so it cannot serve as the rule. The query limit of 64 conditions keeps the match terms to 37; less common institutions are listed by name.", "how_built": "Built from the Bachelor entries of 250 profiles with a current AI title and a current location in China, read in full. The school country recorded on each entry is missing for many mainland institutions (Beihang University, Huazhong University of Science and Technology, and some Tsinghua University entries among them), so each institution's location was decided from its name and reviewed by hand. The list catches 173 of the 173 mainland Bachelor entries in that pool.", "audit": "Two independent samples, read in full. 250 profiles with a current AI title at the ten labs, 25 per lab: the list caught 28 of the 28 mainland Bachelor entries. 200 US-based AI staff with a Bachelor entry at Meta, Amazon, NVIDIA, and Microsoft: the list caught 60 of 63 and matched one institution in Taiwan; the three it missed (Capital Medical University, Southern Medical University, Jimei University) are now listed and the Taiwan match is excluded. On the ten labs' US-based staff the exclude words remove 6 entries and the three added names add 5. Less common institutions still slip through, so shares built on the list are probably slightly low.", "match": [ "Tsinghua", "Peking University", "Fudan", "Tongji", "Nankai", "Beihang", "Renmin", "Huazhong", "Jiaotong", "Jiao Tong", "Xidian", "ShanghaiTech", "SUSTech", "Kunshan", "Ningbo", "Beijing", "Shanghai", "Tianjin", "Chongqing", "Nanjing", "Wuhan", "Harbin", "Hangzhou", "Hefei", "Xiamen", "Dalian", "Qingdao", "Guangzhou", "Shenzhen", "Chengdu", "Zhengzhou", "Jinan", "Zhejiang", "Sichuan", "Shandong", "Jilin", "Hunan" ], "exact": [ "Anhui Agricultural University", "Anhui University", "Capital Medical University", "Capital Normal University", "Capital University of Economics and Business", "Central South University", "Central University of Finance and Economics", "China Agricultural University", "China Pharmaceutical University", "China University of Geosciences", "China University of Mining and Technology", "China University of Petroleum", "Communication University of China", "Donghua University", "Duke Kunshan University", "East China Normal University", "East China University of Science and Technology", "Fujian Normal University", "Fuzhou University", "Guangdong University of Technology", "Guangxi University", "Guizhou University", "Hainan University", "Hebei University of Technology", "Heilongjiang University", "Henan University", "Hohai University", "Hubei University", "Inner Mongolia University", "Jiangnan University", "Jiangsu University", "Jiangxi University of Finance and Economics", "Jilin Agricultural University", "Jimei University", "Kunming University of Science and Technology", "Lanzhou University", "Liaoning University", "Minzu University of China", "Nanchang University", "Nanjing Forestry University", "National University of Defense Technology", "New York University Shanghai", "North China Electric Power University", "Northeast Forestry University", "Northeast Normal University", "Northwest A&F University", "Northwest University", "Northwestern Polytechnical University", "Ocean University of China", "Peking Union Medical College", "Shaanxi Normal University", "Shenyang University of Technology", "Soochow University (CN)", "South China Agricultural University", "South China Normal University", "South China University of Technology", "Southeast University", "Southern Medical University", "Southern University of Science and Technology", "Southwest University", "Sun Yat-Sen University", "Sun Yat-sen University", "Taiyuan University of Technology", "The Chinese University of Hong Kong, Shenzhen", "USTC", "University of Electronic Science and Technology of China", "University of International Business and Economics", "University of Science and Technology of China", "Westlake University", "Xi'an Jiaotong-Liverpool University", "Xinjiang University", "Yunnan University", "上海交通大学", "上海大学", "上海科技大学", "上海财经大学", "东北大学", "东华大学", "东南大学", "中南大学", "中国人民大学", "中国农业大学", "中国海洋大学", "中国科学技术大学", "中央民族大学", "中央财经大学", "中山大学", "佛山大学", "兰州大学", "北京交通大学", "北京大学", "北京师范大学", "北京理工大学", "北京科技大学", "北京航空航天大学", "北京邮电大学", "华东师范大学", "华东理工大学", "华中科技大学", "华南师范大学", "华南理工大学", "南京大学", "南京理工大学", "南京航空航天大学", "南京邮电大学", "南开大学", "南方科技大学", "厦门大学", "合肥工业大学", "吉林大学", "同济大学", "哈尔滨工业大学", "哈尔滨工程大学", "四川大学", "国防科技大学", "复旦大学", "大连理工大学", "天津大学", "对外经济贸易大学", "山东大学", "山东理工大学", "暨南大学", "杭州电子科技大学", "武汉大学", "武汉理工大学", "武汉科技学院", "河海大学", "济南大学", "浙江大学", "深圳大学", "清华大学", "湖南大学", "电子科技大学", "苏州大学", "西交利物浦大学", "西北农林科技大学", "西北工业大学", "西南交通大学", "西安交通大学", "西安电子科技大学", "西湖大学", "郑州大学", "重庆大学", "首都师范大学", "香港中文大学（深圳）" ], "exclude": [ "National Sun Yat-sen", "Taiwan", "Malaysia", "Missouri" ]}
Where they are based
Across every employer, 15,802 profiles in AI roles show a bachelor's degree from a mainland-China institution. 12,194 of them list the United States, followed by Singapore (722), Canada (577), and the United Kingdom (399). These numbers cannot say how many work in China: only 32 AI-role profiles on the whole Platform list China as their country, and employers based in China are barely visible (92 visible profiles in AI roles at Tencent on 2026-09-22).
Run it
Three ways in. Each says what it costs before you start.
1 API Credit buys 25 search results or 5 full records; $1 buys 30 API Credits.
Run it in your agent
235 to 265 API Credits$7.83 to $8.83More than the 100 free API Credits
Your agent stops and asks before spending more than 300 API Credits.
Your agent follows the prompt step by step: it reads the rules, runs the counts, checks the definitions the prompt asks it to check, and writes the files and charts. Use an agent that can write files, such as Claude Code or Codex.
Set up onceKey, connection, and a free check. Skip this if your agent already reaches the Platform.
1Get a key
Create a key on the Metix AI Platform →New accounts get 100 API Credits once, valid for 30 days. Set the key in the shell you start your agent from, or add the line to ~/.zshrc or ~/.bashrc so every new terminal has it:
Shellexport METIX_KEY=metix_xxxxxxxx
2Connect your agent
Claude Code
Registers the Platform for every project. Start claude in any folder and the ten metix tools are there.
MCP setup guide →Shell: "${METIX_KEY:?run step 1 first}" && claude mcp add --scope user --transport http metix \ https://mira-api.metix.ai/mcp \ --header "Authorization: Bearer $METIX_KEY"Codex
Registers the same server. The key stays in your environment instead of the config file.
MCP setup guide →Shellcodex mcp add metix \ --url https://mira-api.metix.ai/mcp \ --bearer-token-env-var METIX_KEY
Skills
Four skills that teach any agent the Platform's endpoints and query rules, for agents without MCP. The installer starts with none ticked: press space on each, then enter.
Skills install guide →Shellnpx skills add MetixAI-Official/metix-skills
Other MCP
Point the client at this endpoint over streamable HTTP with both headers; without the Accept header the server answers 406. Older clients use /sse on the same host.
MCP setup guide →Endpoint and headershttps://mira-api.metix.ai/mcp Authorization: Bearer <your key> Accept: application/json, text/event-stream
3Check the setup
Ask this first. It reads your balance and the field list, runs no search, and costs nothing:
Prompt for your agentUse the Metix AI Platform to check my key status and read the contract; both are free. Then tell me my API Credit balance and which datasets I can query. Do not run any search.
4Paste the prompt
Start your agent in an empty folder, then paste. It writes its files there.
The question
Answer one question with the Metix AI Platform: at ten large AI labs, what share of the people in AI roles earned their bachelor's degree at an institution in mainland China, and where do people in AI roles with that background list themselves now? Work only through the public Platform (REST at https://mira-api.metix.ai, the MCP server, or the metix-skills) with the key in METIX_KEY, and never print the key.
01Read before querying
Call GET /contract (free) and build every condition from querySpecByEntity.profile. Conditions about one job go inside one has_experience entry and conditions about one degree inside one has_education entry, so they describe the same job or degree. A query holds at most 64 conditions and nests at most 6 levels deep. A count with size 1 costs 1 API Credit, so plan every published number as a count. Check the balance with GET /auth/key/status (free) at the start and at the end, and stop and ask before the run passes 300 API Credits.
02Population
An AI role is a current job (experience.is_current eq true) whose experience.title matches any of: machine learning, research scientist, research engineer, deep learning, member of technical staff, applied scientist, artificial intelligence, AI engineer, AI researcher, LLM, NLP, computer vision. The labs, in this order, are OpenAI, Anthropic, Google DeepMind (company names "Google DeepMind" and "DeepMind"), xAI, Meta, NVIDIA, Google, Microsoft, Apple, and Amazon (with "Amazon Web Services (AWS)" and "AWS"). Check every name and variant with a count first. The Platform matches company names word by word, so Google also matches Google DeepMind. Make the rows disjoint: each lab counts people with a current AI role there and no current job at a lab earlier in the order, so the ten-lab figure is the sum of the rows.
03The education rule
A person counts when one education entry has education.degree eq "Bachelor" and an institution located in mainland China. Decide by the institution and never by the person: no names, no languages. The denominator is people with any Bachelor entry. Mainland campuses of foreign and Hong Kong universities count; the list leaves out institutions in Hong Kong, Macau, and Taiwan.
04Build the institution list
The school's country comes back on a detail read but cannot be queried, and it is missing for many mainland institutions. Search 250 profiles with a current AI role located in China (size 250), read them with POST /entity/v1/profiles/detail-by-id and _source on the education fields, and decide each Bachelor institution's location from its name. The Platform matches school names word by word, both match terms and names inside in, so a longer name that contains a listed one also matches. Write match terms only mainland institutions use (cities, provinces, distinctive names such as Tsinghua or Fudan), list other names in full, and add exclude words, in a not inside the same has_education entry, for names elsewhere that contain a listed one (National Sun Yat-sen University in Taiwan contains Sun Yat-sen University; Southeast Missouri State contains Southeast University). Check each exclude word with a count. Stay within the 64 conditions.
05Audit the list
Read 25 AI-role profiles from each lab (250 in all), and check how many mainland Bachelor entries the list catches and whether it matches any institution outside mainland China. Fix the list and check again. Report the sample size and the result, and say that less common institutions can still slip through.
06Location
Compare how often the job-level experience.location.country and the profile's location.country are filled for these people. Use the better covered one for "US-based", keep worldwide numbers as context, and say why.
07Count
For each lab, disjoint as in step 2, US-based: AI roles; with any Bachelor entry; with a mainland-China Bachelor entry; the same two for current jobs that started in the last 24 months (experience.start_date gte "now-24m" in the same entry); the same two for titles that also match "scientist" or "researcher". Add the ten-lab figures as sums, but give no ten-lab total for a column with a suppressed cell. For context, the ten labs worldwide (one query with every name). Then twelve large institutions one at a time at the ten labs, US-based; a name that also matches other universities gets exclude words and is marked a strict count. Then where people in AI roles with a mainland-China Bachelor list themselves, by profile country. Every count of people goes through the small-cell rule (1 to 9 becomes "<10"), and before writing, check every count a reader could get by subtracting published cells; if one is between 1 and 9, suppress more and check again.
08Outputs
Write labs.json, institutions.json, countries.json, and context.json with "unit": "profiles", the snapshot date, and the query files they came from, and commit the institution list with how it was built and audited. Raw records stay in data/raw/ and are never published.
09Charts
The share by lab, US-based, sorted, with the ten-lab share marked; people whose current job started earlier against the last 24 months, for each lab; scientist and researcher titles against the same lab's other AI titles; the twelve institutions as counts, strict counts hatched; countries and regions other than mainland China scaled to the total. Every chart title states its finding in neutral, factual words. A bound keeps its direction when rounded: a floor rounds down, a ceiling rounds up.
10Limits
Counts are visible lower bounds, not headcounts; shares are ratios within visible profiles that list a bachelor's and can sit above or below the share among all staff. People with no Bachelor entry are left out of both sides of every share. Profile location is where a person lists themselves, not always where the job is. Almost no profile lists China as its country, and people whose current job is recorded in China mostly list another country, so the data cannot say how many work in China; say so, give the visible counts for employers based in China, and make no claim about how many work there. The earlier group holds only people still in their job, so a gap between the groups can reflect who stayed as well as who was hired. One day, not a trend. No security, loyalty, or nationality framing.
What you get
The aggregate files and the chart, a note on what the audits found and what they changed, and the API Credits the run spent, read from the balance before and after.
A call that returns 402 insufficient_quota means the key works and the balance is empty.
Reproduce the numbers
104 API Credits$3.47More than the 100 free API Credits
A short standard-library Python script sends the committed queries as counts and writes the aggregate files this page is built from. It needs Python and METIX_KEY set in the shell (step 1 of the agent path); your agent can run these lines for you as well.
curl -fsSL https://platform.metix.ai/casebook/source/china-educated-ai-talent-2026.tar.gz | tar xz
cd china-educated-ai-talent-2026
: "${METIX_KEY:?set METIX_KEY first}" && python3 cases/china-educated-ai-talent-2026/fetch.pyWhat you get
data/*.json and data/receipt.json. Run git diff cases/china-educated-ai-talent-2026/data to see what moved: the numbers should match, apart from what changed in the data since the snapshot.
Adapt it
Costs what your version reads. Write your own ceiling into step 1 of the prompt.
The prompt is the case. Change the parts in this table and your agent answers your question instead, with the same checks and the same way of reporting cost.
| To change | Edit | For example |
|---|---|---|
| The organizations | Step 2, checking each name with a count | Frontier startups: Mistral AI, Cohere, Perplexity |
| The roles | The title terms in step 2 | Data engineering or chip design titles |
| The education rule | Steps 3 to 5 | Institutions in India, built and audited the same way |
What to ask before running it
- Which organizations, and does any name also match a parent company or a different company?
- Which roles, in title words?
- Which country's institutions, and do campuses abroad count?
- US-based, worldwide, or both?
- How many API Credits may the list building and the audit spend on reads?
Answer one question with the Metix AI Platform: at ten large AI labs, what share of the people in AI roles earned their bachelor's degree at an institution in mainland China, and where do people in AI roles with that background list themselves now? Work only through the public Platform (REST at https://mira-api.metix.ai, the MCP server, or the metix-skills) with the key in METIX_KEY, and never print the key. 1. Read before querying. Call GET /contract (free) and build every condition from querySpecByEntity.profile. Conditions about one job go inside one has_experience entry and conditions about one degree inside one has_education entry, so they describe the same job or degree. A query holds at most 64 conditions and nests at most 6 levels deep. A count with size 1 costs 1 API Credit, so plan every published number as a count. Check the balance with GET /auth/key/status (free) at the start and at the end, and stop and ask before the run passes 300 API Credits. 2. Population. An AI role is a current job (experience.is_current eq true) whose experience.title matches any of: machine learning, research scientist, research engineer, deep learning, member of technical staff, applied scientist, artificial intelligence, AI engineer, AI researcher, LLM, NLP, computer vision. The labs, in this order, are OpenAI, Anthropic, Google DeepMind (company names "Google DeepMind" and "DeepMind"), xAI, Meta, NVIDIA, Google, Microsoft, Apple, and Amazon (with "Amazon Web Services (AWS)" and "AWS"). Check every name and variant with a count first. The Platform matches company names word by word, so Google also matches Google DeepMind. Make the rows disjoint: each lab counts people with a current AI role there and no current job at a lab earlier in the order, so the ten-lab figure is the sum of the rows. 3. The education rule. A person counts when one education entry has education.degree eq "Bachelor" and an institution located in mainland China. Decide by the institution and never by the person: no names, no languages. The denominator is people with any Bachelor entry. Mainland campuses of foreign and Hong Kong universities count; the list leaves out institutions in Hong Kong, Macau, and Taiwan. 4. Build the institution list. The school's country comes back on a detail read but cannot be queried, and it is missing for many mainland institutions. Search 250 profiles with a current AI role located in China (size 250), read them with POST /entity/v1/profiles/detail-by-id and _source on the education fields, and decide each Bachelor institution's location from its name. The Platform matches school names word by word, both match terms and names inside in, so a longer name that contains a listed one also matches. Write match terms only mainland institutions use (cities, provinces, distinctive names such as Tsinghua or Fudan), list other names in full, and add exclude words, in a not inside the same has_education entry, for names elsewhere that contain a listed one (National Sun Yat-sen University in Taiwan contains Sun Yat-sen University; Southeast Missouri State contains Southeast University). Check each exclude word with a count. Stay within the 64 conditions. 5. Audit the list. Read 25 AI-role profiles from each lab (250 in all), and check how many mainland Bachelor entries the list catches and whether it matches any institution outside mainland China. Fix the list and check again. Report the sample size and the result, and say that less common institutions can still slip through. 6. Location. Compare how often the job-level experience.location.country and the profile's location.country are filled for these people. Use the better covered one for "US-based", keep worldwide numbers as context, and say why. 7. Count. For each lab, disjoint as in step 2, US-based: AI roles; with any Bachelor entry; with a mainland-China Bachelor entry; the same two for current jobs that started in the last 24 months (experience.start_date gte "now-24m" in the same entry); the same two for titles that also match "scientist" or "researcher". Add the ten-lab figures as sums, but give no ten-lab total for a column with a suppressed cell. For context, the ten labs worldwide (one query with every name). Then twelve large institutions one at a time at the ten labs, US-based; a name that also matches other universities gets exclude words and is marked a strict count. Then where people in AI roles with a mainland-China Bachelor list themselves, by profile country. Every count of people goes through the small-cell rule (1 to 9 becomes "<10"), and before writing, check every count a reader could get by subtracting published cells; if one is between 1 and 9, suppress more and check again. 8. Outputs. Write labs.json, institutions.json, countries.json, and context.json with "unit": "profiles", the snapshot date, and the query files they came from, and commit the institution list with how it was built and audited. Raw records stay in data/raw/ and are never published. 9. Charts. The share by lab, US-based, sorted, with the ten-lab share marked; people whose current job started earlier against the last 24 months, for each lab; scientist and researcher titles against the same lab's other AI titles; the twelve institutions as counts, strict counts hatched; countries and regions other than mainland China scaled to the total. Every chart title states its finding in neutral, factual words. A bound keeps its direction when rounded: a floor rounds down, a ceiling rounds up. 10. Limits. Counts are visible lower bounds, not headcounts; shares are ratios within visible profiles that list a bachelor's and can sit above or below the share among all staff. People with no Bachelor entry are left out of both sides of every share. Profile location is where a person lists themselves, not always where the job is. Almost no profile lists China as its country, and people whose current job is recorded in China mostly list another country, so the data cannot say how many work in China; say so, give the visible counts for employers based in China, and make no claim about how many work there. The earlier group holds only people still in their job, so a gap between the groups can reflect who stayed as well as who was hired. One day, not a trend. No security, loyalty, or nationality framing.
Method and limits
How the population was defined, counted, and checked, and what the numbers cannot show.
Population
Profiles on the Metix AI Platform on September 22, 2026 whose current job (experience.is_current true) is at one of the ten labs with a title matching one of 12 AI terms. The lab, the title, and the start-date condition sit in one has_experience entry, so they describe the same job. Google DeepMind covers "Google DeepMind" and "DeepMind", and Amazon includes AWS. The rows do not overlap: someone with current jobs at two labs counts once, at the first in the list. The Platform matches company names word by word, so "Google" also matches Google DeepMind; Google DeepMind comes first, so its staff count there only.
The institution list
The agent read 250 profiles of AI staff whose current job is in China, collected their bachelor's institutions, and decided each one's location from the institution itself (the recorded school country is often missing and cannot be queried). The list has 37 match words and 145 institution names. The Platform matches both word by word: a name counts when each of its words appears in the entry, so longer names match too, and 4 exclude words remove the ones that belong elsewhere, such as National Sun Yat-sen University in Taiwan, which contains every word of Sun Yat-sen University. Two independent checks: 250 profiles at the ten labs, where the list caught all 28 mainland bachelor's entries, and 200 US-based profiles at Meta, Amazon, NVIDIA, and Microsoft, where it caught 60 of 63 and matched one institution in Taiwan. The three it missed are now listed and the Taiwan match is excluded. Less common institutions can still slip through, so the shares are probably slightly low. The list is in queries/institutions.json.
US-based
The profile's own location.country: every profile in this population lists one, while the job's location is often missing.
Recent jobs and title types
"Current job started in the last 24 months" means it started no earlier than 24 months before the snapshot. Scientist or researcher means the title contains either word; Applied Scientist counts. Whether a gap is larger than chance uses a two-proportion z test that treats the visible profiles as a sample, which is an approximation. Figure 03 makes 7 comparisons, so a |z| between 1.96 and 2.69 is called marginal.
Small numbers
Counts under 10 show as "<10". Because the rows do not overlap and the ten-lab row is their sum, a column with suppressed cells gets no ten-lab total, or subtraction would give them back; that is why figure 03 has none. Before writing any file, the reproduce script checks every count that can be derived by subtraction and stops if one is between 1 and 9.
Limits
Every count is a lower bound on the number of people. The shares are ratios within visible profiles that list a bachelor's, and can sit above or below the share among all US-based AI staff at these labs. The population is defined by where the institution is, never by nationality. Profiles based in China are almost invisible, so nothing here counts how many work there. One day, not a trend.
The last reproduction
What reproducing this case cost the last time the script ran, read from the Platform's own balance before and after.
- Ran on
- 2026-09-22
- Calls
- 105
- Search results
- 104
- Records read
- 0
- API Credits
- 104
Search results are IDs returned by searches, one per count query and one per match on a full search. Records are postings or profiles read in full: this case reads none.
Making this case cost about 203 API Credits more: the agent's audits, trial queries, and earlier runs that the published replay replaced. You do not pay that again.