AI 晶片供給份額｜資料解說 | Trend Core

資料：EPOCH AI ／ 中文整理：TREND CORE

AI 晶片供給份額

比較各設計公司在當季晶片供給中的算力份額，讀懂競爭結構，也保留資料缺口。

各設計公司的當季 AI 算力供給份額

圖表可左右滑動，最新季度在右側；完整數字可展開下方資料表。

各設計公司的當季 AI 算力供給份額依季度堆疊的晶片設計商算力份額；完整數據見圖下資料表。各設計公司的當季 AI 算力供給份額份額 %｜當季新增，非累積存量｜* 資料不完整0%25%50%75%100%2024Q1 · AMD：6.6%2024Q1 · Amazon：4%2024Q1 · Cambricon：0%2024Q1 · Google：14.7%2024Q1 · 華為：5.4%2024Q1 · NVIDIA：69.2%2024Q12024Q2 · AMD：11.5%2024Q2 · Amazon：4.2%2024Q2 · Cambricon：0%2024Q2 · Google：13.7%2024Q2 · 華為：4.4%2024Q2 · NVIDIA：66.3%2024Q22024Q3 · AMD：12.9%2024Q3 · Amazon：5.8%2024Q3 · Cambricon：0.1%2024Q3 · Google：14%2024Q3 · 華為：3.4%2024Q3 · NVIDIA：63.8%2024Q32024Q4 · AMD：10.7%2024Q4 · Amazon：7.7%2024Q4 · Cambricon：0.3%2024Q4 · Google：14.7%2024Q4 · 華為：2.3%2024Q4 · NVIDIA：64.3%2024Q42025Q1 · AMD：6.2%2025Q1 · Amazon：5.4%2025Q1 · Cambricon：0.2%2025Q1 · Google：17.1%2025Q1 · 華為：5.1%2025Q1 · NVIDIA：65.9%2025Q12025Q2 · AMD：3.6%2025Q2 · Amazon：5.5%2025Q2 · Cambricon：0.3%2025Q2 · Google：21.5%2025Q2 · 華為：4.5%2025Q2 · NVIDIA：64.6%2025Q22025Q3 · AMD：5.5%2025Q3 · Amazon：5.1%2025Q3 · Cambricon：0.2%2025Q3 · Google：21.6%2025Q3 · 華為：3.6%2025Q3 · NVIDIA：64.1%2025Q32025Q4 · AMD：5%2025Q4 · Amazon：5%2025Q4 · Cambricon：0.2%2025Q4 · Google：19.8%2025Q4 · 華為：3%2025Q4 · NVIDIA：67%2025Q42026Q1 · AMD：4.6%2026Q1 · Amazon：4.5%2026Q1 · Cambricon：0.2%2026Q1 · Google：18.5%2026Q1 · 華為：4.1%2026Q1 · NVIDIA：68%2026Q12026Q2 · AMD：4.1%2026Q2 · Amazon：4.6%2026Q2 · Cambricon：0.2%2026Q2 · Google：25.4%2026Q2 · 華為：3.2%2026Q2 · NVIDIA：62.5%2026Q22026Q3 * · Google：37.8%，不完整2026Q3 * · NVIDIA：62.2%，不完整2026Q3 *AMDAmazonCambriconGoogle華為NVIDIAEpoch AI ／ Epoch AI｜AI chip market share by designer｜Trend Core 重繪https://epoch.ai/graphs/ai-chip-market-share-by-designer｜CC BY 4.0 https://creativecommons.org/licenses/by/4.0/期間 2024-01-01 至 2026-09-30｜快照 20261019T021600187369+0000-176312511ce6｜版本 10e433e46f916e467295d944df128372c796067cc8f4fcad3d144cc618f01ec6中位估計，不是精確測量；* 資料不完整。來源列上下界不代表總和信賴區間。開啟原文CC BY 4.0

中位估計，並非精確測量。* 與斜線表示不完整，? 表示未知。

資料與原始分析：Epoch AI ／ AI chip market share by designer · Epoch AI · CC BY 4.0 · Trend Core 重繪

資料期間 2024-01-01 至 2026-09-30；快照 20261019T021600187369+0000-176312511ce6；版本 10e433e46f916e467295d944df128372c796067cc8f4fcad3d144cc618f01ec6

2026Q3，資料不完整：當季新增供給中位估計 1,890,556 H100e；NVIDIA 占 62.2%。本季未列 AMD、Amazon、Cambricon、華為，不代表零供給；份額僅以有列出的估計為分母。

每一季的色塊合計為 100%，分母是這份資料追蹤到的當季晶片供給所代表的 H100 等效算力。它比較的是設計公司的份額，不是晶片顆數、銷售金額或晶圓代工廠的份額。

份額下降不一定代表供給減少。若整體供給增加得更快，一家公司新增的絕對算力仍可能上升，只是在總量中占比較小。可搭配每季總算力與設計公司數值一起閱讀。

本圖從 2024 年第一季呈現，依下載檔重算各季份額；最新季度如被來源標為不完整，就保留標記。來源網頁的摘要文字與下載檔涵蓋季度不同，本頁不沿用摘要中的舊季度份額，也不手寫固定的當前排名。 某季未列出一家公司的資料，不代表該公司供給為零；涵蓋不全的季度不宜與完整季度直接比較。

定義與閱讀限制

H100 等效算力，H100e
以一顆 NVIDIA H100 的峰值密集 8 位元運算能力為基準，將各型晶片的規格與估計數量換成共同尺度。這是紙面算力換算，不能直接代表實際工作負載的執行速度。
不完整季度
來源以 Incomplete 標記部分估計，例如財報季度與日曆季度未完全重合。保留數值供閱讀，但不把它當成已涵蓋整季的最終結果。
算力份額
某設計公司當季的 H100 等效算力，除以本資料所有追蹤公司當季的合計算力。百分比只在這個涵蓋範圍內成立。
設計公司
負責設計晶片的公司，例如 NVIDIA、Google；不同於實際製造晶片的晶圓代工廠。CSV 的 Chip manufacturer 欄在此按設計公司解讀。
供給與部署
銷售、出貨或交付後，晶片仍可能等待機櫃、網路與電力設備安裝，不能據此認定已在機房運轉。
  • 下載檔實際包含 NVIDIA、Google、Amazon、AMD、Huawei、Cambricon 六家；資料集常見問答仍將 Cambricon 列為排除，顯示不同時點內容未完全同步。本頁按下載檔列示並保留這項差異。
  • 來源圖頁摘要停在 2026 年第二季，下載檔則包含不完整的 2026 年第三季。季度是否完整以原始列標記為準，不因日曆季度已結束而取消。
  • Microsoft Maia、Meta MTIA、Tesla Dojo、消費級 GPU 與部分較小設計公司未完整涵蓋，因此這是追蹤範圍內的份額，不能視為完整全球市場普查。
  • H100 等效算力依紙面 8 位元規格換算，未涵蓋實際軟體、記憶體、網路與利用率差異；份額不是營收、獲利或實際部署量。
  • 各公司資料起點與估計可信度不同；本圖從 2024 年起呈現，可降低早期涵蓋差異造成的誤讀，但不能消除估計不確定性。

從資料走向公司與專案研究

AI 機房與資本支出總覽微軟：錢花到哪裡Google：錢花到哪裡NVIDIA 資金流向
完整數據與原始估計範圍
季度總量與資料完整程度
季度當季新增 H100e 中位估計完整程度
2024Q1827,913.8來源未標示不完整
2024Q21,032,296.8來源未標示不完整
2024Q31,308,806.8來源未標示不完整
2024Q41,924,022.8來源未標示不完整
2025Q12,571,749.8來源未標示不完整
2025Q22,928,149.8來源未標示不完整
2025Q33,702,941.8來源未標示不完整
2025Q44,411,110.8來源未標示不完整
2026Q14,980,154來源未標示不完整
2026Q26,327,416來源未標示不完整
2026Q31,890,556不完整：Google、NVIDIA；未列：AMD、Amazon、Cambricon、華為，不視為零
各晶片設計商季度供給
季度晶片設計商H100e 中位估計份額完整程度
2024Q1AMD55,0416.6%來源未標示不完整
2024Q1Amazon33,1904%來源未標示不完整
2024Q1Cambricon1420%來源未標示不完整
2024Q1Google121,51314.7%來源未標示不完整
2024Q1華為45,027.85.4%來源未標示不完整
2024Q1NVIDIA573,00069.2%來源未標示不完整
2024Q2AMD118,55311.5%來源未標示不完整
2024Q2Amazon43,0364.2%來源未標示不完整
2024Q2Cambricon2150%來源未標示不完整
2024Q2Google141,22813.7%來源未標示不完整
2024Q2華為45,027.84.4%來源未標示不完整
2024Q2NVIDIA684,23766.3%來源未標示不完整
2024Q3AMD169,48712.9%來源未標示不完整
2024Q3Amazon75,8565.8%來源未標示不完整
2024Q3Cambricon6640.1%來源未標示不完整
2024Q3Google182,66914%來源未標示不完整
2024Q3華為45,027.83.4%來源未標示不完整
2024Q3NVIDIA835,10363.8%來源未標示不完整
2024Q4AMD205,33810.7%來源未標示不完整
2024Q4Amazon148,0597.7%來源未標示不完整
2024Q4Cambricon5,4470.3%來源未標示不完整
2024Q4Google283,36214.7%來源未標示不完整
2024Q4華為45,027.82.3%來源未標示不完整
2024Q4NVIDIA1,236,78964.3%來源未標示不完整
2025Q1AMD160,2296.2%來源未標示不完整
2025Q1Amazon137,8425.4%來源未標示不完整
2025Q1Cambricon4,8690.2%來源未標示不完整
2025Q1Google440,86717.1%來源未標示不完整
2025Q1華為131,963.85.1%來源未標示不完整
2025Q1NVIDIA1,695,97965.9%來源未標示不完整
2025Q2AMD104,9353.6%來源未標示不完整
2025Q2Amazon160,8165.5%來源未標示不完整
2025Q2Cambricon7,7490.3%來源未標示不完整
2025Q2Google629,72221.5%來源未標示不完整
2025Q2華為131,963.84.5%來源未標示不完整
2025Q2NVIDIA1,892,96464.6%來源未標示不完整
2025Q3AMD202,4535.5%來源未標示不完整
2025Q3Amazon187,0715.1%來源未標示不完整
2025Q3Cambricon7,5630.2%來源未標示不完整
2025Q3Google800,81321.6%來源未標示不完整
2025Q3華為131,963.83.6%來源未標示不完整
2025Q3NVIDIA2,373,07864.1%來源未標示不完整
2025Q4AMD218,3705%來源未標示不完整
2025Q4Amazon219,8915%來源未標示不完整
2025Q4Cambricon8,2770.2%來源未標示不完整
2025Q4Google875,05519.8%來源未標示不完整
2025Q4華為131,963.83%來源未標示不完整
2025Q4NVIDIA2,957,55467%來源未標示不完整
2026Q1AMD229,5494.6%來源未標示不完整
2026Q1Amazon226,4554.5%來源未標示不完整
2026Q1Cambricon12,3980.2%來源未標示不完整
2026Q1Google923,37418.5%來源未標示不完整
2026Q1華為202,4164.1%來源未標示不完整
2026Q1NVIDIA3,385,96268%來源未標示不完整
2026Q2AMD260,9204.1%來源未標示不完整
2026Q2Amazon288,2054.6%來源未標示不完整
2026Q2Cambricon12,3650.2%來源未標示不完整
2026Q2Google1,607,42725.4%來源未標示不完整
2026Q2華為204,6763.2%來源未標示不完整
2026Q2NVIDIA3,953,82362.5%來源未標示不完整
2026Q3Google715,33937.8%不完整
2026Q3NVIDIA1,175,21762.2%不完整

來源列保留數值精度，圖表與摘要適度四捨五入。上下界未知不補零，不得逐列相加當成總和信賴區間。

來源列與原始估計範圍
來源列設計商開始結束H100e 中位估計來源下界來源上界完整程度／原註記
Q1 2022 - A100NVIDIA2022-01-012022-03-31668835154787166Estimates generated on: 08-27-2026 15:06
Q2 2022 - A100NVIDIA2022-04-012022-06-30714725508393146Estimates generated on: 08-27-2026 15:06
Q3 2022 - A100NVIDIA2022-07-012022-09-30680225242588650Estimates generated on: 08-27-2026 15:06
Q3 2022 - H100/H200NVIDIA2022-07-012022-09-30611951267199Estimates generated on: 08-27-2026 15:06
Q4 2022 - TPU v4iGoogle2022-10-012022-12-3114358842182不完整。Incomplete: based on Broadcom fiscal quarters beginning 10/31/2022. Estimates generated on: 09-30-2026 14:07
Q4 2022 - TPU v4Google2022-10-012022-12-319232669912186不完整。Incomplete: based on Broadcom fiscal quarters beginning 10/31/2022. Estimates generated on: 09-30-2026 14:07
Q4 2022 - TPU v5eGoogle2022-10-012022-12-31311319614658不完整。Incomplete: based on Broadcom fiscal quarters beginning 10/31/2022. Estimates generated on: 09-30-2026 14:07
Q4 2022 - A100NVIDIA2022-10-012022-12-31358502762946721Estimates generated on: 08-27-2026 15:06
Q4 2022 - H100/H200NVIDIA2022-10-012022-12-31324812721238218Estimates generated on: 08-27-2026 15:06
Q4 2022 - A800NVIDIA2022-10-012022-12-318870678411506Estimates generated on: 08-27-2026 15:06
Q1 2023 - TPU v4iGoogle2023-01-012023-03-316714131020Estimates generated on: 09-30-2026 14:07
Q1 2023 - TPU v4Google2023-01-012023-03-31165901182222242Estimates generated on: 09-30-2026 14:07
Q1 2023 - TPU v5eGoogle2023-01-012023-03-317773372212465Estimates generated on: 09-30-2026 14:07
Q1 2023 - A100NVIDIA2023-01-012023-03-31186101434324254Estimates generated on: 08-27-2026 15:06
Q1 2023 - H100/H200NVIDIA2023-01-012023-03-31566404745266645Estimates generated on: 08-27-2026 15:06
Q1 2023 - A800NVIDIA2023-01-012023-03-319240706711986Estimates generated on: 08-27-2026 15:06
Q1 2023 - H800NVIDIA2023-01-012023-03-31734259239136Estimates generated on: 08-27-2026 15:06
Q2 2023 - TPU v4Google2023-04-012023-06-309201518413700Estimates generated on: 09-30-2026 14:07
Q2 2023 - TPU v5eGoogle2023-04-012023-06-30239431571933616Estimates generated on: 09-30-2026 14:07
Q2 2023 - A100NVIDIA2023-04-012023-06-30203401567526508Estimates generated on: 08-27-2026 15:06
Q2 2023 - H100/H200NVIDIA2023-04-012023-06-30151413126851178158Estimates generated on: 08-27-2026 15:06
Q2 2023 - A800NVIDIA2023-04-012023-06-308319636310792Estimates generated on: 08-27-2026 15:06
Q2 2023 - H800NVIDIA2023-04-012023-06-30319162574739718Estimates generated on: 08-27-2026 15:06
Q3 2023 - TPU v4Google2023-07-012023-09-3015564482736Estimates generated on: 09-30-2026 14:07
Q3 2023 - TPU v5eGoogle2023-07-012023-09-30497053553266184Estimates generated on: 09-30-2026 14:07
Q3 2023 - A100NVIDIA2023-07-012023-09-30173271335422581Estimates generated on: 08-27-2026 15:06
Q3 2023 - H100/H200NVIDIA2023-07-012023-09-30260088217897306028Estimates generated on: 08-27-2026 15:06
Q3 2023 - A800NVIDIA2023-07-012023-09-308987687411658Estimates generated on: 08-27-2026 15:06
Q3 2023 - H800NVIDIA2023-07-012023-09-30563334544670105Estimates generated on: 08-27-2026 15:06
Q4 2023 - TPU v5eGoogle2023-10-012023-12-317719956024101014Estimates generated on: 09-30-2026 14:07
Q4 2023 - TPU v5pGoogle2023-10-012023-12-316693345510381Estimates generated on: 09-30-2026 14:07
Q4 2023 - A100NVIDIA2023-10-012023-12-31476636736212Estimates generated on: 08-27-2026 15:06
Q4 2023 - H100/H200NVIDIA2023-10-012023-12-31423514354811498320Estimates generated on: 08-27-2026 15:06
Q4 2023 - A800NVIDIA2023-10-012023-12-31286521923717Estimates generated on: 08-27-2026 15:06
Q4 2023 - H800NVIDIA2023-10-012023-12-31198381600424688Estimates generated on: 08-27-2026 15:06
Q1 2024 - TPU v5eGoogle2024-01-012024-03-318284759719109791Estimates generated on: 09-30-2026 14:07
Q1 2024 - TPU v5pGoogle2024-01-012024-03-31386662674051745Estimates generated on: 09-30-2026 14:07
Q1 2024 - H100/H200NVIDIA2024-01-012024-03-31552252462665649798Estimates generated on: 08-27-2026 15:06
Q1 2024 - H20NVIDIA2024-01-012024-03-31207481596326991Estimates generated on: 08-27-2026 15:06
FY24 Q1 - MI250XAMD2024-01-012024-03-31587939548280Estimates generated on: 08-27-2026 16:14
FY24 Q1 - MI300XAMD2024-01-012024-03-31456323303463037Estimates generated on: 08-27-2026 16:14
FY24 Q1 - MI300AAMD2024-01-012024-03-31353011096293Estimates generated on: 08-27-2026 16:14
Q1 2024 Ascend 910B華為2024-01-012024-03-3135378.7527362.7545990.5Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q1 2024 Ascend 910C華為2024-01-012024-03-3196494876.7518995.25Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q1 2024 - Trainium1Amazon2024-01-012024-03-31233451061153057Estimates generated on: 10-14-2026 14:34
Q1 2024 - Trainium2Amazon2024-01-012024-03-319845328116409Estimates generated on: 10-14-2026 14:34
2024 Q1 - Siyuan 590Cambricon2024-01-012024-03-31142136153Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; quarterly allocation proportional to reported quarterly revenue
Q2 2024 - TPU v5eGoogle2024-04-012024-06-307436449752105356Estimates generated on: 09-30-2026 14:07
Q2 2024 - TPU v5pGoogle2024-04-012024-06-30668644512392622Estimates generated on: 09-30-2026 14:07
Q2 2024 - H100/H200NVIDIA2024-04-012024-06-30650874545290765841Estimates generated on: 08-27-2026 15:06
Q2 2024 - H20NVIDIA2024-04-012024-06-30333632566943403Estimates generated on: 08-27-2026 15:06
FY24 Q2 - MI250XAMD2024-04-012024-06-3012762052394Estimates generated on: 08-27-2026 16:14
FY24 Q2 - MI300XAMD2024-04-012024-06-3010664283327135478Estimates generated on: 08-27-2026 16:14
FY24 Q2 - MI300AAMD2024-04-012024-06-3010635379517597Estimates generated on: 08-27-2026 16:14
Q2 2024 Ascend 910B華為2024-04-012024-06-3035378.7527362.7545990.5Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q2 2024 Ascend 910C華為2024-04-012024-06-3096494876.7518995.25Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q2 2024 - Trainium1Amazon2024-04-012024-06-30233451061153057Estimates generated on: 10-14-2026 14:34
Q2 2024 - Trainium2Amazon2024-04-012024-06-30196911312732819Estimates generated on: 10-14-2026 14:34
2024 Q2 - Siyuan 590Cambricon2024-04-012024-06-30215207233Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; quarterly allocation proportional to reported quarterly revenue
Q3 2024 - TPU v5eGoogle2024-07-012024-09-307085641815107090Estimates generated on: 09-30-2026 14:07
Q3 2024 - TPU v5pGoogle2024-07-012024-09-307404045412109147Estimates generated on: 09-30-2026 14:07
Q3 2024 - TPU v6eGoogle2024-07-012024-09-30377731568366996Estimates generated on: 09-30-2026 14:07
Q3 2024 - H100/H200NVIDIA2024-07-012024-09-30794383665518934698Estimates generated on: 08-27-2026 15:06
Q3 2024 - H20NVIDIA2024-07-012024-09-30407203132952972Estimates generated on: 08-27-2026 15:06
FY24 Q3 - MI250XAMD2024-07-012024-09-308482551503Estimates generated on: 08-27-2026 16:14
FY24 Q3 - MI300XAMD2024-07-012024-09-3012586699498156137Estimates generated on: 08-27-2026 16:14
FY24 Q3 - MI300AAMD2024-07-012024-09-30427733225654899Estimates generated on: 08-27-2026 16:14
Q3 2024 Ascend 910B華為2024-07-012024-09-3035378.7527362.7545990.5Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q3 2024 Ascend 910C華為2024-07-012024-09-3096494876.7518995.25Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q3 2024 - Trainium1Amazon2024-07-012024-09-30233451061153057Estimates generated on: 10-14-2026 14:34
Q3 2024 - Trainium2Amazon2024-07-012024-09-30525113281968921Estimates generated on: 10-14-2026 14:34
2024 Q3 - Siyuan 590Cambricon2024-07-012024-09-30664637717Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; quarterly allocation proportional to reported quarterly revenue
Q4 2024 - TPU v5eGoogle2024-10-012024-12-31596513653289524Estimates generated on: 09-30-2026 14:07
Q4 2024 - TPU v5pGoogle2024-10-012024-12-31621993905190751Estimates generated on: 09-30-2026 14:07
Q4 2024 - TPU v6eGoogle2024-10-012024-12-31161512107878225856Estimates generated on: 09-30-2026 14:07
Q4 2024 - H100/H200NVIDIA2024-10-012024-12-31681105570616801411Estimates generated on: 08-27-2026 15:06
Q4 2024 - H20NVIDIA2024-10-012024-12-31485173732963116Estimates generated on: 08-27-2026 15:06
Q4 2024 - B200NVIDIA2024-10-012024-12-31507167432569594047Estimates generated on: 08-27-2026 15:06
FY24 Q4 - MI250XAMD2024-10-012024-12-316903461114Estimates generated on: 08-27-2026 16:14
FY24 Q4 - MI300XAMD2024-10-012024-12-31152770121506188325Estimates generated on: 08-27-2026 16:14
FY24 Q4 - MI300AAMD2024-10-012024-12-31518783947864597Estimates generated on: 08-27-2026 16:14
Q4 2024 Ascend 910B華為2024-10-012024-12-3135378.7527362.7545990.5Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q4 2024 Ascend 910C華為2024-10-012024-12-3196494876.7518995.25Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q4 2024 - Trainium1Amazon2024-10-012024-12-31233451061153057Estimates generated on: 10-14-2026 14:34
Q4 2024 - Trainium2Amazon2024-10-012024-12-3112471482049173943Estimates generated on: 10-14-2026 14:34
2024 Q4 - Siyuan 590Cambricon2024-10-012024-12-31544752295882Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; quarterly allocation proportional to reported quarterly revenue
Q1 2025 - TPU v5eGoogle2025-01-012025-03-31190091237528027Estimates generated on: 09-30-2026 14:07
Q1 2025 - TPU v5pGoogle2025-01-012025-03-31441302001569248Estimates generated on: 09-30-2026 14:07
Q1 2025 - TPU v6eGoogle2025-01-012025-03-31377728277636494191Estimates generated on: 09-30-2026 14:07
Q1 2025 - H100/H200NVIDIA2025-01-012025-03-31329018275645387134Estimates generated on: 08-27-2026 15:06
Q1 2025 - H20NVIDIA2025-01-012025-03-31540964162170374Estimates generated on: 08-27-2026 15:06
Q1 2025 - B200NVIDIA2025-01-012025-03-31131286511197541537764Estimates generated on: 08-27-2026 15:06
FY25 Q1 - MI300XAMD2025-01-012025-03-3111292583505148965Estimates generated on: 08-27-2026 16:14
FY25 Q1 - MI300AAMD2025-01-012025-03-31460021867678Estimates generated on: 08-27-2026 16:14
FY25 Q1 - MI325XAMD2025-01-012025-03-31427042598468328Estimates generated on: 08-27-2026 16:14
Q1 2025 Ascend 910B華為2025-01-012025-03-3116816.511946.523883.25Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q1 2025 Ascend 910C華為2025-01-012025-03-31115147.2597198.5136010.75Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q1 2025 - Trainium2Amazon2025-01-012025-03-31137842118150164098Estimates generated on: 10-14-2026 14:34
2025 Q1 - Siyuan 590Cambricon2025-01-012025-03-31486944256638Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; quarterly allocation proportional to reported quarterly revenue
Q2 2025 - TPU v5pGoogle2025-04-012025-06-3014505413224519Estimates generated on: 09-30-2026 14:07
Q2 2025 - TPU v6eGoogle2025-04-012025-06-30615217470078769444Estimates generated on: 09-30-2026 14:07
Q2 2025 - H100/H200NVIDIA2025-04-012025-06-30228707191607269105Estimates generated on: 08-27-2026 15:06
Q2 2025 - H20NVIDIA2025-04-012025-06-30228381757129710Estimates generated on: 08-27-2026 15:06
Q2 2025 - B200NVIDIA2025-04-012025-06-3010977319362651285778Estimates generated on: 08-27-2026 15:06
Q2 2025 - B300NVIDIA2025-04-012025-06-30543688460596638832Estimates generated on: 08-27-2026 15:06
FY25 Q2 - MI300XAMD2025-04-012025-06-30607494093486348Estimates generated on: 08-27-2026 16:14
FY25 Q2 - MI325XAMD2025-04-012025-06-30346732079157090Estimates generated on: 08-27-2026 16:14
FY25 Q2 - MI350XAMD2025-04-012025-06-309513386916362Estimates generated on: 08-27-2026 16:14
Q2 2025 Ascend 910B華為2025-04-012025-06-3016816.511946.523883.25Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q2 2025 Ascend 910C華為2025-04-012025-06-30115147.2597198.5136010.75Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q2 2025 - Trainium2Amazon2025-04-012025-06-30160816141124187071Estimates generated on: 10-14-2026 14:34
2025 Q2 - Siyuan 590Cambricon2025-04-012025-06-307749704410567Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; quarterly allocation proportional to reported quarterly revenue
Q3 2025 - TPU v6eGoogle2025-07-012025-09-30681240517161864478Estimates generated on: 09-30-2026 14:07
Q3 2025 - TPU v7Google2025-07-012025-09-3011957353913183569Estimates generated on: 09-30-2026 14:07
Q3 2025 - H100/H200NVIDIA2025-07-012025-09-30127307106655149793Estimates generated on: 08-27-2026 15:06
Q3 2025 - H20NVIDIA2025-07-012025-09-30256019693330Estimates generated on: 08-27-2026 15:06
Q3 2025 - B200NVIDIA2025-07-012025-09-308744017457851024191Estimates generated on: 08-27-2026 15:06
Q3 2025 - B300NVIDIA2025-07-012025-09-30136881011596131608347Estimates generated on: 08-27-2026 15:06
FY25 Q3 - MI300XAMD2025-07-012025-09-30606943790788542Estimates generated on: 08-27-2026 16:14
FY25 Q3 - MI325XAMD2025-07-012025-09-30257961034746573Estimates generated on: 08-27-2026 16:14
FY25 Q3 - MI350XAMD2025-07-012025-09-308439764076109082Estimates generated on: 08-27-2026 16:14
FY25 Q3 - MI355XAMD2025-07-012025-09-30315662071644515Estimates generated on: 08-27-2026 16:14
Q3 2025 Ascend 910B華為2025-07-012025-09-3016816.511946.523883.25Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q3 2025 Ascend 910C華為2025-07-012025-09-30115147.2597198.5136010.75Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q3 2025 - Trainium2Amazon2025-07-012025-09-30187071164098216609Estimates generated on: 10-14-2026 14:34
2025 Q3 - Siyuan 590Cambricon2025-07-012025-09-307563687710313Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; quarterly allocation proportional to reported quarterly revenue
Q4 2025 - TPU v6eGoogle2025-10-012025-12-31378850217549543071Estimates generated on: 09-30-2026 14:07
Q4 2025 - TPU v7Google2025-10-012025-12-31496205343980668417Estimates generated on: 09-30-2026 14:07
Q4 2025 - H100/H200NVIDIA2025-10-012025-12-31233841959127515Estimates generated on: 08-27-2026 15:06
Q4 2025 - B200NVIDIA2025-10-012025-12-3110551318999311235879Estimates generated on: 08-27-2026 15:06
Q4 2025 - B300NVIDIA2025-10-012025-12-31187903915918642207867Estimates generated on: 08-27-2026 15:06
FY25 Q4 - MI300XAMD2025-10-012025-12-3114795885622778Estimates generated on: 08-27-2026 16:14
FY25 Q4 - MI325X AMD2025-10-012025-12-3116786813230540Estimates generated on: 08-27-2026 16:14
FY25 Q4 - MI350XAMD2025-10-012025-12-318202856529110056Estimates generated on: 08-27-2026 16:14
FY25 Q4 - MI355XAMD2025-10-012025-12-319426369459124606Estimates generated on: 08-27-2026 16:14
Q4 2025 Ascend 910B華為2025-10-012025-12-3116816.511946.523883.25Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q4 2025 Ascend 910C華為2025-10-012025-12-31115147.2597198.5136010.75Confidence intervals in units are relatively unprincipled due to lack of revenue model. Don't take them too literally. H100e conversion incorporates some uncertainty over average chip specifications. See methodology
Q4 2025 - Trainium2Amazon2025-10-012025-12-31219891187071255992Estimates generated on: 10-14-2026 14:34
FY25 Q4 - MI308XAMD2025-10-012025-12-3110498855812883Estimates generated on: 08-27-2026 16:14
2025 Q4 - Siyuan 590Cambricon2025-10-012025-12-318277752511287Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; quarterly allocation proportional to reported quarterly revenue
Q1 2026 - TPU v6eGoogle2026-01-012026-03-3114385120600260010Estimates generated on: 09-30-2026 14:07
Q1 2026 - TPU v7Google2026-01-012026-03-317795235738661012205Estimates generated on: 09-30-2026 14:07
Q1 2026 - B200NVIDIA2026-01-012026-03-318774387483751027746Estimates generated on: 08-27-2026 15:06
Q1 2026 - B300NVIDIA2026-01-012026-03-31250852421251442947508Estimates generated on: 08-27-2026 15:06
FY26 Q1 - MI300XAMD2026-01-012026-03-31203351256431216Estimates generated on: 08-27-2026 16:14
FY26 Q1 - MI325XAMD2026-01-012026-03-3117196983728954Estimates generated on: 08-27-2026 16:14
FY26 Q1 - MI350XAMD2026-01-012026-03-317504052781102291Estimates generated on: 08-27-2026 16:14
FY26 Q1 - MI355XAMD2026-01-012026-03-3111697885580146034Estimates generated on: 08-27-2026 16:14
Q1 2026 Ascend 910C華為2026-01-012026-03-3120241665000368000Estimates generated on: 10-01-2026 17:15; H1 2026 update. See methodology: https://docs.google.com/document/d/1nPlAAB0HioI9LSk6u-LGEnVBvSAoc50MGXmMxOLv0fk/edit?usp=sharing
Q1 2026 - Trainium2Amazon2026-01-012026-03-31226455147688344605Estimates generated on: 10-14-2026 14:34
2026 Q1 - Siyuan 590Cambricon2026-01-012026-03-31123981130013700Estimates generated on: 10-01-2026 17:15; Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; units from reported quarterly revenue and gross margin over estimated cost per chip
Q2 2026 - TPU v6eGoogle2026-04-012026-06-3038772075813Estimates generated on: 09-30-2026 14:07
Q2 2026 - TPU v7Google2026-04-012026-06-30143772710831761807234Estimates generated on: 09-30-2026 14:07
Q2 2026 - H100/H200NVIDIA2026-04-012026-06-30193551621622774Estimates generated on: 08-27-2026 15:06
Q2 2026 - B200NVIDIA2026-04-012026-06-30563976481020660588Estimates generated on: 08-27-2026 15:06
Q2 2026 - B300NVIDIA2026-04-012026-06-30337049228553763960320Estimates generated on: 08-27-2026 15:06
FY26 Q2 - MI300XAMD2026-04-012026-06-30222841400733099Estimates generated on: 08-27-2026 16:14
FY26 Q2 - MI325XAMD2026-04-012026-06-30187051115031128Estimates generated on: 08-27-2026 16:14
FY26 Q2 - MI350XAMD2026-04-012026-06-30544422946578803Estimates generated on: 08-27-2026 16:14
FY26 Q2 - MI355XAMD2026-04-012026-06-30165489135506198701Estimates generated on: 08-27-2026 16:14
Q2 2026 Ascend 910C華為2026-04-012026-06-3013494444000245000Estimates generated on: 10-01-2026 17:15; H1 2026 update. See methodology: https://docs.google.com/document/d/1nPlAAB0HioI9LSk6u-LGEnVBvSAoc50MGXmMxOLv0fk/edit?usp=sharing
Q2 2026 - Trainium2Amazon2026-04-012026-06-30167379105022259274Estimates generated on: 10-14-2026 14:34
2026 Q2 - Siyuan 590Cambricon2026-04-012026-06-30123651110013800Estimates generated on: 10-01-2026 17:15; Cambricon methodology: https://docs.google.com/document/d/1LmVwqNAxL33wipBo7boVf5TRJusubmiLmACy2m0E2ew/edit?usp=sharing; units from reported quarterly revenue and gross margin over estimated cost per chip
Q2 2026 - TPU v8iGoogle2026-04-012026-06-3013092837480234658Estimates generated on: 09-30-2026 14:07
Q2 2026 Ascend 950PR華為2026-04-012026-06-30697325500085000Estimates generated on: 10-08-2026 10:46; H1 2026 update. See methodology: https://docs.google.com/document/d/1nPlAAB0HioI9LSk6u-LGEnVBvSAoc50MGXmMxOLv0fk/edit?usp=sharing
Q2 2026 - Trainium3Amazon2026-04-012026-06-3012082638155375197Estimates generated on: 10-14-2026 14:34
Q3 2026 - TPU v7Google2026-07-012026-09-30640846485526799602不完整。Incomplete: based on Broadcom fiscal quarters ending 08/02/2026. Estimates generated on: 09-30-2026 14:07
Q3 2026 - H100/H200NVIDIA2026-07-012026-09-30774264869109不完整。Incomplete: based on Nvidia fiscal quarters ending 7/26/2026. Estimates generated on: 08-27-2026 15:06
Q3 2026 - B200NVIDIA2026-07-012026-09-30134386114621157407不完整。Incomplete: based on Nvidia fiscal quarters ending 7/26/2026. Estimates generated on: 08-27-2026 15:06
Q3 2026 - B300NVIDIA2026-07-012026-09-3010330898752021213878不完整。Incomplete: based on Nvidia fiscal quarters ending 7/26/2026. Estimates generated on: 08-27-2026 15:06
Q3 2026 - TPU v8iGoogle2026-07-012026-09-307449321325133512不完整。Incomplete: based on Broadcom fiscal quarters ending 08/02/2026. Estimates generated on: 09-30-2026 14:07
來源、方法與版本紀錄

呈現2024Q1起，Nvidia、Google、Amazon、AMD、Huawei、Cambricon；Maia、MTIA、Dojo、消費GPU及較小廠商未完整涵蓋。資料集說明仍把Cambricon列為排除，與下載檔不一致，以實際列示並揭露差異。

下載檔涵蓋至2026Q3，原圖文字僅至2026Q2；中文圖採下載檔，不沿用原圖舊份額數字。

中文翻譯、編排與圖表重繪：Trend Core。依 CC BY 4.0 使用，未經 Epoch AI 審閱或背書。

原始檔涵蓋 2022-01-01 至 2026-09-30；圖表呈現期間列於圖下。

閱讀來源方法 · 下載來源資料

原始研究發表日期
未提供
來源資料更新日期
2026-10-14
資料取得時間
2026-10-19T02:09:59.744963+00:00
接受版本時間
2026-10-19T02:21:14.905539+00:00
中文內容更新
2026-10-19
來源 SHA-256
d73f0a772be916d3bd0fcb825351e97c773946293cfbaf494b28548c9ee728d3

晶片銷售資料集：來源更新 2026-10-14；原文發表 未提供。

完整方法：來源更新 未提供；原文發表 未提供。

資料與原始分析：Epoch AI ／ AI chip market share by designer · Epoch AI · CC BY 4.0 · Trend Core 重繪

資料期間 2024-01-01 至 2026-09-30；快照 20261019T021600187369+0000-176312511ce6；版本 10e433e46f916e467295d944df128372c796067cc8f4fcad3d144cc618f01ec6

本頁為公開資料的中文整理與視覺化，不構成投資建議。估計、預定與實際完成須分別核對。