NVIDIA（NVDA）— 2026-08-26 FY2027 Q2 法說會逐字稿（中譯）

日期：2026-08-26（美國時間 2026-08-26 盤後）｜時長：約 60 分鐘 逐字稿為 AlphaMemo 對 NVIDIA 官方 earnings call 音檔 AI 轉寫後中譯，專有名詞與數字以官方 press release／10-Q 為準。

原始內容

返回列表 NVDA· FY27Q2NVIDIA 2026-08-26 1:00:00 複製全文 下載 TXT 傳送到 Notion 與 Gemini 討論 NVDA_2027q2_20260826

這份檔案由 AlphaMemo.ai 製作，歡迎到網站了解更多：https://alphamemo.ai。

如果這份整理有為你帶來價值，歡迎支持我們。

警語：此 Memo 和逐字稿為 AI 生成，可能因為音檔品質導致轉錄逐字稿有誤，或是 AI 曲解原意／幻覺導致 Memo 內容有誤，請以原始音檔為準。

Memo 亮點 • NVIDIA 交出另一個卓越季度，創下營收、營業利益及每股盈餘新高。總營收達 960 億美元，較去年同期翻倍以上，且成長已連續第四季加速（推斷自音訊日期 20260826）。 [CFO - 00:02:13] • 第二季資料中心營收季增 18%，達 890 億美元，兩大子部門 Hyperscale 與 ACINE 均有強勁貢獻。Hyperscale 營收 490 億美元，較上一季成長 13%，主要受惠於 Blackwell 持續強勁表現。 [CFO - 00:02:57] • NVIDIA 預期 2028 財年營收將成長約 70%，屬供應受限的展望，儘管客戶預測明年成長將翻倍。 [CFO - 00:02:45] [CFO - 00:05:28] • 宣佈擴大與 AWS 的合作夥伴關係：AWS 將從本季起至 2029 財年第二季，部署額外 200 萬個 GPU，並將採用 NVIDIA 完整的實體 AI 堆疊於其倉儲機器人。 [CFO - 00:03:53] • NVIDIA 的次世代平台 Vera Rubin，相較於 Grace Blackwell Ultra，提供每瓦特 30 倍的吞吐量及 35 倍更低的 token 成本。生產出貨已於本月初開始，預期將成為 NVIDIA 史上最快的產品成長曲線。 [CFO - 00:08:07] • NVIDIA 於第二季向股東返還創紀錄的 260 億美元：其中 200 億美元透過股票回購，60 億美元以季度股息形式發放。年初至今，已將 60% 的自由現金流返還給股東。 [CFO - 00:23:22] • 第三季總營收預計為 1080 億美元，正負 2%。季增主要由資料中心 AI 驅動，Vera Rubin 預計將佔資料中心營收約 20%。 [CFO - 00:23:58] • 毛利率預計於第四季觸底，介於 71-72%，因記憶體價格極端波動，隨著價格調漲於 2028 財年生效，毛利率將回升至 72-73%。 [CFO - 00:25:13] 地緣政治 • 第二季，依據美國政府許可，向中國客戶出貨的 Hopper 200 產品佔資料中心總營收不到 1%。鑑於持續的地緣政治不確定性，NVIDIA 未將中國資料中心運算營收納入未來展望。 [CFO - 00:22:03] 產業循環週期 • AI 需求激增推動全球基礎建設擴建，並由日益多元的成長機會支撐。即使在 NVIDIA 規模下，需求加速仍持續，客戶預測明年成長將翻倍。 [CFO - 00:02:29] 產業供需 • NVIDIA 對 2028 財年 70% 營收成長的展望受限於供應，儘管客戶預測需求可支持 100% 成長。公司正努力縮小供需缺口，但預期供應限制至少持續至 2028 財年底。 [CFO - 00:05:28] [CEO - 00:30:58] • NVIDIA Compute 在所有服務的雲端中均已充分利用。其為 Hyperscale、NeoCloud 及 AI 實驗室合作夥伴創造的經濟價值持續提升。 [CFO - 00:05:40] • 整個供應鏈面臨挑戰且運作於滿載狀態。更多產能持續上線，但約 70% 的需求供應可用。公司致力與供應商密切合作以提升產能。 [CEO - 00:54:30] 產業狀況 • 雲端產業積壓訂單現已超過 2 兆美元，前五大 Hyperscaler 的資本支出預計於 2026 年達近 8000 億美元，2027 年達 1.3 兆美元。 [CFO - 00:03:37] 獲利 • GAAP 及非 GAAP 毛利率均為 75%，與上季相當，主因產品組合相似。毛利率預計於第四季觸底於 71-72%，並於 2028 財年隨價格調漲回升至 72-73%。 [CFO - 00:22:31] • Frontier Labs 銷售激增且毛利優異。其產生的 token 具盈利能力，唯一限制為可用運算資源量。 [CEO - 00:47:11] 公司展望、未來成長動能 • NVIDIA 預期 2028 財年營收將成長約 70%，受強勁 AI 需求及供應受限條件推動。公司正努力縮小供需缺口，但預期限制將持續至 2028 財年底。 [CFO - 00:02:45] • NVIDIA 的三大獨特能力——可運行所有模型的平台、完整堆疊 AI 工廠平台及 CUDA 生態系統——相互強化並推動成長。 [CFO - 00:17:07] • NVIDIA 預期 Vera Rubin 將達成公司史上最快的產品成長曲線，出貨已展開，主要合作夥伴包括 OCI、SpaceX AI 及 AWS。 [CFO - 00:08:27] 市場競爭策略 • NVIDIA 是全球唯一打造並提供完整 AI 工廠平台（全堆疊系統）的公司。客戶可自由組合元件，但多數缺乏自行組裝的技能或意願。 [CEO - 00:28:44] • 包括 OpenAI 與 Anthropic 在內的多家 AI 實驗室正設計自有定製晶片，但 NVIDIA 認為其平台具卓越經濟效益，預期這些公司將長期維持客戶及合作夥伴關係。 [CEO - 00:42:30] 公司對市場觀點的看法 • 投資者高度關注 NVIDIA 在推論市場的市佔率。公司認為其優勢非凡，因平台能運行所有模型並整合 AI 生命週期。 [CJ Muse - 00:31:53] [CEO - 00:32:18] 營收 • 總營收 960 億美元，較去年翻倍以上，第二季資料中心營收季增 18% 至 890 億美元。Hyperscale 營收 490 億美元，季增 13%；ACINE 營收 400 億美元，季增 25%，年增 138%。 [CFO - 00:02:19] • 第三季總營收預計為 1080 億美元，正負 2%。 [CFO - 00:23:58] 訂單 • NVIDIA 已收到所有主要 Hyperscaler、AI 雲端及系統 OEM 對 Vera Rubin 的採購訂單。 [CFO - 00:08:27] 產品組合 • NVIDIA 產品組合包括 Hopper、Blackwell、Vera Rubin（CPU、GPU、NVLink、InfiniBand 乙太網、Groq LPU）及 Grace CPU。公司已全面量產 Vera CPU，其 SPEC 基準測試完成速度比其他資料中心 CPU 快 1.8 倍，且每瓦特帶寬提升 5 倍。 [CFO - 00:07:25] 營收地區 • 第二季，依據美國政府許可，向中國客戶出貨的 Hopper 200 產品佔資料中心總營收不到 1%。未將中國資料中心運算營收納入未來展望。 [CFO - 00:22:03] 上游廠商 • NVIDIA 與三大主要記憶體供應商維持長期且深厚的合作關係，正密切合作以進一步提升其路線圖所需產能。 [CFO - 00:26:04] 下游客戶 • NVIDIA 客戶涵蓋 Hyperscaler、AI 實驗室、AI 原生企業及主權客戶。即使在 NVIDIA 規模下，需求加速仍持續，預測明年成長將翻倍。 [CFO - 00:02:38] 同業 • 包括 OpenAI 與 Anthropic 在內的多家 AI 實驗室正設計自有定製晶片，但 NVIDIA 認為其平台具卓越經濟效益，預期這些公司將長期維持客戶及合作夥伴關係。 [CEO - 00:42:30] 稼動率 • NVIDIA Compute 在所有服務的雲端中均已充分利用。NeoCloud 合作夥伴預計年底總裝機容量將達 8 吉瓦，較 2025 年底約 3 吉瓦大幅提升。 [CFO - 00:05:11] 存貨庫存 • 資產負債表上，因應 Vera Rubin 上市，庫存增加至 320 億美元。 [CFO - 00:23:00] 出貨週期與交期 • 應收帳款天數增加至 60 天，反映部分投資等級客戶大額採購延長付款條件，且出貨分多季進行。 [CFO - 00:23:09] 新產品／新技術 • Vera Rubin 展現 NVIDIA 每代帶來指數級性能提升的能力，相較 Grace Blackwell Ultra，每瓦特吞吐量提升 30 倍，token 成本降低 35 倍。生產出貨已於本月初開始。 [CFO - 00:08:07] • Groq 3 LPX 為 NVIDIA 首款機架級 LPU 系統，已全面量產並創紀錄，在 NVIDIA 人工分析基準測試中，token 每秒數接近次佳方案的四倍。 [CFO - 00:10:38] 產品價格 • NVIDIA 面臨記憶體價格極端波動，價格漲幅超出先前預期，且預計明年將持續上升。已執行的價格調漲預計於 2028 財年第一季生效。 [CFO - 00:24:55] 出貨量 • Vera Rubin 生產出貨已於本月初開始，預期將成為 NVIDIA 史上最快的產品成長曲線。出貨已展開，主要合作夥伴包括 OCI、SpaceX AI 及 AWS。 [CFO - 00:08:23] 法規影響 • 第二季，依據美國政府許可，向中國客戶出貨的 Hopper 200 產品佔資料中心總營收不到 1%。未將中國資料中心運算營收納入未來展望。 [CFO - 00:22:03] 成本毛利 • GAAP 及非 GAAP 毛利率均為 75%，與上季相當，主因產品組合相似。毛利率預計於第四季觸底於 71-72%，並於 2028 財年隨價格調漲回升至 72-73%。記憶體稀缺主要由 AI 建設推動。 [CFO - 00:22:31] 原物料／關鍵零組件 • NVIDIA 面臨記憶體價格極端波動，價格漲幅超出先前預期，且預計明年將持續上升。公司正與三大主要記憶體供應商密切合作以提升產能。 [CFO - 00:24:55] 銷售／管理／研發費用 • GAAP 及非 GAAP 營運費用分別季增 10% 及 11%，主要因運算基礎設施成本及薪酬福利支出增加。全年營運費用預計成長約 50%，受益於產品組合擴大及 AI 工具使用增加。 [CFO - 00:22:42] 擴廠規劃 • AWS 將從本季起至 2029 財年第二季，部署額外 200 萬個 GPU，並將採用 NVIDIA 完整的實體 AI 堆疊於其倉儲機器人。 [CFO - 00:03:53] 營業利益 • NVIDIA 交出另一個卓越季度，創下營業利益新高。 [CFO - 00:02:13] 稅率 • 非 GAAP 有效稅率升至 16%，較去年同期增加，主要因營收提升。2027 財年 GAAP 及非 GAAP 稅率預計介於 16% 至 18%，不含任何特殊項目或重大變動。 [CFO - 00:22:54] 股利政策 • 第二季，NVIDIA 向股東返還創紀錄的 260 億美元：其中 200 億美元透過股票回購，60 億美元以每股 0.25 美元季度股息發放。年初至今，已將 60% 的自由現金流返還給股東。未來公司計劃增加並返還扣除策略性用途後的超額自由現金流。 [CFO - 00:23:22] 現金、現金流 • NVIDIA 於第二季向股東返還創紀錄的 260 億美元，年初至今佔自由現金流 60%。公司計劃增加並返還扣除策略性用途後的超額自由現金流。 [CFO - 00:23:22] 應收帳款 • 應收帳款天數增加至 60 天，反映部分投資等級客戶大額採購延長付款條件，且出貨分多季進行。 [CFO - 00:23:09] Transcript Presentation Tiffany - Operator

(00:00:01)Good afternoon. My name is Tiffany, and I will be your conference operator today. At this time, I would like to welcome everyone to NVIDIA's second quarter earnings call. All lines have been placed on mute to prevent any background noise. After the speakers' remarks, there will be a question-and-answer session. If you would like to ask a question during this time, simply press star followed by the number one on your telephone keypad. If you would like to withdraw your question, press star one again. Thank you.

(00:00:01)下午好。我叫 Tiffany，今天將擔任您的會議操作員。此時，我想歡迎大家參加 NVIDIA 第二季度的財報電話會議。所有線路已經靜音，以防止任何背景噪音。在發言者發言後，將會有一個問答環節。如果您想在此期間提問，只需按下電話鍵盤上的星號，然後按數字一。如果您想撤回您的問題，再次按下星號一。謝謝。

Tiffany - Operator

(00:00:36)Toshiya Hari, you may begin your conference.

(00:00:36)Toshiya Hari，您可以開始您的會議。

Toshiya Hari - NVIDIA, Investor Relation

(00:00:40)Thank you. Good afternoon, and welcome to NVIDIA's conference call for the second quarter of fiscal 2027. With me today from NVIDIA are Jensen Huang, President and Chief Executive Officer, and Colette Kress, Executive Vice President and Chief Financial Officer. Our call is being webcast live on NVIDIA's Investor Relations website. The webcast will be available for replay until the conference call to discuss our financial results for the third quarter of fiscal 2027. The content of today's call is NVIDIA's property.

(00:00:40)謝謝。下午好，歡迎參加 NVIDIA 2027 財年的第二季度電話會議。今天與我一起參加會議的有 NVIDIA 的總裁兼首席執行官 Jensen Huang 和執行副總裁兼首席財務官 Colette Kress。我們的會議正在 NVIDIA 投資者關係網站上進行直播。會議的重播將可用，直到我們討論 2027 財年第三季度財務結果的電話會議。今天會議的內容是 NVIDIA 的財產。

Toshiya Hari - NVIDIA, Investor Relation

(00:01:11)It cannot be reproduced or transcribed without our prior written consent. During this call, we may make forward-looking statements based on current expectations. These are subject to a number of significant risks and uncertainties, and our actual results may differ materially. For a discussion of factors that could affect our future financial results and business, please refer to the disclosures in today's earnings release, our most recent Forms 10-K and 10-Q, and the reports we may file on Form 8-K with the Securities and Exchange Commission.

(00:01:11)未經我們事先書面同意，不得複製或轉錄。在此次會議中，我們可能會根據當前預期發表前瞻性聲明。這些聲明受到多種重大風險和不確定性的影響，我們的實際結果可能會有重大差異。有關可能影響我們未來財務結果和業務的因素的討論，請參閱今天的財報公告、我們最近的 10-K 和 10-Q 表格，以及我們可能向證券交易委員會提交的 8-K 表格報告。

Toshiya Hari - NVIDIA, Investor Relation

(00:01:43)All our statements are made as of today, August 26, 2026, based on information currently available to us. Except as required by law, we assume no obligation to update any such statements. During this call, we will discuss non-GAAP financial measures. You can find a reconciliation of these non-GAAP financial measures to GAAP financial measures in our CFO commentary, which is posted on our website. With that, let me turn the call over to Colette.

(00:01:43)我們的所有聲明均以今天，即 2026 年 8 月 26 日為準，基於目前可用的信息。除非法律要求，否則我們不承擔更新任何此類聲明的義務。在此次會議中，我們將討論非 GAAP 財務指標。您可以在我們的首席財務官評論中找到這些非 GAAP 財務指標與 GAAP 財務指標的對比，該評論已發布在我們的網站上。有了這些，我將會議轉交給 Colette。

Colette Kress - NVIDIA, CFO

(00:02:11)Thanks, Toshiya Hari. We delivered another outstanding quarter with record revenue, operating income, and EPS. Total revenue of $96 billion more than doubled year over year, as growth accelerated for the fourth consecutive quarter. The surge in AI demand is driving a global infrastructure build-out, supported by an expanding and diverse set of growth opportunities. These opportunities span hyperscalers, AI labs, AI natives, enterprises, and sovereign customers. We expect to grow revenue by approximately 70% in fiscal 2028.

(00:02:11)謝謝，Toshiya Hari。我們又交出了一個卓越的季度，創下了收入、營業收入和每股收益的紀錄。總收入達到 960 億美元，同比增長超過一倍，增長速度在連續第四個季度加快。人工智慧需求的激增正在推動全球基礎設施的擴建，並得到一系列擴展和多樣化的增長機會的支持。這些機會涵蓋了超大規模數據中心、人工智慧實驗室、人工智慧原生企業、企業和主權客戶。我們預計在 2028 財年收入將增長約 70%。

Colette Kress - NVIDIA, CFO

(00:02:51)This is a supply-constrained outlook. Q2 data center revenue increased 18% quarter over quarter to $89 billion, with strong contributions from both sub-segments: Hyperscale and ACINE, which includes our NeoCloud, industrial, and enterprise customers. Hyperscale revenue of $49 billion grew 13% sequentially, driven by sustained strength in Blackwell. Reinforcing that more compute drives more revenue, as new GPU capacity comes online, our hyperscale customers delivered strong financial results this quarter with accelerating revenue growth and expanding margins.

(00:02:51)這是一個供應受限的展望。第二季度數據中心收入環比增長 18%，達到 890 億美元，兩個子細分市場均有強勁貢獻：超大規模和 ACINE，包括我們的 NeoCloud、工業和企業客戶。超大規模收入達到 490 億美元，環比增長 13%，這得益於 Blackwell 的持續強勁表現。強調了更多計算能力帶來更多收入，隨著新的 GPU 能力上線，我們的超大規模客戶在本季度交出了強勁的財務結果，收入增長加速，利潤率擴大。

Colette Kress - NVIDIA, CFO

(00:03:37)With the cloud industry backlog now greater than $2 trillion, CapEx by the top five hyperscalers is expected to reach nearly $800 billion in 2026 and $1.3 trillion in 2027. Today, we are delighted to announce an expansion of our partnership with AWS. Building on its already vast installed base of NVIDIA compute, AWS is deploying an additional 2 million GPUs starting this quarter through the second quarter of fiscal 2029. Along with Vera CPUs, some integrated with Rubin, others standalone. AWS will serve the NVIDIA Nemotron family of open models on Amazon Bedrock and SageMaker.

(00:03:37)隨著雲行業的積壓訂單現在超過 2 萬億美元，前五大超大規模數據中心的資本支出預計在 2026 年將達到近 8000 億美元，並在 2027 年達到 1.3 萬億美元。今天，我們很高興宣佈與 AWS 的合作夥伴關係擴展。在其已經龐大的 NVIDIA 計算基礎上，AWS 將從本季度開始再部署 200 萬個 GPU，直到 2029 財年的第二季度。連同 Vera CPU，有些與 Rubin 集成，其他則獨立運行。AWS 將在 Amazon Bedrock 和 SageMaker 上提供 NVIDIA Nemotron 系列開放模型。

Colette Kress - NVIDIA, CFO

(00:04:24)Amazon will also adopt our full physical AI stack—Omniverse, Cosmos, Isaac, and Jetson—to power its fleet of warehouse robots. ACINE revenue of $40 billion increased 25% sequentially and 138% year over year. Growth was driven by NeoCloud capacity additions to meet rising demand from enterprises, AI startups, sovereigns, as well as hyperscalers purchasing capacity to supplement their own build-outs. Using NVIDIA DGX reference designs, our NeoCloud partners are bringing capacity online faster and at lower token cost.

(00:04:24)亞馬遜還將採用我們完整的物理 AI 堆疊——Omniverse、Cosmos、Isaac 和 Jetson——為其倉庫機器人提供動力。ACINE 收入達到 400 億美元，環比增長 25%，同比增長 138%。增長是由於 NeoCloud 的產能增加，以滿足企業、人工智慧初創公司、主權客戶以及超大規模數據中心購買產能以補充其自身擴建的需求。使用 NVIDIA DGX 參考設計，我們的 NeoCloud 合作夥伴正在更快地上線產能，並以更低的代幣成本運行。

Colette Kress - NVIDIA, CFO

(00:05:11)They are expected to exit the year with eight gigawatts in total installed capacity, up from approximately three gigawatts at the end of 2025. Incredibly, we are seeing demand acceleration even at our scale. Customers' forecasts point to our growth doubling next year. However, as I mentioned earlier, we expect to grow approximately 70% as we remain supply constrained. NVIDIA Compute is fully utilized across every cloud we serve. The economic value it generates for our hyperscale, NeoCloud, and AI lab partners keeps rising.

(00:05:11)他們預計在年底前將總裝機容量提升至 8 吉瓦，較 2025 年底的約 3 吉瓦有所增加。令人難以置信的是，我們甚至在這樣的規模下也看到了需求加速。客戶的預測顯示我們的增長明年將翻倍。然而，正如我之前提到的，我們預計將增長約 70%，因為我們仍然受到供應限制。NVIDIA 計算在我們服務的每個雲端中都得到了充分利用。它為我們的超大規模、NeoCloud 和 AI 實驗室合作夥伴所創造的經濟價值不斷上升。

Colette Kress - NVIDIA, CFO

(00:05:54)Besides building the best AI computing technologies and the most capable supply chain, NVIDIA has three unique capabilities that are engines powering our growth. First, NVIDIA's architecture supports every model, and we are increasing our market share as both closed and open model adoption accelerates. Adoption of both closed and open models is skyrocketing. NVIDIA powers leading closed models such as OpenAI, Anthropic, Groq, Meta, and Gemini, as well as top open models including TML, Mistral, Qwen, and Kimi, GLM, DeepSeek, Minimax, and Nemotron.

(00:05:54)除了建立最佳的 AI 計算技術和最強大的供應鏈外，NVIDIA 還擁有三項獨特的能力，這些能力是推動我們增長的引擎。首先，NVIDIA 的架構支持每一種模型，隨著封閉和開放模型的採用加速，我們的市場份額正在增加。封閉和開放模型的採用正在迅速上升。NVIDIA 驅動著領先的封閉模型，如 OpenAI、Anthropic、Groq、Meta 和 Gemini，以及頂尖的開放模型，包括 TML、Mistral、Qwen 和 Kimi，GLM、DeepSeek、Minimax 和 Nemotron。

Colette Kress - NVIDIA, CFO

(00:06:38)We excel with both small and large models—whether video, autoregressive, or diffusion—deployed in the cloud or at the edge. NVIDIA delivers outstanding performance in training, inference, and agentic workloads. Our platform is fungible across every model and workload, and durable throughout the entire AI lifecycle. This combination of performance, fungibility, and durability makes NVIDIA the most productive and financeable compute infrastructure. Our second unique strength is our full-stack AI factory platform, which is expanding our share of the data center total addressable market (TAM).

(00:06:38)我們在小型和大型模型方面都表現出色——無論是視頻、自回歸還是擴散——無論是在雲端還是邊緣部署。NVIDIA 在訓練、推理和代理工作負載方面提供卓越的性能。我們的平台在每一種模型和工作負載中都是可互換的，並且在整個 AI 生命週期中都是耐用的。性能、可互換性和耐用性的結合使 NVIDIA 成為最具生產力和可融資的計算基礎設施。我們的第二個獨特優勢是我們的全棧 AI 工廠平台，這正在擴大我們在數據中心總可尋址市場 （TAM） 的份額。

Colette Kress - NVIDIA, CFO

(00:07:25)Since Hopper, our revenue opportunity has grown from approximately $18 billion per gigawatt to $25 billion with Blackwell, and now $40 billion with Vera Rubin, which includes Vera CPU, Rubin GPU, NVLink, InfiniBand Ethernet, and the Groq LPU announced earlier this week. Our ability to co-design GPU, CPU, NVLink-scale-up networking, scale-out networking systems, algorithms, and software enables us to deliver exponential performance gains every generation. Vera Rubin exemplifies this, delivering 30 times higher throughput per megawatt and 35 times lower token costs compared to Grace Blackwell Ultra.

(00:07:25)自 Hopper 以來，我們的收入機會從每千瓦約 180 億美元增長到與 Blackwell 的 250 億美元，現在與 Vera Rubin 的 400 億美元，包括 Vera CPU、Rubin GPU、NVLink、InfiniBand 以太網，以及本週早些時候宣佈的 Groq LPU。我們共同設計 GPU、CPU、NVLink 擴展網絡、擴展網絡系統、算法和軟件的能力使我們能夠每一代都提供指數級的性能增長。Vera Rubin 便是這一點的典範，提供每兆瓦 30 倍的吞吐量，並且與 Grace Blackwell Ultra 相比，令代幣成本降低 35 倍。

Colette Kress - NVIDIA, CFO

(00:08:23)We began production shipments of Vera Rubin earlier this month. Having already received purchase orders from every major hyperscaler, AI cloud, and system OEM, we expect Vera Rubin to achieve the fastest product ramp in NVIDIA's history. Our networking business had another record quarter, with revenue growing 18%. This growth was sequential. Spectrum-X Ethernet, which grew 2.6 times year-over-year, is helping us become the largest and fastest-growing networking company globally. Increasing adoption of agentic AI is accelerating demand for data center CPUs.

(00:08:23)我們本月早些時候開始了 Vera Rubin 的生產出貨。我們已經從每個主要的超大規模雲服務商、AI 雲和系統 OEM 收到訂單，我們預計 Vera Rubin 將實現 NVIDIA 歷史上最快的產品增長。我們的網絡業務又創下了記錄季度，收入增長 18%。這一增長是連續的。Spectrum-X 以太網年增長 2.6 倍，幫助我們成為全球最大和增長最快的網絡公司。代理 AI 的採用增加正在加速對數據中心 CPU 的需求。

Colette Kress - NVIDIA, CFO

(00:09:10)Our Grace CPU, introduced in 2021, has been highly successful, with trailing 12-month revenue exceeding $5 billion. We are now in full production of our next-generation Vera CPU. As a standalone product, Vera further expands our TAM. Vera completes the SPEC benchmark 1.8 times faster and delivers five times the bandwidth per watt compared to any other data center CPU. We expect Vera to be deployed by every major hyperscaler, neocloud, AI lab, and system OEM, with shipments already underway to lead partners including OCI, SpaceX AI, and starting this quarter, AWS.

(00:09:10)我們的 Grace CPU，自 2021 年推出以來，表現非常成功，過去 12 個月的收入超過 50 億美元。我們現在已經全面生產下一代 Vera CPU。作為一個獨立產品，Vera 進一步擴大了我們的 TAM。Vera 完成 SPEC 基準測試的速度比任何其他數據中心 CPU 快 1.8 倍，並且每瓦特的帶寬是其他產品的五倍。我們預計 Vera 將被每個主要的超大規模雲服務商、新雲、AI 實驗室和系統 OEM 部署，並且已經開始向包括 OCI、SpaceX AI 和 AWS 在內的主要合作夥伴發貨。

Colette Kress - NVIDIA, CFO

(00:10:04)We continue to see demand for approximately $20 billion in total server CPUs. Based on customer demand and improving supply outlook, our preliminary expectation is for CPU revenue to more than double in fiscal 2028, positioning us as one of the world's leading server CPU suppliers. Since announcing our Groq partnership last year, we have been integrating NVIDIA's high-throughput architecture with Groq's high-interactivity design. At Hot Chips earlier this week, we announced that Groq 3 LPX, our first rack-scale LPU system, is in full production and already setting records.

(00:10:04)我們繼續看到約 200 億美元的伺服器 CPU 需求。根據客戶需求和供應前景的改善，我們的初步預期是 CPU 收入在 2028 財年將增長超過兩倍，使我們成為全球領先的伺服器 CPU 供應商之一。自去年宣佈與 Groq 的合作以來，我們一直在將 NVIDIA 的高吞吐量架構與 Groq 的高互動設計進行整合。在本週早些時候的 Hot Chips 大會上，我們宣佈 Groq 3 LPX，我們的第一個機架級 LPU 系統，已經全面生產並創下了記錄。

Colette Kress - NVIDIA, CFO

(00:10:52)It demonstrates nearly four times the tokens per second compared to the next best alternative on our artificial analysis benchmark. We expect to ship Groq 3 LPX in volume later this quarter to early adopters. Nebius will be the first recipient. Today, we are not just selling the best chips. We offer a full-stack AI factory platform that delivers superior economics for customers and captures a larger share of the data center TAM. Our third unique capability is the combination of our full-stack AI factory and rich CUDA ecosystem, enabling us to extend AI into markets unreachable by a single chip alone.

(00:10:52)與我們的人工分析基準相比，它每秒的代幣數量幾乎是下一個最佳替代方案的四倍。我們預計在本季度晚些時候向早期採用者發貨 Groq 3 LPX。Nebius 將是第一個接受者。今天，我們不僅僅是在銷售最好的晶片。我們提供一個全棧 AI 工廠平台，為客戶提供優越的經濟效益，並捕獲更大份額的數據中心 TAM。我們的第三個獨特能力是我們的全棧 AI 工廠和豐富的 CUDA 生態系統的結合，使我們能夠將 AI 擴展到單一晶片無法觸及的市場。

Colette Kress - NVIDIA, CFO

(00:11:42)Beyond hyperscalers lies a vast market eager to adopt AI, consisting of customers uninterested in designing their own custom silicon. NVIDIA's fully proven full-stack platform is uniquely suited to help sovereigns, neoclouds, and enterprises build AI infrastructure, bring it to full operation, continuously optimize it through CUDA software, and connect it to demand from our extensive developer ecosystem. While hyperscalers remain a major growth driver, non-hyperscaler growth—our AIC&E segment spanning sovereign regional neoclouds, enterprise edge, and air-gapped data centers—will represent roughly half of our data center business.

(00:11:42)超大規模雲服務商之外是一個渴望採用 AI 的廣闊市場，這些客戶對設計自己的定製矽片不感興趣。NVIDIA 完全驗證的全棧平台獨特地適合幫助主權國家、新雲和企業建立 AI 基礎設施，將其全面運行，通過 CUDA 軟件持續優化，並將其與我們廣泛的開發者生態系統的需求相連接。雖然超大規模雲服務商仍然是主要的增長驅動力，但非超大規模雲的增長——我們的 AIC&E 部門涵蓋主權區域新雲、企業邊緣和隔離數據中心——將佔我們數據中心業務的大約一半。

Colette Kress - NVIDIA, CFO

(00:12:32)Our AI-native startup ecosystem, developed and running primarily on the NVIDIA compute platform, is scaling rapidly. Global venture capital funding in AI, about 70% of which is spent on compute, exceeded $400 billion in the first half of 2026, surpassing the $265 billion raised in all of 2025. Nearly 20 companies, including Cursor (owned by SpaceX), Figma, and Together AI, now exceed $1 billion in annualized run-rate revenue, up from 13 companies in Q4 last year, with vertical enterprise software showing the fastest growth.

(00:12:32)我們的 AI 原生初創生態系統，主要在 NVIDIA 計算平台上開發和運行，正在迅速擴展。全球對 AI 的風險投資資金，約 70% 用於計算，在 2026 年上半年超過 4000 億美元，超過了 2025 年全年籌集的 2650 億美元。包括 Cursor（由 SpaceX 擁有）、Figma 和 Together AI 在內的近 20 家公司現在的年化運行收入超過 10 億美元，較去年第四季度的 13 家公司有所增加，垂直企業軟件顯示出最快的增長。

Colette Kress - NVIDIA, CFO

(00:13:20)In enterprise, on a trailing 12-month basis, on-premises revenue in the automotive vertical reached $8 billion, while financial services, manufacturing, and healthcare combined contributed $7 billion. Hudson River Trading and Jane Street are leveraging NVIDIA-powered AI factories to accelerate quantitative trading. Samsung Electronics is using NVIDIA cuLitho to achieve up to 20 times greater performance in computational lithography, while Bristol-Myers Squibb is investing in the Vera Rubin AI Factory, following Roche and Lilly build-outs, as drug R&D timelines compress from years to months.

(00:13:20)在企業方面，根據過去 12 個月的數據，汽車垂直領域的本地收入達到 80 億美元，而金融服務、製造業和醫療保健的總和貢獻了 70 億美元。Hudson River Trading 和 Jane Street 正在利用 NVIDIA 驅動的 AI 工廠加速量化交易。三星電子正在使用 NVIDIA cuLitho 在計算光刻中實現高達 20 倍的性能提升，而百時美施貴寶則在 Vera Rubin AI 工廠中進行投資，隨著羅氏和禮來的建設，藥物研發的時間表從數年縮短到數月。

Colette Kress - NVIDIA, CFO

(00:14:06)In Sovereign AI, our business primarily through regional neoclouds grew 35% sequentially and more than tripled year-over-year in Q2. A country or region can allocate land and power directly to a regional cloud partner in ways it would not to a foreign hyperscaler. We do not own a cloud ourselves. We are a neutral partner to every sovereign and neocloud. Because NVIDIA Compute is productive, fungible, rentable, and durable, regional cloud interest is surging worldwide. We helped CoreWeave, Nebius, and Nscale build entire infrastructure businesses, and neoclouds are emerging everywhere.

(00:14:06)在主權 AI 領域，我們通過區域新雲的業務在第二季度增長了 35%，同比增長超過三倍。一個國家或地區可以以不會對外國超大規模雲服務商的方式，直接將土地和電力分配給區域雲合作夥伴。我們自己並不擁有雲。我們是每個主權國家和新雲的中立合作夥伴。由於 NVIDIA 計算是高效的、可互換的、可租賃的和耐用的，區域雲的興趣在全球範圍內激增。我們幫助 CoreWeave、Nebius 和 Nscale 建立整個基礎設施業務，並且新雲正在各地出現。

Colette Kress - NVIDIA, CFO

(00:14:52)Firebird in Armenia, Cassava Technologies across Africa, GMI Cloud in Taiwan, Yoda and NASA in India, Hermes in Australia, YTL-AI Cloud in Malaysia, pairing local land, power, and operating expertise with our platform. Last month, we announced a partnership with NOATRA, Japan's national AI company, to build an NVIDIA DGX AI factory that will create open models powering AI agents, digital twins, robotics, and physical AI applications. South Korea's LG and Hyundai Motor Group are partnering with NVIDIA to develop and scale AI technologies.

(00:14:52)亞美尼亞的 Firebird、非洲的 Cassava Technologies，台灣的 GMI Cloud、印度的 Yoda 和 NASA、澳大利亞的 Hermes、馬來西亞的 YTL-AI Cloud，將當地的土地、電力和運營專業知識與我們的平台相結合。上個月，我們宣佈與日本國家人工智慧公司 NOATRA 建立合作夥伴關係，將建造一座 NVIDIA DGX 人工智慧工廠，該工廠將創建開放模型，驅動人工智慧代理、數位雙胞胎、機器人和實體人工智慧應用。韓國的 LG 和現代汽車集團正在與 NVIDIA 合作開發和擴展人工智慧技術。

Colette Kress - NVIDIA, CFO

(00:15:44)In Europe, a record 35 new NVIDIA-powered AI supercomputers were unveiled to advance industry and scientific breakthroughs. Neoclouds are experiencing strong demand pipelines from a diverse range of offtakers. Instead of allocating their entire capacity to a single long-term offtake guarantee, which lenders typically require to finance a data center independently, we have introduced a revenue-sharing structure. NVIDIA provides a take-or-pay commitment on a portion of the facility's capacity, offering a minimum revenue guarantee that gives lenders confidence to underwrite the project; in exchange, we share a portion of the Neocloud's revenue earned above that floor.

(00:15:44)在歐洲，推出了創紀錄的 35 台 NVIDIA 驅動的人工智慧超級計算機，以推進行業和科學突破。Neoclouds 正面臨來自多樣化用戶的強勁需求管道。我們引入了一種收益分享結構，而不是將其全部產能分配給單一的長期購買保證，這是貸款方通常要求的，以獨立融資數據中心。NVIDIA 對設施產能的一部分提供了取或付承諾，提供最低收入保證，讓貸款方有信心承保該項目；作為交換，我們分享 Neocloud 超過該底線的收入的一部分。

Colette Kress - NVIDIA, CFO

(00:16:31)Independent capital still underwrites every deal on its own merits. We are not making loans. In this model, we get paid twice: once on the hardware sale and again through a share of rental revenue, a highly recurring stream layered on top of a one-time equipment purchase. Over time, this model can expand our addressable market and create recurring usage-linked revenue streams alongside our core platform revenue, with the potential to drive billions in revenue over the medium to long term.

(00:16:31)獨立資本仍然根據每筆交易的自身優勢進行承保。我們並不提供貸款。在這種模式下，我們獲得兩次收入：一次是硬體銷售，另一次是通過租金收入的分享，這是一種高度重複的收入流，層疊在一次性設備購買之上。隨著時間的推移，這種模式可以擴大我們的可服務市場，並創造與我們核心平台收入並行的重複使用相關的收入流，具有在中長期內推動數十億收入的潛力。

Colette Kress - NVIDIA, CFO

(00:17:07)Together, NVIDIA's three unique capabilities—a platform that runs every model, a full-stack AI factory platform capturing more of the data center TAM, and a CUDA ecosystem that extends AI into markets no single chip could reach alone—reinforce one another and drive our growth. Let me update you on our progress with our frontier AI labs. The Frontier AI labs have extraordinary demand for training and inference compute, but they are growing faster than their balance sheets and credit profiles can support.

(00:17:07)NVIDIA 的三項獨特能力——一個運行每個模型的平台、一個全堆疊的人工智慧工廠平台，捕捉更多的數據中心市場總量，以及一個將人工智慧擴展到單一晶片無法獨立達到的市場的 CUDA 生態系統——相互強化，推動我們的增長。讓我向您更新我們在前沿人工智慧實驗室的進展。前沿人工智慧實驗室對訓練和推理計算的需求異常強勁，但它們的增長速度超過了其資產負債表和信用狀況所能支持的範圍。

Colette Kress - NVIDIA, CFO

(00:17:46)They have rapidly growing customer demand but still lack the decades-long infrastructure contracts and investment-grade financing capacity needed to secure AI factory infrastructure independently. In other words, their growth isn't limited by technology or customer demand. It is limited by compute capacity. For these companies, more compute means more intelligence, more users, and more revenue. NVIDIA is essential to powering this flywheel. First, we've invested nearly $50 billion in the Frontier AI Labs.

(00:17:46)他們的客戶需求迅速增長，但仍然缺乏獨立確保人工智慧工廠基礎設施所需的數十年基礎設施合同和投資級融資能力。換句話說，他們的增長並不受技術或客戶需求的限制。它受到計算能力的限制。對於這些公司來說，更多的計算意味著更多的智慧、更多的用戶和更多的收入。NVIDIA 對於驅動這個飛輪至關重要。首先，我們在前沿人工智慧實驗室投資了近 500 億美元。

Colette Kress - NVIDIA, CFO

(00:18:26)This was a meaningful commitment but represented a small fraction of our expected free cash flow over the same period. Furthermore, to support the Frontier Labs infrastructure build-outs, we recently announced partnerships with six of the world's leading infrastructure capital providers—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to establish financing platforms that will raise over $500 billion of third-party capital. With these partnerships, building on our unique, fungible, and durable computing platform, the AI labs will be able to build and access AI infrastructure funded by long-term institutional capital at relatively attractive rates.

(00:18:26)這是一項重要的承諾，但僅佔我們在同一時期預期自由現金流的一小部分。此外，為了支持前沿實驗室的基礎設施建設，我們最近宣佈與六家全球領先的基礎設施資本提供商——阿波羅、貝萊德、黑石、布魯克菲爾德、高盛和 KKR——建立合作夥伴關係，以建立融資平台，將籌集超過 5000 億美元的第三方資本。通過這些合作夥伴關係，基於我們獨特、可替代且耐用的計算平台，人工智慧實驗室將能夠建立和訪問由長期機構資本資助的人工智慧基礎設施，並以相對有吸引力的利率進行融資。

Colette Kress - NVIDIA, CFO

(00:19:13)Last week, we announced that we secured land, power, and shell capacity through our partnership with SoftBank Energy to exclusively host NVIDIA Compute at their PORTS-Pike campus. The initial deployment, expected to support 4.25 gigawatts of AI factory capacity, will be utilized by OpenAI. Each generation of NVIDIA AI factory systems deployed at PORTS-Pike could represent approximately 1.5 million NVIDIA GPUs. Over more than 20 years, the site could support multiple upgrade cycles. Here's the essential economic point.

(00:19:13)上週，我們宣佈通過與軟銀能源的合作夥伴關係，獲得土地、電力和外殼產能，專門在他們的 PORTS-Pike 校園內託管 NVIDIA 計算。預計支持 4.25 吉瓦人工智慧工廠產能的初始部署將由 OpenAI 使用。每一代在 PORTS-Pike 部署的 NVIDIA 人工智慧工廠系統可能代表約 150 萬個 NVIDIA GPU。在 20 多年內，該地點可以支持多次升級週期。這裡是關鍵的經濟觀點。

Colette Kress - NVIDIA, CFO

(00:19:53)The LPS commitment secures a long-lived AI factory site, while the NVIDIA compute within the data center can be upgraded repeatedly. This project deepens our longstanding partnership with OpenAI. OpenAI has committed to substantial deployments of NVIDIA AI infrastructure through 2030. OpenAI's existing and planned commitments represent approximately 12 gigawatts of NVIDIA compute. For another frontier AI lab, we will provide selective credit enhancement for nearly two gigawatts of compute. This complements the substantial NVIDIA compute capacity they've secured independently without NVIDIA's credit support.

(00:19:53)LPS 承諾確保了一個長期存在的人工智慧工廠地點，而數據中心內的 NVIDIA 計算可以不斷升級。這個項目加深了我們與 OpenAI 的長期合作夥伴關係。OpenAI 已承諾在 2030 年前大規模部署 NVIDIA 人工智慧基礎設施。OpenAI 現有和計劃中的承諾代表約 12 吉瓦的 NVIDIA 計算。對於另一個前沿人工智慧實驗室，我們將提供近 2 吉瓦計算的選擇性信用增強。這補充了他們在沒有 NVIDIA 信用支持的情況下獨立確保的可觀 NVIDIA 計算能力。

Colette Kress - NVIDIA, CFO

(00:20:43)We recognize the scale of this support, and we know some will call this circular financing. We see it differently. We're undergoing a major computing platform shift. The creation of one of the most important technologies in human history. These are once-in-a-generation companies. Their technology leadership is proven, and their customer traction and usage are skyrocketing. We expect them to become the largest technology companies in history. We believe these investments, measured against the strength of their demand, the business they create for us, the ecosystem they build on NVIDIA's platform, and the equity returns on our invested capital, will be excellent.

(00:20:43)我們認識到這種支持的規模，我們知道有些人會稱這為循環融資。我們的看法有所不同。我們正在經歷一場重大的計算平台轉變。創造人類歷史上最重要的技術之一。這些是千載難逢的公司。他們的技術領導地位已得到證明，客戶的吸引力和使用率正在飆升。我們預計他們將成為歷史上最大的科技公司。我們相信這些投資，與他們的需求強度、為我們創造的業務、他們在 NVIDIA 平台上建立的生態系統以及我們投資資本的股權回報相比，將是優秀的。

Colette Kress - NVIDIA, CFO

(00:21:31)And our risk is limited. The NVIDIA Compute platform is fungible and durable and can be redeployed to support other customers. For context, we expect demand from the AI labs, for which we expect to leverage our balance sheet, to contribute roughly a quarter of our business next year. This remains compute we ship will be consumed by investment-grade customers or those backed by one. In Q2, we shipped less than 1% of our total data center revenue in Hopper 200 products to customers based in China, in accordance with U.S.

(00:21:31)而且我們的風險是有限的。NVIDIA 計算平台是可替代且耐用的，可以重新部署以支持其他客戶。為了提供背景，我們預計來自人工智慧實驗室的需求，將使我們預期利用資產負債表，明年將貢獻大約四分之一的業務。我們運送的計算將被投資級客戶或那些由其支持的客戶消耗。在第二季度，我們向中國客戶運送的 Hopper 200 產品佔我們總數據中心收入的不到 1%，這是根據美國政府的許可。

Colette Kress - NVIDIA, CFO

(00:22:14)government licenses. Current Hopper shipments are dilutive to corporate gross margins. Given ongoing geopolitical uncertainty, there is no China data center compute revenue in our forward outlook. Moving to the rest of the P&L, GAAP and non-GAAP gross margins were both 75%, largely unchanged from last quarter due to a similar product mix. GAAP and non-GAAP operating expenses increased by 10% and 11% sequentially, primarily due to higher compute infrastructure costs and increased compensation and benefits expenses.

(00:22:14)政府許可。當前的 Hopper 出貨對公司毛利率造成稀釋。鑒於持續的地緣政治不確定性，我們的前瞻性展望中沒有中國數據中心計算收入。轉向損益表的其他部分，GAAP 和非 GAAP 毛利率均為 75%，與上個季度基本持平，因為產品組合相似。GAAP 和非 GAAP 的營運支出分別增加了 10% 和 11%，主要是由於計算基礎設施成本上升以及薪酬和福利支出增加。

Colette Kress - NVIDIA, CFO

(00:22:54)Our non-GAAP effective tax rate rose to 16% compared to a year ago, mainly driven by higher revenue. On our balance sheet, inventory grew to $32 billion as we prepared for the Vera Rubin launch. Days sales outstanding increased to 60 days, reflecting extended payment terms for large purchases by certain investment-grade customers, with shipments scheduled over multiple quarters. In Q2, we returned a record $26 billion to shareholders: $20 billion through share repurchases and $6 billion via our quarterly dividend of $0.25 per share.

(00:22:54)我們的非 GAAP 實際稅率較去年上升至 16%，主要是由於收入增加。在我們的資產負債表上，庫存增長至 320 億美元，因為我們為 Vera Rubin 的推出做準備。應收帳款天數增加至 60 天，反映出某些投資級客戶的大宗採購延長了付款條件，運輸安排跨越多個季度。在第二季度，我們向股東回報了創紀錄的 260 億美元：200 億美元通過股票回購，60 億美元通過每股 0.25 美元的季度股息。

Colette Kress - NVIDIA, CFO

(00:23:35)Relative to our plan to return 50% or more of free cash flow, we have returned 60% on a year-to-date basis. Moving forward, we intend to increase and return excess free cash flow, net of strategic uses. Let me now turn to the outlook for the third quarter. Total revenue is expected to be $108 billion, plus or minus 2%. We expect sequential growth to be driven primarily by AI in the data center, while growth in hyperscale is expected to reaccelerate in Q4 and into fiscal year 28, as Vera Rubin supply increases over time.

(00:23:35)相對於我們計劃回報 50% 或更多的自由現金流，我們截至目前已回報 60%。展望未來，我們打算增加並回報超額自由現金流，扣除戰略用途。現在讓我轉向第三季度的展望。總收入預計為 1080 億美元，誤差範圍為正負 2%。我們預計，數據中心的 AI 將主要推動環比增長，而超大規模的增長預計將在第四季度和 28 財年重新加速，隨著 Vera Rubin 的供應隨時間增加。

Colette Kress - NVIDIA, CFO

(00:24:21)We anticipate Vera Rubin will account for about 20% of data center revenue in Q3. Looking ahead, our preliminary expectation is for fiscal year 28 revenue to grow approximately 70% year over year. Although we will work to close the supply-demand gap, we expect supply constraints to remain a bottleneck at least through the end of fiscal year 28. Many of you have expressed concerns regarding our gross margins as component costs have risen significantly. As you know, we are experiencing extreme pricing conditions in memory.

(00:24:21)我們預計 Vera Rubin 將在第三季度佔數據中心收入的約 20%。展望未來，我們的初步預期是 28 財年的收入將同比增長約 70%。雖然我們將努力縮小供需差距，但我們預計供應限制至少會在 28 財年結束之前成為瓶頸。許多人對我們的毛利率表示擔憂，因為元件成本大幅上升。如您所知，我們在記憶體方面正面臨極端的定價條件。

Colette Kress - NVIDIA, CFO

(00:25:01)The magnitude of the price increase has exceeded our prior expectations and is expected to rise further into next year. As a result, we are resetting expectations today. For Q3, we expect GAAP and non-GAAP gross margins to be 74%, plus or minus 50 basis points. We expect margins to bottom in Q4, ranging between 71% and 72%. Margins are then expected to settle at 72% to 73% in fiscal year 28 as executed price increases take effect in Q1. We want to be direct about this rather than leave it as an open question.

(00:25:01)價格上漲的幅度超出了我們之前的預期，並預計明年將進一步上升。因此，我們今天將重新設定預期。對於第三季度，我們預計 GAAP 和非 GAAP 的毛利率將為 74%，誤差範圍為正負 50 個基點。我們預計毛利率在第四季度觸底，範圍在 71% 到 72% 之間。然後預計毛利率在 28 財年將穩定在 72% 到 73% 之間，因為執行的價格上漲將在第一季度生效。我們希望對此直言不諱，而不是讓它成為一個懸而未決的問題。

Colette Kress - NVIDIA, CFO

(00:25:47)Memory scarcity today is largely driven by the AI build-out itself and, unlike a component that simply raises our costs without offsetting benefits, tighter memory supply is a symptom of the same demand surge that is driving our own growth. We have longstanding, deep relationships with all three major memory suppliers and are working closely with them to further increase the capacity our roadmap requires. GAAP and non-GAAP operating expenses are expected to be approximately $9.2 billion and $9.0 billion, respectively.

(00:25:47)當前的記憶體短缺主要是由 AI 建設本身驅動的，與單純提高我們成本而沒有相應好處的元件不同，更緊張的記憶體供應是驅動我們自身增長的同一需求激增的症狀。我們與所有三家主要記憶體供應商有著長期而深厚的關係，並與他們密切合作，以進一步增加我們路線圖所需的產能。GAAP 和非 GAAP 的營運支出預計分別約為 92 億美元和 90 億美元。

Colette Kress - NVIDIA, CFO

(00:26:27)For the full year, we now expect OPEX to grow in the low 50% range, driven by a broadening product portfolio and increased use of AI tools, which are already enhancing engineering productivity and will continue to do so. For fiscal year 27, we continue to expect GAAP and non-GAAP tax rates to be between 16% and 18%, excluding any discrete items or material changes to our tax environment.

(00:26:27)對於整個年度，我們現在預計 OPEX 將增長在低 50% 的範圍內，這是由於產品組合的擴大和 AI 工具的使用增加，這些工具已經提高了工程生產力，並將繼續這樣做。對於 27 財年，我們仍然預計 GAAP 和非 GAAP 的稅率將在 16% 到 18% 之間，不包括任何一次性項目或對我們稅務環境的重大變化。

Q&A Colette Kress - NVIDIA, CFO

(00:26:57)With that, we will now transition to the Q&A session. Operator, please open the line for questions.

(00:26:57)有鑑於此，我們現在將過渡到問答環節。操作員，請開放提問線。

Tiffany - Operator

(00:27:07)At this time, I would like to remind everyone that to ask a question, please press star then the number one on your telephone keypad. We'll pause briefly to compile the Q&A roster. Your first question comes from Joseph Moore with Morgan Stanley. Your line is open.

(00:27:07)此時，我想提醒大家，若要提問，請按電話鍵盤上的星號，然後按數字一。我們將暫停片刻以編制問答名單。您的第一個問題來自摩根士丹利的 Joseph Moore。您的線路已開通。

Joseph Moore - Morgan Stanley

(00:27:29)Great. Thank you. I wonder if you could provide color on the 70% growth and what gives you confidence to guide a full year out, especially if you haven't been doing that previously? Also, what explains the gap between that growth and the 100% demand growth? What is the key constraint that separates those numbers? And could you close those gaps over time?

(00:27:29)很好。謝謝。我想知道您是否可以對 70% 的增長提供一些背景，以及您有什麼信心能夠指導整個年度，特別是如果您之前沒有這樣做的話？此外，什麼解釋了該增長與 100% 需求增長之間的差距？什麼是將這些數字分開的關鍵限制？您能否隨著時間的推移縮小這些差距？

Jensen Huang - NVIDIA, CEO

(00:27:51)Yeah, thanks, Joe. As you probably know, AI has become very useful. And the AI agents being adopted everywhere use an enormous amount of compute. First, the large language models are larger than ever because they are smarter than ever. These agents perform reasoning and planning, involving multiple turns of tool use. The compute required for an agent compared to a human using it is probably 15 to 100 times greater, depending on the problem type. So, the amount of compute necessary is truly extraordinary.

(00:27:51)是的，謝謝你，Joe。如您所知，AI 已變得非常有用。而且AI 代理在各地被採用，使用了大量的計算資源。首先，大型語言模型比以往更大，因為它們比以往更聰明。這些代理進行推理和計劃，涉及多次工具使用。相較於人類使用它，代理所需的計算量可能大 15 到 100 倍，具體取決於問題類型。因此，所需的計算量確實是非凡的。

Jensen Huang - NVIDIA, CEO

(00:28:31)That is a factor almost everyone recognizes. The part that people don't see about our growth is that we are practically singular because of how we deliver products. We are the only company in the world that creates, builds, and offers a complete AI factory platform—a full-stack system. Customers still have the flexibility to mix and match components. However... Most companies lack the skills or desire to do that themselves. Therefore, there is a significant part of the market where we are experiencing growth.

(00:28:31)這是幾乎每個人都認識到的一個因素。人們對我們增長的看法是，我們幾乎是獨一無二的，因為我們的產品交付方式。我們是全球唯一一家創建、建設並提供完整 AI 工廠平台的公司——一個全堆疊系統。客戶仍然可以靈活地混合和匹配組件。然而……大多數公司缺乏自己這樣做的技能或意願。因此，市場上有一個重要部分正在經歷增長。

Jensen Huang - NVIDIA, CEO

(00:29:05)This includes sovereign AI, regional AIs, neoclouds, AI startups, and enterprises, which together represent about half of our business and are growing at 100% annually. This segment of the world's computing is likely to become larger over time than even the current cloud market. The demand we see is driven by all these factors. It is also true that you can no longer simply procure technology and set up this infrastructure. You must secure land, power, and shell, which often takes two to three years.

(00:29:05)這包括主權 AI、區域 AI、新雲、AI 初創公司和企業，這些合起來約佔我們業務的一半，並以每年 100% 的速度增長。這部分全球計算市場隨著時間的推移可能會變得比當前的雲市場更大。我們看到的需求是由所有這些因素驅動的。也確實如此，你不能再僅僅採購技術並建立這個基礎設施。你必須確保土地、電力和外殼，這通常需要兩到三年。

Jensen Huang - NVIDIA, CEO

(00:29:48)Additionally, the entire supply chain must be aligned, including construction, power, cooling, and all necessary labor. AI infrastructure is creating numerous jobs across the United States. And globally, it requires much more planning. We are now involved in securing infrastructure further down the pipeline. Long ago, people asked why we worked with memory suppliers when we are a chip company. Today, it is understood that working upstream in our supply chain was a strategic move. We collaborate with power generator companies downstream.

(00:29:48)此外，整個供應鏈必須協調一致，包括建設、電力、冷卻和所有必要的勞動力。AI 基礎設施正在美國創造大量工作機會。而在全球範圍內，這需要更多的規劃。我們現在參與確保更下游的基礎設施。很久以前，人們問我們為什麼與記憶供應商合作，當我們是一家晶片公司時。今天，人們明白在我們的供應鏈上游工作是一個戰略舉措。我們與下游的發電公司合作。

Jensen Huang - NVIDIA, CEO

(00:30:30)We also work with land, power, and shell companies worldwide. This preparation supports all the computing infrastructure that will be built and deployed for our ecosystem and customers. As a result, we now have much greater visibility both upstream and downstream. We have never forecasted or guided a full year in advance. Although our demand exceeds 70%, our supply allows us to confidently deliver 70%. We will continue working with our supply chain to increase that capacity. Our goal is to provide consistent information so customers, shareholders, and suppliers all share the same view.

(00:30:30)我們還與全球的土地、電力和外殼公司合作。這些準備支持將為我們的生態系統和客戶建設和部署的所有計算基礎設施。因此，我們現在在上游和下游都有了更大的可見性。我們從未預測或指導過一整年的預測。雖然我們的需求超過 70%，但我們的供應使我們能夠自信地交付 70%。我們將繼續與供應鏈合作以增加該能力。我們的目標是提供一致的信息，以便客戶、股東和供應商都能共享相同的觀點。

Jensen Huang - NVIDIA, CEO

(00:31:22)This consistency is important because everyone is allocating significant resources. We want to ensure everyone has access to the same information. We have a huge year ahead. It is going to be quite extraordinary.

(00:31:22)這種一致性很重要，因為每個人都在分配大量資源。我們希望確保每個人都能獲得相同的信息。我們有一個巨大的年度計劃。這將是相當非凡的。

Tiffany - Operator

(00:31:43)Your next question comes from CJ Muse with Cantor Fitzgerald. Your line is open.

(00:31:43)您的下一個問題來自 Cantor Fitzgerald 的 CJ Muse。您的線路已開通。

CJ Muse - Cantor Fitzgerald

(00:31:50)Good afternoon. Thank you for taking my question. There is tremendous investor focus on your inference market share. Can you discuss the evolving workloads you are seeing with generative AI and how you expect your market share to evolve over time, especially considering the growing TAM with each new full-stack generation, your expectations for greater growth from AI, and the inclusion of Groq 3 LPX? We would appreciate your insights.

(00:31:50)下午好。感謝您回答我的問題。投資者對您的推理市場份額非常關注。您能否討論一下您在生成 AI 中看到的演變工作負載，以及您預期市場份額如何隨著時間的推移而演變，特別是考慮到每一代全堆疊的 TAM 不斷增長、您對 AI 更大增長的期望，以及 Groq 3 LPX 的納入？我們將感謝您的見解。

Jensen Huang - NVIDIA, CEO

(00:32:16)Thank you, CJ. The AI lifecycle is becoming much more complex than before. It is increasingly integrated... into NVIDIA's architecture far more deeply than in the past. You can think of it as four phases. The first phase involves preparing all the necessary data. Some of this data is synthetic. Some is real. Some is generated and labeled by human labor. Then, the models are pre-trained. The third phase is post-training. Finally, there is the agentic inference phase. Agentic inference is extremely complex. Each of these phases is complicated.

(00:32:16)謝謝你，CJ。AI 生命週期變得比以前更為複雜。它越來越深入地整合進入……NVIDIA 的架構中，遠比過去更深。您可以將其視為四個階段。第一階段涉及準備所有必要的數據。其中一些數據是合成的。有些是真實的。有些是由人力生成和標記的。然後，模型會進行預訓練。第三階段是後訓練。最後，進入代理推理階段。代理推理是極其複雜的。這些階段中的每一個都是複雜的。

Jensen Huang - NVIDIA, CEO

(00:32:59)The remarkable aspect of the NVIDIA architecture is that we developed it with NVLink 72. It was a major breakthrough when we first introduced the world's first rack-scale architecture. Building the first generation was extremely challenging and far from easy. We are now in the third generation of NVLink 72 rack-scale systems. We had to reinvent the entire supply chain, systems, technology, and completely refactor our software—every aspect was difficult. However, this enabled us to create a single fungible system that transitions seamlessly from data creation and preparation, through pre-training and post-training, to agentic inference.

(00:32:59)NVIDIA 架構的顯著特點是我們使用 NVLink 72 開發了它。當我們首次推出全球首個機架規模架構時，這是一個重大突破。建造第一代產品是極具挑戰性的，遠非易事。我們現在已經進入 NVLink 72 機架規模系統的第三代。我們不得不重新構建整個供應鏈、系統、技術，並完全重構我們的軟體——每一個方面都很困難。然而，這使我們能夠創建一個單一的可替代系統，無縫地從數據創建和準備，經過預訓練和後訓練，轉換到代理推理。

Jensen Huang - NVIDIA, CEO

(00:33:48)The benefits to customers are incredible. This is because, as we mentioned, each gigawatt of technology and NVIDIA's revenue exposure in the Hopper timeframe includes Hopper plus InfiniBand. Now, with Vera Rubin CPU and three different types of networking, it requires that many networking types to address the global data center needs, including scale in security and multi-campus networking. So, you could say there are five different types of networking systems involved. And, of course, Groq. All of that has increased our revenue contribution or revenue opportunity per gigawatt to $40 billion.

(00:33:48)對客戶的好處是不可思議的。這是因為，正如我們提到的，每一吉瓦的技術和 NVIDIA 在 Hopper 時間範圍內的收入暴露包括 Hopper 加上 InfiniBand。現在，隨著 Vera Rubin CPU 和三種不同類型的網絡，它需要這麼多的網絡類型來滿足全球數據中心的需求，包括安全性擴展和多校園網絡。所以，你可以說涉及五種不同類型的網絡系統。當然，還有 Groq。所有這些都增加了我們的每吉瓦的收入貢獻或收入機會達到 400 億美元。

Jensen Huang - NVIDIA, CEO

(00:34:35)Each gigawatt of data center investment has grown from about $30 billion five years ago to $60 billion today. Naturally, the productivity gains are tremendous. The performance improvements are incredible by comparison. We're talking about a $60 billion investment. And since it can be used across multiple phases of the AI lifecycle, running every type of model imaginable—diffusion, autoregressive, state space, or hybrids—every AI variant. Every attention mechanism, whether for small or large models, ensures that the investment remains valuable and productive for a long time.

(00:34:35)每吉瓦的數據中心投資從五年前的約 300 億美元增長到今天的 600 億美元。自然，生產力的提升是巨大的。相比之下，性能的改善是不可思議的。我們談論的是 600 億美元的投資。而且，由於它可以用於 AI 生命週期的多個階段，運行每一種想像中的模型——擴散、自回歸、狀態空間或混合模型——每一種 AI 變體。每一種注意力機制，無論是針對小型還是大型模型，都確保了投資在長期內保持有價值和高效。

Jensen Huang - NVIDIA, CEO

(00:35:31)I believe our advantage in this new world is truly extraordinary. This could explain why our growth is actually accelerating. It was already substantial, but now it's gaining momentum. You asked about Groq. We are very excited about Groq 3. We achieved a record token interactivity rate with extremely low latency in performance generation. The team is performing fantastically. Over the past several months, we have been integrating the NVLink architecture, which will serve as the core engine. And then for services that require super high interactivity and very fast token generation, the throughput will be lower.

(00:35:31)我相信我們在這個新世界中的優勢確實非凡。這可能解釋了為什麼我們的增長實際上正在加速。它已經相當可觀，但現在正在獲得動力。你問到了 Groq。我們對 Groq 3 感到非常興奮。我們在性能生成中達到了創紀錄的令牌互動率，延遲極低。團隊的表現非常出色。在過去幾個月中，我們一直在整合 NVLink 架構，這將作為核心引擎。然後對於需要超高互動性和非常快速令牌生成的服務，吞吐量將會較低。

Jensen Huang - NVIDIA, CEO

(00:36:27)The cost per token will be higher, but this can be associated with high ASP services. For those companies, you can add one of our Groq accelerators. I'm very excited about that. However, the vast majority of the world's data centers will primarily use Vera Rubin and NVLink 72.

(00:36:27)每個令牌的成本將會更高，但這可以與高 ASP 服務相關聯。對於那些公司，你可以添加我們的 Groq 加速器之一。我對此感到非常興奮。然而，世界上絕大多數的數據中心將主要使用 Vera Rubin 和 NVLink 72。

Tiffany - Operator

(00:36:49)Your next question comes from Stacy Rasgon with Bernstein Research. Your line is open.

(00:36:49)你的下一個問題來自 Bernstein Research 的 Stacy Rasgon。你的線路已開通。

Stacy Rasgon - Bernstein Research

(00:36:57)Hi, everyone. Thank you for taking my questions. Regarding the 70% growth in fiscal 2028, which corresponds roughly to calendar 2027, that represents about a $200 billion increase compared to the prior outlook. Previously, the outlook was for a trillion dollars over three years. So this is approximately $200 billion more. Could you walk us through the contributors to this increase across different products like URA, CPUs, Groq, and others? Also, you mentioned a price increase taking effect in Q1, so I assume some of this growth is due to pricing.

(00:36:57)嗨，大家好。感謝你們回答我的問題。關於 2028 財年的 70% 增長，這大約對應於 2027 日曆年，這代表與之前的預測相比約增加 2000 億美元。之前的預測是三年內達到一萬億美元。所以這大約是多了 2000 億美元。你能否向我們解釋一下這一增長的貢獻者，包括 URA、CPU、Groq 和其他產品？此外，你提到在第一季度會有價格上漲，所以我假設這部分增長是由於定價。

Stacy Rasgon - Bernstein Research

(00:37:32)And I realize I might be asking too many questions, but I'm also curious since this is a constrained number, what would the figure be if it were unconstrained?

(00:37:32)我意識到我可能問了太多問題，但我也很好奇，因為這是一個受限的數字，如果不受限制，這個數字會是多少？

Jensen Huang - NVIDIA, CEO

(00:37:45)The unconstrained number would be significantly higher. We grew 100% year over year this year. The unconstrained opportunity is substantial. So, we're going to have to work hard to increase our capacity. We have a large supply chain. In fact, our supply chain is enormous. We have incredible partners and have secured a significant amount of supply. However, we need much more to meet demand. To put it simply, most people only see the hyperscalers. But that's only half the picture. The other half is what we call AC.

(00:37:45)不受限制的數字將會高得多。我們今年的年增長率達到 100%。不受限制的機會是相當可觀的。所以，我們必須努力增加我們的產能。我們有一個龐大的供應鏈。事實上，我們的供應鏈是巨大的。我們有令人難以置信的合作夥伴，並且已經確保了大量的供應。然而，我們需要更多的供應來滿足需求。簡單來說，大多數人只看到超大規模的公司。但這只是整體的一半。另一半是我們所稱的 AC。

Jensen Huang - NVIDIA, CEO

(00:38:30)This includes enterprises, neoclouds, and sovereign AI systems. That segment is largely invisible to most people. This is because they don't purchase custom chips individually. They don't buy chips one at a time. Instead, they require an entire factory platform built for their needs. We add tremendous value in that space. Of course, the hyperscale segment is also growing rapidly. That's correct. They now have backlogs totaling around two trillion dollars. When they deploy video compute, their revenues and earnings contributions increase exponentially.

(00:38:30)這包括企業、新雲端和主權 AI 系統。這一部分對大多數人來說幾乎是看不見的。這是因為他們不會單獨購買定製芯片。他們不會一次購買一個芯片。相反，他們需要一整個工廠平台來滿足他們的需求。我們在這個領域增添了巨大的價值。當然，超大規模的市場也在迅速增長。沒錯。他們現在的積壓訂單總額約為兩萬億美元。當他們部署視頻計算時，他們的收入和盈利貢獻會指數增長。

Jensen Huang - NVIDIA, CEO

(00:39:13)Compute is highly profitable today. Compute directly drives increased revenues. There is a race to bring more NVIDIA compute online, especially at hyperscalers. But there are also tremendous opportunities outside the hyperscalers. That's why we mapped this out for you. For Hopper, the value is about $18 billion. For Grace Blackwell, it's approximately $25 billion per gigawatt. For Vera Rubin, it's about $40 billion per gigawatt. Productivity increases by multiple factors with each generation. Customers want to move to the next generation as quickly as possible.

(00:39:13)現在計算是非常有利可圖的。計算直接推動收入的增長。現在有一場競賽，旨在讓更多的 NVIDIA 計算上線，特別是在超大規模公司中。但在超大規模公司之外也有巨大的機會。這就是為什麼我們為您繪製了這個圖。對於 Hopper，價值約為 180 億美元。對於 Grace Blackwell，每千瓦的價值約為 250 億美元。對於 Vera Rubin，每千瓦的價值約為 400 億美元。隨著每一代的進步，生產力會增加多倍。客戶希望儘快轉向下一代。

Jensen Huang - NVIDIA, CEO

[output truncated at 50000 of 69976 characters. Pass a larger max_chars (default 50000) to see more, or use read_page with a ref_id to focus on a smaller section.]

Jensen Huang - NVIDIA, CEO

(00:40:05)Because NVIDIA's compute is so productive, the tokens generated and GPU hours rented are extremely profitable. Their margins are excellent. All of this is happening simultaneously. The big picture is that we are undergoing a platform shift affecting every computer company. Every industry in the world uses computers. Therefore, every industry is impacted. Every company is affected. This new computing paradigm is intelligent. It is not based on file retrieval but is generative, producing intelligence. This requires significant compute power.

(00:40:05)由於 NVIDIA 的計算效率極高，生成的代幣和租用的 GPU 小時數非常有利可圖。他們的利潤率非常出色。所有這一切都是同時發生的。大局是我們正在經歷一場影響每家電腦公司的平台轉變。世界上每個行業都在使用計算機。因此，每個行業都受到影響。每家公司都受到影響。這種新的計算範式是智能的。它不是基於文件檢索，而是生成性的，產生智能。這需要大量的計算能力。

Jensen Huang - NVIDIA, CEO

(00:40:52)The results achieved are phenomenal. The outcomes are significantly improved. We are witnessing this trend worldwide. Everyone wants to be part of it. The AI revolution means everyone must participate in this computing shift and build infrastructure.

(00:40:52)所取得的結果是驚人的。結果顯著改善。我們在全球範圍內見證了這一趨勢。每個人都想成為其中的一部分。AI 革命意味著每個人都必須參與這一計算轉變並建立基礎設施。

Tiffany - Operator

(00:41:13)Your next question comes from Vivek Arya of Bank of America Securities. Your line is open.

(00:41:13)您的下一個問題來自美國銀行證券的 Vivek Arya。您的線路已開通。

Vivek Arya - BofA Securities

(00:41:22)Thank you for taking my question and for the transparency and commitments you've provided over the years. When I add up everything mentioned in the CFO commentary, I arrive at about $500 billion over the next several years. I have a few questions related to that. First, is it correct to understand that the total of your ecosystem investments over the next several years is in that range, or are there additional equity or other investments anticipated? That's one. Secondly, is there a specific cash component we should consider for fiscal 28?

(00:41:22)感謝您回答我的問題，以及多年來提供的透明度和承諾。當我將 CFO 評論中提到的所有內容加起來時，我得出未來幾年約 5000 億美元的數字。我有幾個相關的問題。首先，理解您未來幾年的生態系統投資總額在這個範圍內是否正確，還是預期會有額外的股權或其他投資？這是一個。其次，我們是否應該考慮在財政 28 中有特定的現金組成部分？

Vivek Arya - BofA Securities

(00:42:00)And Jensen, many of these investments aim to support frontier labs, especially OpenAI and Anthropic, but both are designing their own custom chips. In fact, OpenAI just recently spoke about their claims of being superior to Blackwell and so forth. How do you balance investing heavily in the ecosystem when part of it is developing competitive solutions? Thank you.

(00:42:00)而且，Jensen，這些投資中的許多旨在支持前沿實驗室，特別是 OpenAI 和 Anthropic，但兩者都在設計自己的定製芯片。事實上，OpenAI 最近剛剛談到他們聲稱優於 Blackwell 等。當其中一部分正在開發競爭解決方案時，您如何平衡在生態系統中的重投資？謝謝。

Jensen Huang - NVIDIA, CEO

(00:42:30)Well, we're building something very different. Many of these XPUs are inference-specific chips designed for one cloud, or one service, but NVIDIA offers a platform—an entire AI factory platform spanning the full AI lifecycle that can be used in any cloud. It's available in every cloud. You can run it anywhere. We'll help you set it up anywhere. So, we built something very different. All AI services will eventually want to operate globally. And those data centers won't necessarily be built solely by them.

(00:42:30)我們正在構建一些非常不同的東西。這些 XPUs 中的許多是專為一個雲端設計的推理專用芯片，或一項服務，但 NVIDIA 提供了一個平台——一個涵蓋整個 AI 生命週期的完整 AI 工廠平台，可以在任何雲端使用。它在每個雲端都可用。您可以在任何地方運行它。我們將幫助您在任何地方設置它。所以，我們構建了一些非常不同的東西。所有 AI 服務最終都希望在全球運行。而這些數據中心不一定僅由他們建造。

Jensen Huang - NVIDIA, CEO

(00:43:25)Also, I fully expect they will run on NVIDIA technology worldwide. I have 100% confidence our technology will remain extraordinary for them. From data processing to training, post-training, and agentic processing, our technology offers exceptional economics for them. They will use it. I'm very confident they will be our customers and partners for a long time. That said, stepping back, investing in these companies—and several AI labs we've backed—is a once-in-a-generation opportunity. My only regret is not investing more and sooner.

(00:43:25)此外，我完全預期他們將在全球範圍內運行 NVIDIA 技術。我對我們的技術將對他們保持卓越的信心是 100%。從數據處理到訓練、後訓練和代理處理，我們的技術為他們提供了卓越的經濟效益。他們會使用它。我非常有信心他們將成為我們的客戶和合作夥伴很長一段時間。話雖如此，退一步說，投資這些公司——以及我們支持的幾個 AI 實驗室——是一個千載難逢的機會。我唯一的遺憾是沒有投資更多和更早。

Jensen Huang - NVIDIA, CEO

(00:44:20)Two of these companies will likely go public soon, with others to follow. These will be among the most consequential technology companies in history. I'm delighted to be their friend. I'm delighted to partner with them. I'm delighted they're building an ecosystem on NVIDIA architecture. I'm delighted they're counting on us to scale. I have 100% confidence that over a long period, they will utilize NVIDIA Compute for much of their computing. So, I feel great about it.

(00:44:20)這兩家公司可能很快會上市，其他公司也會隨之而來。這些將成為歷史上最具影響力的科技公司之一。我很高興能成為他們的朋友。我很高興能與他們合作。我很高興他們正在基於 NVIDIA 架構建立生態系統。我很高興他們依賴我們來擴展。我對於在長期內，他們將利用 NVIDIA 計算進行大部分計算的信心是 100%。所以，我對此感到非常好。

Colette Kress - NVIDIA, CFO

(00:45:00)Vivek, let me add more about the commitments, specifically our supply commitments. This is essential. It's critical for raising Vera Rubin today and throughout next year. Most of these commitments are concentrated in the first three years, which we'll use to build necessary products. This also gives us confidence in our growth and revenue, given how much we've already aligned in commitments regarding supply and capacity needs.

(00:45:00)Vivek，讓我多說一些關於承諾，特別是我們的供應承諾。這是至關重要的。這對於今天和明年整個年度提高 Vera Rubin 至關重要。大多數這些承諾集中在前三年，我們將利用這些來建造必要的產品。這也讓我們對我們的增長和收入充滿信心，考慮到我們已經在供應和產能需求方面達成了多少承諾。

Tiffany - Operator

(00:45:40)Your next question comes from Timothy Arcuri with UBS. Your line is open.

(00:45:40)您的下一個問題來自 UBS 的 Timothy Arcuri。您的線路已開通。

Timothy Arcuri - UBS

(00:45:47)Hi, thanks a lot. Jensen, I want to ask about open source. There's much discussion about these models gaining workload share in the U.S. You're obviously well positioned with Nemotron, but on the other hand, much of the end demand is driven by these large frontier model companies. Many investors view open models as negative for the growth of those companies. How do you interpret and respond to that? Do you see the rise of open models as beneficial or ultimately negative for NVIDIA?

(00:45:47)嗨，非常感謝。Jensen，我想問一下開源的問題。關於這些模型在美國獲得工作負載份額的討論很多。您顯然在 Nemotron 方面處於良好位置，但另一方面，最終需求的很大一部分是由這些大型前沿模型公司驅動的。許多投資者認為開放模型對這些公司的增長是負面的。您如何解釋並回應這一點？您認為開放模型的興起對 NVIDIA 是有利的還是最終是負面的？

Jensen Huang - NVIDIA, CEO

(00:46:16)Thanks. The world will need both closed and open models. Both closed and open models are experiencing skyrocketing usage. I would say nearly all open models run on NVIDIA. This is because NVIDIA's global footprint is the largest. Additionally, our architecture is the most versatile. It's everywhere. It's in PCs and edge devices like DGX Spark, which is performing well, extending to robots, workstations, and on-premises data centers. Open models are performing exceptionally well. Closed models are also doing incredibly well.

(00:46:16)謝謝。世界將需要封閉和開放模型。封閉和開放模型的使用量都在急劇上升。我可以說幾乎所有的開放模型都運行在 NVIDIA 上。這是因為 NVIDIA 的全球足跡是最大的。此外，我們的架構是最具多樣性的。它無處不在。它在 PC 和邊緣設備中，如 DGX Spark，表現良好，並延伸到機器人、工作站和本地數據中心。開放模型的表現非常出色。封閉模型的表現也非常好。

Jensen Huang - NVIDIA, CEO

(00:47:11)Frontier Labs' sales are skyrocketing. Their margins are excellent. They are generating profitable tokens. Their only limitation is the amount of compute available. This is equally true for open models. Our position in open models is strong because the CUDA ecosystem is literally everywhere. Open models are foundational to nearly every AI startup and enterprise worldwide. They are vital to these organizations. The reason is that intelligence should be rented from strong, smart sources wherever possible.

(00:47:11)Frontier Labs 的銷售額正在急劇上升。他們的利潤率非常高。他們正在產生盈利的代幣。他們唯一的限制是可用的計算量。對於開放模型來說也是如此。我們在開放模型中的地位很強，因為 CUDA 生態系統幾乎無處不在。開放模型是全球幾乎所有 AI 初創公司和企業的基礎。它們對這些組織至關重要。原因是智能應該盡可能從強大、聰明的來源租用。

Jensen Huang - NVIDIA, CEO

(00:47:49)That is why we rent intelligence. I encourage my employees to use cloud services as much as possible. However, every major company, country, and startup needs to build their domain-specific, proprietary AI—their proprietary alpha. The advancement of open models to frontier levels has enabled this. One critical area where frontier models are vital is cybersecurity. Numerous cybersecurity companies leverage frontier models to operate distributed, continuously running systems. These autonomous cybersecurity systems defend networks effectively.

(00:47:49)這就是為什麼我們租用智能。我鼓勵我的員工盡可能使用雲服務。然而，每個主要公司、國家和初創公司都需要建立他們特定領域的專有 AI——他們的專有 alpha。開放模型的進步達到了前沿水平，使這一切成為可能。前沿模型至關重要的一個關鍵領域是網絡安全。許多網絡安全公司利用前沿模型來運行分佈式、持續運行的系統。這些自主網絡安全系統有效地保護網絡。

Jensen Huang - NVIDIA, CEO

(00:48:44)Such companies are emerging rapidly. There are some amazing companies in this space. They could not operate without open models. Open models are both incredibly successful and essential to the American economy. They are also vital to the global economy. They are crucial for companies building proprietary AI. You cannot succeed without either open or closed models. Both will be extraordinarily successful. Lastly, as you know, Regarding our market footprint across all AI models, we believe we are the only platform that runs every frontier model.

(00:48:44)這類公司正在迅速崛起。在這個領域有一些令人驚嘆的公司。他們無法在沒有開放模型的情況下運行。開放模型對美國經濟來說既成功又至關重要。它們對全球經濟也至關重要。它們對於建立專有 AI 的公司至關重要。沒有開放或封閉模型，你無法成功。兩者都將非常成功。最後，如你所知，關於我們在所有 AI 模型中的市場足跡，我們相信我們是唯一運行每個前沿模型的平台。

Jensen Huang - NVIDIA, CEO

(00:49:31)This includes both closed and open models. Most frontier models were built on NVIDIA. Consequently, they run exceptionally well on NVIDIA hardware. We are very pleased about this. I am happy whenever any model succeeds. Both closed and open models will succeed and are simultaneously driving our sales.

(00:49:31)這包括封閉和開放模型。大多數前沿模型都是在 NVIDIA 上構建的。因此，它們在 NVIDIA 硬件上運行得非常好。我們對此感到非常高興。每當任何模型成功時，我都感到高興。封閉和開放模型都將成功，並同時推動我們的銷售。

Tiffany - Operator

(00:49:59)Your next question comes from Ben Reitzes with Melius Research. Your line is open.

(00:49:59)你的下一個問題來自 Melius Research 的 Ben Reitzes。你的線路是開放的。

Ben Reitzes - Melius Research

(00:50:06)Yeah, hey, thanks. I wanted to ask, Jensen, a question about demand from a different perspective. You mentioned demand growing 100% next year, and I wanted to understand some of the drivers behind that and beyond. There are two concepts here. One is recursive self-improvement, which at Anthropic and OpenAI is progressing well with AI that improves itself. Even OpenAI has stated they could achieve AGI by the end of this year. With developments in RSI as well as AGI, what impact will this have on industry demand?

(00:50:06)是的，嘿，謝謝。我想問，Jensen，關於需求的問題，從不同的角度來看。你提到明年需求增長 100%，我想了解一些背後的驅動因素及其以外的因素。這裡有兩個概念。一個是遞歸自我改進，在 Anthropic 和 OpenAI 中，這一進展良好，AI 正在自我改進。甚至 OpenAI 也表示他們可能在今年年底實現 AGI。隨著 RSI 和 AGI 的發展，這將對行業需求產生什麼影響？

Ben Reitzes - Melius Research

(00:50:44)Will demand inflect further? And what implications does this have for NVIDIA when these developments occur? How are you viewing this as a catalyst for demand? Thank you.

(00:50:44)需求會進一步增長嗎？當這些發展發生時，這對 NVIDIA 有什麼影響？你如何看待這作為需求的催化劑？謝謝你。

Jensen Huang - NVIDIA, CEO

(00:50:55)I appreciate the question. Demand is going to inflect further. Currently, the vast majority of AI is prompted by humans. I believe that just last month, this changed. Most AI systems are now agentic. In the future, every company will have numerous agents. We currently have roughly 40,000 employees. In the future, we will have 400,000 or even 4 million agents. These agents operate continuously. They run in the background. If you know anyone building edge personal AI agents, they often run them on a DGX Spark.

(00:50:55)我很感謝這個問題。需求將會進一步增長。目前，絕大多數的 AI 是由人類驅動的。我相信就在上個月，這一點發生了變化。現在大多數 AI 系統都是自主的。在未來，每家公司將擁有眾多的代理。我們目前大約有 40,000 名員工。在未來，我們將擁有 400,000 甚至 400 萬名代理。這些代理會持續運作。它們在背景中運行。如果你認識任何正在構建邊緣個人 AI 代理的人，他們通常會在 DGX Spark 上運行它們。

Jensen Huang - NVIDIA, CEO

(00:51:37)I know many who run these agents on DGX Stations, an incredible workstation we built that you can purchase from Dell; they are truly impressive. These AI agents running on a DGX Station operate 24/7 because there is always work for them to do. When the world fully adopts agentic systems, agents will be running constantly, collaborating with other agents. These agents will work in the background, improving companies and enhancing lives. In many ways, we are already at a recursive point. You could argue this is coarse-grained, but each time an agent runs, it reflects on how to improve and updates its skill file.

(00:51:37)我知道很多人在 DGX Station 上運行這些代理，這是一個我們建造的令人驚嘆的工作站，你可以從 Dell 購買；它們真的很令人印象深刻。在 DGX Station 上運行的這些 AI 代理全天候運作，因為它們總是有工作要做。當世界完全採用自主系統時，代理將不斷運行，與其他代理協作。這些代理將在背景中工作，改善公司並提升生活。在許多方面，我們已經達到了一個遞歸的點。你可以說這是粗略的，但每次代理運行時，它都會反思如何改進並更新其技能文件。

Jensen Huang - NVIDIA, CEO

(00:52:28)The skills document, typically in markdown, is updated after every run. So, the next time it runs, it performs better. This represents a loosely coarse-grained form of self-improvement. You see this pattern everywhere, consistently. In many respects, For many tasks, one could say we have already achieved AGI. I believe these milestones have become somewhat meaningless at this point. The most important factors for the industry are: one, AI is now performing productive and useful work. Two, AI is generating profitable tokens.

(00:52:28)技能文件通常是以 markdown 格式更新的，每次運行後都會更新。所以，下次運行時，它的表現會更好。這代表了一種鬆散的粗略自我改進形式。你在各處都能看到這種模式，一直如此。在許多方面，對於許多任務，可以說我們已經實現了 AGI。我相信這些里程碑在此時已經變得有些無意義。對於行業來說，最重要的因素是：第一，AI 現在正在執行生產性和有用的工作。第二，AI 正在生成有利可圖的代幣。

Jensen Huang - NVIDIA, CEO

(00:53:11)And three, with more compute, we could generate more profitable tokens, resulting in greater profits for all services. This is exactly the phase we are in, which explains why everyone is fully engaged.

(00:53:11)第三，隨著計算能力的增加，我們可以生成更多有利可圖的代幣，從而為所有服務帶來更大的利潤。這正是我們所處的階段，這解釋了為什麼每個人都全力以赴。

Tiffany - Operator

(00:53:26)Your next question comes from Jim Schneider of Goldman Sachs. Your line is open.

(00:53:26)你的下一個問題來自高盛的 Jim Schneider。你的線路已開通。

Jim Schneider - Goldman Sachs

(00:53:34)Good afternoon. Thank you for taking my question. Regarding the 100% growth you mentioned in unconstrained demand, you expect to fulfill 70% in supply. Could you discuss the rank order of the most acute constraints, such as data center power and shell availability, DRAM, wafer foundry capacity, etc.? If you could help us understand which constraints are the most significant, that would be very helpful. Thank you.

(00:53:34)下午好。謝謝你接受我的問題。關於你提到的在無約束需求中 100% 的增長，你預計供應能滿足 70%。你能談談最嚴重的約束的排名，例如數據中心的電力和外殼可用性、DRAM、晶圓廠的產能等嗎？如果你能幫助我們瞭解哪些約束是最重要的，那將非常有幫助。謝謝你。

Jensen Huang - NVIDIA, CEO

(00:54:11)There's something humorous I could say, but I will refrain. Last year, one of the more amusing things was figuring out where I was going for dinner and with whom. Often, their stock price would double the next day. I believe the answer is that our entire supply chain is challenged. Everyone is operating at full capacity. More capacity is coming online continuously, which is one advantage expected this year. This capacity won't come online all at once but will be added daily. Yields are going to improve.

(00:54:11)我可以說一些幽默的話，但我會剋制。去年，有一件比較有趣的事情是弄清楚我將去哪裡吃晚餐以及和誰一起。通常，他們的股價會在第二天翻倍。我相信答案是我們整個供應鏈都面臨挑戰。每個人都在全力運作。更多的產能正在不斷上線，這是今年預期的一個優勢。這些產能不會一次性上線，而是每天增加。收益將會改善。

Jensen Huang - NVIDIA, CEO

(00:54:58)We will be focusing on yield improvement. We are committed to working closely with each of our suppliers, and since it's not even next year yet, we have plenty of time to work hard every day. Currently, we have supply for about 70% of demand. We have more supply than 70%, but approximately 70%. Our demand significantly exceeds that level. We need to work hard, or else, we risk disappointing our customers. And we do not want to disappoint our customers. We are committed to working hard for them. I will need the full support of the entire supply chain to assist me here.

(00:54:58)我們將專注於提高收益。我們致力於與每一位供應商密切合作，而由於明年還沒有到來，我們有足夠的時間每天努力工作。目前，我們的供應大約能滿足 70% 的需求。我們的供應超過 70%，但大約是 70%。我們的需求顯著超過這個水平。我們需要努力工作，否則，我們有可能讓客戶失望。我們不想讓客戶失望。我們致力於為他們努力工作。我需要整個供應鏈的全力支持來協助我。

Jensen Huang - NVIDIA, CEO

(00:55:44)Everyone involved is aware of this. What I am sharing about our needs for next year is exactly what I have communicated to them. Everyone is on the same page, and I am striving to be as transparent as possible because we are discussing large numbers.

(00:55:44)參與的每個人都知道這一點。我所分享的關於明年需求的內容正是我與他們溝通的內容。每個人都在同一頁面上，我努力做到盡可能透明，因為我們在討論大數字。

Tiffany - Operator

(00:56:06)Your final question comes from Aaron Rakers of Wells Fargo. Your line is now open.

(00:56:06)你的最後一個問題來自於 Wells Fargo 的 Aaron Rakers。你的線路現在已經開通。

Aaron Rakers - Wells Fargo

(00:56:13)Thank you for taking my question. I want to revisit the gigawatt figures, specifically the 25 to 40 gigawatts. Jensen, you mentioned at recent conferences that this will scale further. As we consider the path beyond Vera Rubin, including Vera Rubin Ultra and beyond, should we conceptualize growth from 40 billion to 60 billion or even 80 billion? Underlying that question is how we should think about your capacity to scale deployments. Is it a linear progression, or is there a factor that unlocks greater supply capacity as we approach fiscal 2027?

(00:56:13)感謝你回答我的問題。我想重新討論吉瓦數字，特別是 25 到 40 吉瓦。Jensen，你在最近的會議上提到這將進一步擴展。當我們考慮超越 Vera Rubin 的路徑，包括 Vera Rubin Ultra 及其後，我們應該將增長概念化為從 400 億到 600 億甚至 800 億嗎？這個問題的根本在於我們應該如何看待你們擴大部署的能力。這是一個線性進展，還是有某種因素在接近 2027 財年時解鎖更大的供應能力？

Jensen Huang - NVIDIA, CEO

(00:56:56)That's a great question. Or fiscal 2028, sorry. Yes, great question. Indeed, a great question. The answer is quite simple. Our goal is to maximize compute capacity on a given plot of land. We aim to increase compute per gigawatt, not decrease it. Obviously, we would prefer that. The ideal answer would be infinite compute per gigawatt. If we could fit a trillion dollars' worth of compute into one gigawatt and a single power shell, that would be an outstanding outcome. So, directionally, that is the direction we are heading.

(00:56:56)這是一個很好的問題。或者 2028 財年，抱歉。是的，很好的問題。確實是一個很好的問題。答案相當簡單。我們的目標是在特定的土地上最大化計算能力。我們的目標是提高每吉瓦的計算能力，而不是降低它。顯然，我們會更喜歡這樣。理想的答案是每吉瓦無限的計算能力。如果我們能在一吉瓦和一個電源殼中放入價值一萬億美元的計算，那將是一個卓越的結果。所以，方向上，我們正朝著這個方向前進。

Jensen Huang - NVIDIA, CEO

(00:57:38)We began in the era of general-purpose computing during Moore's Law. Back then, the compute value was roughly $3 to $5 billion per gigawatt. That was with general-purpose computing. With Hopper architecture, it increased to $18 billion. Grace Blackwell now represents $25 billion. Next, Vera Rubin will represent $40 billion. And beyond that, the value will be even higher. This is excellent news. It's fantastic for the industry. It's also fantastic for customers. As long as productivity, durability, and fungibility continue to improve, investors remain confident.

(00:57:38)我們開始於摩爾定律的通用計算時代。當時，計算價值大約是每吉瓦 30 到 50 億美元。那是通用計算的情況。隨著 Hopper 架構的出現，這一數字增加到 180 億美元。Grace Blackwell 現在代表 250 億美元。接下來，Vera Rubin 將代表 400 億美元。而在此之上，價值將更高。這是個好消息。對於行業來說，這是個好消息。對於客戶來說，這也是個好消息。只要生產力、耐用性和可替代性持續改善，投資者就會保持信心。

Jensen Huang - NVIDIA, CEO

(00:58:18)They invest in assets that generate revenue, profits, and enable rapid return on investment. I recently heard that the return on invested capital is now less than one year. And we are discussing $50 billion data centers. This highlights the productivity and profitability of NVIDIA's technology. I want to thank everyone for joining us today.

(00:58:18)他們投資於能夠產生收入、利潤並實現快速投資回報的資產。我最近聽說，投資資本的回報現在少於一年。我們正在討論 500 億美元的數據中心。這突顯了 NVIDIA 技術的生產力和盈利能力。我想感謝大家今天的參與。

Tiffany - Operator

(00:58:53)There are no further questions at this time. Toshiya Hari, I will now turn the call back over to you.

(00:58:53)目前沒有其他問題。Toshiya Hari，現在我將把電話交回給你。

Toshiya Hari - NVIDIA, Investor Relation

(00:58:59)Thank you. Before we close, please note that Jensen will participate in a keynote fireside chat at the Goldman Sachs Communicopia and Technology Conference in San Francisco on September 10th. He will also deliver a keynote at GTC Berlin on October 21st. Our earnings call to discuss the results of the third quarter of fiscal 2027 is scheduled for November 17th. Thank you for joining us today. Operator, please close the call.

(00:58:59)謝謝。在我們結束之前，請注意，Jensen 將於 9 月 10 日在舊金山的高盛 Communicopia 和科技大會上參加主題座談會。他還將於 10 月 21 日在 GTC 柏林發表主題演講。我們將於 11 月 17 日舉行財政年度 2027 年第三季度業績的電話會議。感謝您今天的參與。操作員，請結束通話。

Tiffany - Operator

(00:59:27)This concludes today's conference call. You may now disconnect.

(00:59:27)這次電話會議到此結束。您現在可以斷開連接。