International Business Machines Corp (IBM) 2026 Q3 法說會逐字稿

Welcome and thank you for standing by. (Operator Instructions) Today's conference is being recorded. If you have any objections, you may disconnect at this time.

歡迎各位，感謝您稍候。(接線員指示) 今天的電話會議將被錄音。如有任何異議，您可於此時中斷連線。

Now, I will turn the meeting over to Olympia McNerney, IBM's Global Head of Investor Relations. Olympia, you may begin.

現在，我將把會議交給 IBM 全球投資人關係主管 Olympia McNerney。Olympia，您可以開始了。

Thank you. I'd like to welcome you to IBM's third-quarter 2026 earnings presentation.

謝謝。歡迎各位參加 IBM 2026 年第三季財報簡報。

I'm Olympia McNerney. And I'm here today with Arvind Krishna, IBM's Chairman, President, and Chief Executive Officer; and Jim Kavanaugh, IBM's Senior Vice President and Chief Financial Officer.

我是 Olympia McNerney。今天與我一同出席的有 IBM 董事長、總裁兼執行長 Arvind Krishna，以及 IBM 資深副總裁兼財務長 Jim Kavanaugh。

We'll post today's prepared remarks and a replay of today's webcast on the IBM Investor website within a couple of hours. The earnings presentation is already available.

我們將在數小時內於 IBM 投資人網站上發布今天的事先準備講稿以及今日網路直播的重播。本次財報簡報已可取得。

To provide additional information to our investors, our presentation includes certain non-GAAP measures. For example, all of our references to revenue and signings growth are at constant currency. We provided reconciliation charts for these and other non-GAAP financial measures at the end of the presentation, which is posted to our Investor website.

為向投資人提供更多資訊，我們的簡報包含若干非 GAAP 指標。例如，我們所有對營收與簽約成長的提及皆以固定匯率計算。我們在簡報末尾提供了這些及其他非 GAAP 財務指標的調節表，並已張貼於我們的投資人網站。

Finally, some comments made in this presentation may be considered forward-looking under the Private Securities Litigation Reform Act of 1995. These statements involve factors that could cause our actual results to differ materially. Additional information about these factors is included in the company's SEC filings.

最後，本簡報中的部分評論可能依據 1995 年《私人證券訴訟改革法》被視為前瞻性陳述。這些陳述涉及可能導致我們實際結果出現重大差異的因素。關於這些因素的更多資訊載於公司向 SEC 提交的文件中。

So with that, I'll turn the call over to Arvind.

那麼，接下來我把電話交給 Arvind。

Thank you for joining us today.

感謝各位今天加入我們。

Revenue growth in the third quarter accelerated to 6% above the base case we laid out in July. This performance reflects acceleration across our segments, led by Software growth of 9%, Infrastructure growth of 5%, and modest improvement in Consulting.

第三季營收成長加速至 6%，高於我們在 7 月提出的基準情境。此一表現反映各業務部門普遍加速，其中以軟體成長 9%、基礎架構成長 5%領先，顧問服務亦有小幅改善。

We entered the quarter focused on disciplined execution, which also drove strong performance in profitability and free cash flow.

我們在本季開始時聚焦於嚴謹且有紀律的執行，這也帶動了獲利能力與自由現金流的強勁表現。

These results demonstrate the strength of our portfolio, the continued value clients place in mission-critical platforms, and our confidence in both our strategy and the outlook for the balance of the year.

這些結果展現了我們產品組合的實力、客戶對關鍵任務平台持續重視的價值，以及我們對策略與今年剩餘期間展望的信心。

Technology innovation in both hybrid infrastructure and Super Intelligence is leading to better business performance.

混合式基礎架構與超級智慧（Super Intelligence）兩方面的技術創新，正帶來更佳的業務表現。

This quarter's results reinforce the strategic choices we have made over the last several years to advance IBM as a software-led hybrid-cloud and AI-platform company.

本季結果再次印證了我們過去數年所做的策略選擇，推動 IBM 成為以軟體為主導的混合雲與 AI 平台公司。

In Software, Red Hat, again, grew 11%, with bookings growth accelerating back to the high-teens. OpenShift ARR growth accelerated to over 30%, as organizations scale Super Intelligence applications and agents into production, reinforcing our conviction that we remain in the early stages of a long-term application-modernization cycle.

在軟體方面，Red Hat 再次成長 11%，訂單（bookings）成長加速回到高十幾%的水準。OpenShift 的 ARR 成長加速至超過 30%；隨著各組織將超級智慧應用與代理（agents）擴大導入生產環境，進一步強化我們的信念：我們仍處於長期應用現代化週期的早期階段。

Transaction processing grew 10%, underscoring the strategic importance of the mainframe and validating our view that the softness we saw in the second quarter reflected timing, rather than underlying demand. Data grew 20%, supported by continued strength in Confluent.

交易處理成長 10%，凸顯主機（mainframe）的策略重要性，並驗證我們的看法：第二季所見的疲軟反映的是時點因素，而非基本需求。資料業務成長 20%，受惠於 Confluent 持續強勁的表現。

Infrastructure was up 5%, driven by continued momentum in distributed Infrastructure, growing 32% this quarter.

基礎架構成長 5%，主要由分散式基礎架構持續動能帶動，本季成長 32%。

AI continues to increase demand for storage. And we are gaining share through differentiated offerings across Flash, Fusion, Tape, and IBM Z storage, including AI-enabled capabilities that help clients scale and manage data for AI.

AI 持續推升對儲存的需求。我們透過在 Flash、Fusion、Tape 與 IBM Z 儲存等領域的差異化產品持續提升市占，其中包含 AI 啟用的能力，協助客戶為 AI 擴展並管理資料。

Consulting revenue grew 2%, as clients modernized to deploy AI securely and at scale. The durability of our portfolio remains a defining feature of IBM today.

顧問服務營收成長 2%，因客戶進行現代化以安全且可擴展地部署 AI。我們產品組合的韌性仍是今日 IBM 的關鍵特徵。

z17 continued the strongest innovation cycle in the platform's history, generating more than $1 billion of incremental revenue over six quarters compared with prior programs. This reflects the security, reliability, scalability, and compelling economics of the platform.

z17 延續了該平台史上最強的創新週期，相較於先前方案，在六個季度內帶來超過 10 億美元的增量營收。這反映了平台的安全性、可靠性、可擴展性與具吸引力的經濟效益。

As infrastructure costs rise, IBM Z delivers a 2 times to 15 times total-cost-of-ownership advantage versus off-platform alternatives, positioning us to capture additional value, as AI workloads grow.

隨著基礎架構成本上升，IBM Z 相較於非平台替代方案，在總持有成本（TCO）上可提供 2 倍至 15 倍的優勢，使我們在 AI 工作負載成長之際得以擷取更多價值。

Building on that advantage, we introduced new single-frame and rack-mount IBM Z systems in July that help clients address data-center space, cost, and deployment constraints without compromising performance or security.

在此優勢基礎上，我們於 7 月推出新的單機框（single-frame）與機架式（rack-mount）IBM Z 系統，協助客戶在不犧牲效能或安全性的前提下，因應資料中心空間、成本與部署限制。

Client response has been strong, contributing to the highest mid-range revenue launch in IBM Z this quarter in nearly two decades.

客戶反應強勁，促成本季 IBM Z 近二十年來最高的中階產品營收上市表現。

Last quarter, we told you Lightwell represents a multibillion-dollar opportunity for IBM and Red Hat to help clients secure their open-source software.

上季我們曾提到，Lightwell 對 IBM 與 Red Hat 而言是一個數十億美元規模的機會，可協助客戶保護其開源軟體。

Since then, some of the world's largest enterprises have signed up as clients. And we have built a pipeline approaching $700 million in a short period of time.

自那之後，全球一些最大的企業已簽約成為客戶。我們也在短時間內建立了接近 7 億美元的商機管線（pipeline）。

We have released more than 40,000 remediated packages to the Lightwell network, demonstrating the rapid scale-up of our remediation capabilities.

我們已向 Lightwell 網路發布超過 40,000 個完成修補（remediated）的套件，展現我們修補能力的快速擴張。

Now, turning to our AI strategy, enterprise AI is entering its next phase. Deploying more models or building more agents will not be enough. The advantage will belong to a company that can operate AI securely, efficiently, and sustainably across the enterprise.

接著談到我們的 AI 策略，企業 AI 正進入下一個階段。僅僅部署更多模型或打造更多代理並不足夠。優勢將屬於能在整個企業範圍內安全、高效且可持續地運行 AI 的公司。

Organizations are recognizing that AI is only as effective as the system around it: the data that provides context, the workflows that orchestrate action, the infrastructure that secures and optimizes operations, and the hybrid cloud foundation that provides flexibility without locking.

各組織正意識到，AI 的效能取決於其周邊系統：提供情境的資料、協調行動的工作流程、確保並最佳化營運的基礎架構，以及在不造成綁定（lock-in）的前提下提供彈性的混合雲基礎。

IBM has assembled these capabilities across four platforms:

IBM 已在四個平台上整合了這些能力：

Confluent provides continuous business context through a real-time data foundation.

Confluent 透過即時資料基礎，提供持續的業務情境。

watsonx, including Bob and Orchestrate, support AI-driven software development and agent orchestration.

watsonx（包含 Bob 與 Orchestrate）支援 AI 驅動的軟體開發與代理協調（agent orchestration）。

HashiCorp and Apptio automate identity, access, and policy, while optimizing cost and performance.

HashiCorp 與 Apptio 自動化身分、存取與政策，同時最佳化成本與效能。

Red Hat provides the open-hybrid architecture, spanning public cloud, on-premises, sovereign environments, and the edge.

Red Hat 提供開放式混合架構，涵蓋公有雲、地端（on-premises）、主權環境（sovereign environments）以及邊緣端。

Together, these platforms help clients move from isolated deployments to AI operating at enterprise scale.

這些平台共同協助客戶從孤立的部署，邁向在企業規模下運行的 AI。

Data sits at the center of the AI operating model. 72% of organizations report insufficient real-time data infrastructure, which holds back AI at scale.

資料位於 AI 營運模型的核心。72% 的組織表示即時資料基礎架構不足，因而限制了 AI 的規模化。

Agents need the current state of the business. That is why we are excited about Streamhouse, a capability from Confluent that continuously captures, transforms, and queries real-time data. It complements warehouses and lakehouses.

代理需要掌握企業的即時狀態。因此我們對 Streamhouse 感到振奮；這是 Confluent 的一項能力，可持續擷取、轉換並查詢即時資料。它可與資料倉儲（warehouses）與資料湖倉（lakehouses）互補。

A Streamhouse makes real-time governed data continuously available to applications, analytics, and AI, unifying operational and analytical data into a single architecture.

Streamhouse 讓受治理的即時資料持續可供應用程式、分析與 AI 使用，將營運與分析資料整合為單一架構。

We believe Confluent is uniquely positioned with a complete open and hybrid approach that helps clients avoid lock-in, while deploying wherever the data resides.

我們相信 Confluent 以完整的開放與混合方法具備獨特優勢，能協助客戶避免被綁定，同時可在資料所在之處進行部署。

As AI scales, clients ask two questions: What will it cost? And who controls it? We are innovating on both.

隨著 AI 規模化，客戶會問兩個問題：成本會是多少？以及由誰來控制？我們在這兩方面都持續創新。

Apptio now connects AI spend to business results, providing visibility into AI usage, token consumption, and the cost of running it across the enterprise.

Apptio 現在可將 AI 支出與業務成果連結，提供對 AI 使用情況、token 消耗，以及在企業範圍內運行成本的可視性。

llm-d, an open-source project developed with Red Hat and Google, reuses previously processed context, helping clients run large AI models at 5 times to 10 times lower cost by making more efficient use of existing GPU infrastructure.

llm-d 是一個與 Red Hat 與 Google 共同開發的開源專案，可重用先前處理過的情境，透過更有效率地利用既有 GPU 基礎架構，協助客戶以低 5 倍至 10 倍的成本運行大型 AI 模型。

And Bob now runs on-prem so clients in regulated and sovereign environments can keep code and models on their own infrastructure.

而 Bob 現在可在地端（on-prem）運行，使受監管與主權環境中的客戶能將程式碼與模型保留在自有基礎架構上。

Governance matters as much as intelligence. As agents multiply, enterprises need to know who each agent is, what authority it has, and what it can access.

治理與智慧同等重要。隨著代理數量增加，企業需要知道每個代理是誰、擁有哪些權限，以及可存取哪些資源。

HashiCorp Vault allows clients to register, authorize, audit, and revoke agents and secure their credentials.

HashiCorp Vault 讓客戶能夠註冊、授權、稽核並撤銷代理，並保護其憑證。

IBM Agent Identity and HashiCorp Vault are partnering with the NVIDIA Open Agent Safety Platform to strengthen agent security. We believe Agent Identity will become as foundational to enterprise AI as identity has become to the Internet.

IBM Agent Identity 與 HashiCorp Vault 正與 NVIDIA Open Agent Safety Platform 合作，以強化代理安全。我們相信，Agent Identity 將如同身分識別之於網際網路一般，成為企業 AI 的基礎能力。

Finally, Quantum remains a strategic area of differentiation for IBM and a significant long-term opportunity.

最後，量子仍是 IBM 的策略性差異化領域，也是重要的長期機會。

In July, three independent demonstrations of quantum advantage were achieved on IBM hardware with the University of Chicago, Algorithmiq, and QEDMA, providing further evidence that quantum computing is moving from scientific promise towards practical value.

7 月，與芝加哥大學、Algorithmiq 與 QEDMA 合作，在 IBM 硬體上達成三項獨立的量子優勢展示，進一步證明量子運算正從科學承諾走向實際價值。

Our Nighthawk r2 processor has achieved important roadmap milestones across speed, scale, and quality, delivering 7,500 gigs, executing 100,000 circuits per second, and enabling performance improvements of up to 19 times over prior generation systems.

我們的 Nighthawk r2 處理器在速度、規模與品質方面達成重要的路線圖里程碑，提供 7,500 gigs、每秒執行 100,000 個電路，並使效能相較前一代系統最高提升達 19 倍。

That performance changes both time-to-solution and the economics of quantum computation. Workloads costing approximately $100 to run on IBM could cost up to $10 million on slower competitive modalities.

這樣的效能同時改變了解題時間（time-to-solution）與量子運算的經濟性。在 IBM 上執行成本約 100 美元的工作負載，在較慢的競爭替代方案上可能高達 1,000 萬美元。

Recently, IBM was selected for Stage C, the third phase of the Quantum Benchmarking Initiative led by DARPA, the United States' Defense Advanced Research Projects Agency, another validation of the strength of IBM's path to building the fault-tolerant quantum computer.

近期，IBM 入選由美國國防高等研究計畫署（DARPA）主導的 Quantum Benchmarking Initiative 之 Stage C（第三階段），再次驗證 IBM 在打造容錯量子電腦路徑上的實力。

Commercial momentum is also growing. This year, IBM Quantum has signed 16 enterprise engagements across industries including automotive, aerospace, telecommunications, financial services, and manufacturing.

商業動能也在增強。今年，IBM Quantum 已在汽車、航太、電信、金融服務與製造等產業簽下 16 項企業合作案。

This includes AT&T and General Motors, amongst many others who are now all part of the IBM Quantum Network, the world's largest network of quantum innovators.

其中包括 AT&T 與通用汽車等多家企業，現皆已加入 IBM Quantum Network——全球最大的量子創新者網路。

We are also advancing the technology and infrastructure required for industrial-scale fault-tolerant computing, including the successful connection of our first modular cryogenic systems and continued progress on our 300,000-square-foot Poughkeepsie manufacturing facility.

我們也在推進工業規模容錯運算所需的技術與基礎設施，包括成功連接我們第一套模組化低溫系統，並持續推動我們位於 Poughkeepsie、面積 30 萬平方英尺的製造設施進展。

This, along with the closing of the $1 billion award from the Department of Commerce in Anderon for manufacturing quantum wafers, gives us confidence that, after the stalling demonstration, we are ready to scale.

此外，商務部在 Anderon 提供的 10 億美元獎助（用於製造量子晶圓）已完成結案，這些進展讓我們有信心：在停滯的展示之後，我們已準備好擴大規模。

We look forward to talking more about our progress at our Quantum Day on December 10.

我們期待在 12 月 10 日的 Quantum Day 上，進一步分享我們的進展。

In closing, our third-quarter performance reflects the strength of our portfolio and the progress we are making against our strategic priorities of delivering both sustainable revenue growth and higher profitability.

總結而言，我們第三季的表現反映了產品組合的實力，以及我們在策略優先事項上的推進進度——同時實現可持續的營收成長與更高的獲利能力。

Our results reinforce our confidence in delivering revenue growth for 2026 in a range of 4% to 5% and enable us to raise our year-over-year free cash flow growth expectation to more than $1 billion.

我們的結果強化了我們對 2026 年營收成長 4% 至 5% 區間的信心，並使我們得以將年度自由現金流年增的預期上調至超過 10 億美元。

With that, let me hand it over to Jim.

接下來，我把時間交給 Jim。

Thanks, Arvind.

謝謝你，Arvind。

In the third quarter, we delivered solid results, with 6% revenue growth, 70 basis points of operating pre-tax margin expansion, and 11% diluted operating earnings per share growth. And, through the first nine months, we generated about $8 billion of free cash flow, up 8% year over year.

第三季，我們交出穩健的成果：營收成長 6%，營業稅前利潤率擴張 70 個基點，稀釋後營業每股盈餘成長 11%。此外，前九個月我們產生約 80 億美元自由現金流，年增 8%。

These results reflect strength across our portfolio and solid execution against our priorities. Let me walk through the performance in each of our segments:

這些結果反映了我們產品組合的全面強勁，以及對優先事項的扎實執行。我將逐一說明各事業部門的表現：

Software revenue growth accelerated 4 points sequentially to 9% this quarter, including over 4 points of organic growth. This performance was driven by our high-value, annual, recurring revenue base, which grew 8% to $24.9 billion.

本季軟體營收成長較前一季加速 4 個百分點至 9%，其中有超過 4 個百分點來自有機成長。此表現主要由我們高價值的年度經常性收入（ARR）基礎所帶動，該基礎成長 8% 至 249 億美元。

3 of our 4 Software sub-segments grew revenue double-digits, led by the combination of strong new business performance and pipeline-closure rates, including the successful closure of around two-thirds of the large deals that slipped from the second quarter.

我們 4 個軟體子事業中有 3 個營收呈現雙位數成長，主要受惠於強勁的新業務表現與管線結案率，包括成功結案約三分之二從第二季遞延的大型交易。

Our results represent the high end of the range we outlined in July, reinforcing our confidence in the strength of our portfolio.

我們的結果落在 7 月所提出區間的高端，進一步強化我們對產品組合實力的信心。

Turning to our sub-segments, Red Hat revenue grew 11%, again, this quarter, fueled by bookings accelerating back to the high-teens.

接著看子事業部門，Red Hat 本季營收再次成長 11%，主要受惠於訂單（bookings）成長加速回到高十位數。

Lightwell was recently made generally available. And we have strong early demand, with a pipeline that is already approaching $700 million.

Lightwell 近期已正式全面上市（generally available）。我們在早期就看到強勁需求，管線金額已接近 7 億美元。

OpenShift is now a $2.4 billion ARR business, with growth accelerating to more than 30%.

OpenShift 現已成為 ARR 達 24 億美元的業務，成長加速至超過 30%。

Our virtualization offering continues to scale, with nearly $800 million in contracts signed since early 2024 and about $750 million in pipeline.

我們的虛擬化產品持續擴張，自 2024 年初以來已簽署近 8 億美元合約，且管線約 7.5 億美元。

Data grew 20%, driven by another strong performance from Confluent, with bookings growth of about 20%.

Data 營收成長 20%，主要由 Confluent 再次交出強勁表現所帶動，訂單成長約 20%。

Transaction processing grew 10% in the quarter, driven by demand for this mission-critical software and strong execution against our pipeline.

交易處理（Transaction processing）本季成長 10%，主要受此關鍵任務軟體需求帶動，並且我們對管線的執行表現強勁。

HashiCorp continues to perform well, growing bookings across all geographies and accelerating revenue growth sequentially. However, automation was impacted by a tougher compare.

HashiCorp 持續表現良好，在所有地區的訂單皆成長，且營收成長率較前一季加速。然而，自動化業務受到較高比較基期影響。

Turning to Infrastructure, revenue growth accelerated to 5%, driven by 32% growth in distributed infrastructure.

再看基礎設施（Infrastructure），營收成長加速至 5%，主要由分散式基礎設施成長 32% 所帶動。

Storage delivered record revenue growth and continued to gain market share, with broad-based demand across Flash, Fusion, Tape, and IBM Z storage.

儲存業務創下營收成長新高並持續提升市占率，Flash、Fusion、Tape 與 IBM Z 儲存皆呈現廣泛需求。

Strength across storage and power reflects the growing importance of differentiated infrastructure, as clients modernize their environments, address data growth, and support AI at scale.

儲存與電力（power）的全面強勁，反映出在客戶現代化其環境、因應資料成長並支援規模化 AI 的過程中，差異化基礎設施的重要性日益提升。

With the sizable backlog in distributed infrastructure exiting the quarter, we are well-positioned for continued momentum in the fourth quarter.

在本季結束時，分散式基礎設施仍有可觀的積壓訂單（backlog），使我們在第四季具備延續動能的良好條件。

z17 continues to perform exceptionally well, wrapping on the strongest start to a Z program last year. Program-to-date performance remains well ahead of the prior generation, up over $1 billion. And this quarter, we had the strongest mid-range launch in IBM Z in nearly two decades.

z17 持續表現非常出色，延續去年 Z 計畫史上最強勁的開局。截至目前的計畫表現仍大幅領先前一代，增加超過 10 億美元。而在本季，我們也完成了近二十年來 IBM Z 最強勁的中階機型上市。

Beyond the strength of the program, IBM mainframe clients, representing approximately 85% of installed MIPS capacity, are maintaining or growing their capacity. And aggregated install capacity has tripled over the last few programs.

除了計畫本身的強勁之外，IBM 大型主機客戶（約占已安裝 MIPS 容量的 85%）正維持或提升其容量。而彙總後的已安裝容量在過去幾個計畫中已成長至三倍。

Consulting revenue grew 2% this quarter, reflecting continued client demand for AI-enabled transformation. Growth was led by strategy and technology, up 4%, with strength in application modernization and data and AI transformation, while cybersecurity delivered nearly double-digit growth in intelligent operations.

本季顧問（Consulting）營收成長 2%，反映客戶對 AI 賦能轉型的需求持續。成長主要由策略與科技帶動、成長 4%，其中應用程式現代化以及資料與 AI 轉型表現強勁；同時，資安在智慧營運方面也交出接近雙位數的成長。

AI is increasingly embedded in broader transformations our clients are undertaking, representing about 30% of our Consulting revenue in the quarter and 35% of our backlog.

AI 正日益嵌入客戶所推動的更廣泛轉型之中，本季約占我們顧問營收的 30%，並占我們積壓訂單的 35%。

We are also expanding the ecosystem of capabilities we bring to clients, including through our new strategic partnerships with OpenAI and our expanded relationship with Google Cloud, as we help clients scale enterprise AI.

我們也在擴大帶給客戶的能力生態系，包括與 OpenAI 的新策略合作夥伴關係，以及與 Google Cloud 擴大的合作關係，以協助客戶擴展企業級 AI。

Clients remain at different stages of adoption. But we continue to see them modernizing applications, strengthening their data foundations, and re-engineering workflows, as they move forward toward embedding AI into core business processes at scale.

客戶仍處於不同的採用階段。但我們持續看到他們在現代化應用程式、強化資料基礎，以及重新設計工作流程，並持續推進，朝向在規模化情境下將 AI 嵌入核心業務流程。

Our integrated model, combining deep domain expertise, technology, and reusable assets, positions us to help clients capture business value from AI, while accelerating time-to-value.

我們的整合式模式結合深厚的產業領域專業、技術與可重複使用資產，使我們能協助客戶從 AI 中取得商業價值，同時加速價值實現時間（time-to-value）。

Let me now discuss profitability:

接下來我談談獲利能力：

Adjusted EBITDA and operating pre-tax margins expanded by 60 basis points and 70 basis points, respectively. As I mentioned last quarter, we accelerated our productivity initiatives to help enhance our operating profit margin and free cash flow commitments, while continuing to invest for growth.

調整後 EBITDA 與營業稅前利潤率分別擴張 60 個基點與 70 個基點。如同我上季提到的，我們加速推動生產力提升計畫，以強化營業利潤率與自由現金流承諾，同時持續投資成長。

We made significant progress this quarter, driving efficiencies across G&A and sales, reinforcing the power of our productivity flywheel.

本季我們取得了顯著進展，推動整體 G&A 與銷售端的效率提升，進一步強化我們生產力飛輪的威力。

Segment -profit margins expanded by 210 basis points in Software and 120 basis points in Infrastructure, driven by ongoing productivity actions.

在持續的生產力行動帶動下，分部—利潤率在軟體業務擴大 210 個基點、在基礎設施業務擴大 120 個基點。

Consulting segment profit margin declined modestly, as we reinvested productivity gains into digital assets, accelerating our transition toward a Services-as-Software model to capture the growing AI opportunity.

顧問服務分部利潤率小幅下滑，因為我們將生產力收益再投資於數位資產，加速向「服務即軟體（Services-as-Software）」模式轉型，以掌握日益成長的 AI 機會。

Through the first nine months of the year, we generated about $8 billion of free cash flow, up about $600 million year over year, driven by about $1 billion increase in adjusted EBITDA, offset by higher CapEx, cash taxes, and net interest expense.

今年前九個月，我們產生約 80 億美元的自由現金流，年增約 6 億美元；主要由調整後 EBITDA 約增加 10 億美元所帶動，但部分被較高的資本支出（CapEx）、現金稅負與淨利息費用所抵銷。

Our free cash flow margin for the first nine months of the year is the highest in reported history, highlighting our strong execution against productivity initiatives and laser focus on operational discipline.

今年前九個月的自由現金流利潤率為公司揭露史上最高，凸顯我們在生產力計畫上的強勁執行力，以及對營運紀律的高度聚焦。

We exited the third quarter with a strong liquidity position and a solid investment-grade balance sheet, with cash of $11.1 billion.

第三季結束時，我們的流動性部位強勁，且投資等級資產負債表穩健，現金為 111 億美元。

We returned $4.8 billion to shareholders in the form of dividends through the first nine months of the year. Our debt balance ending the quarter was $63.8 billion, including debt of $12.7 billion for our Financing business, with a receivables portfolio that is almost 80% investment grade.

今年前九個月，我們以股利形式向股東回饋 48 億美元。本季期末負債餘額為 638 億美元，其中包含融資業務的 127 億美元負債；其應收帳款組合中近 80% 為投資等級。

Now, let me talk about what we are seeing, going forward:

接下來，讓我談談我們對未來的觀察：

The strength and diversity of our portfolio, disciplined capital allocation, and sustained productivity execution underpin our revenue growth, earnings expansion, and free cash flow performance.

我們投資組合的強度與多元性、嚴謹的資本配置，以及持續落實的生產力執行，支撐了我們的營收成長、獲利擴張與自由現金流表現。

Through the first nine months of the year, we delivered over 4% revenue growth, 9% operating pre-tax income growth, 10% operating earnings per share growth, and 8% free cash flow growth.

今年前九個月，我們實現超過 4% 的營收成長、9% 的營運稅前收入成長、10% 的營運每股盈餘成長，以及 8% 的自由現金流成長。

We continue to expect revenue growth for 2026 in the range of 4% to 5% and are raising our expectation for free cash flow growth to more than $1 billion.

我們仍預期 2026 年營收成長落在 4% 至 5% 區間，並將自由現金流成長預期上調至超過 10 億美元。

90 days ago, we provided a full-year revenue growth range, with profit and free cash flow expectations anchored to a 4% base case. We continue to take the same approach for the full year, anchoring profit, free cash flow, and segment expectations to that base case.

90 天前，我們提供了全年營收成長區間，並以 4% 的基準情境作為獲利與自由現金流預期的錨定。我們全年仍採取相同做法，將獲利、自由現金流與各分部預期錨定在該基準情境。

Given our strong third-quarter Software performance, our confidence in the full-year outlook has increased. We now expect full-year Software growth of approximately 7% to 8%, with about 7% as the revised base case. For the full year, we expect double-digit growth in hybrid cloud, mid-teens growth in data, and low single-digit growth in automation.

鑑於第三季軟體業務表現強勁，我們對全年展望的信心提升。我們目前預期全年軟體成長約 7% 至 8%，並將約 7% 作為修訂後的基準情境。就全年而言，我們預期混合雲將呈現兩位數成長、資料業務中十位數（mid-teens）成長，以及自動化業務低個位數成長。

For transaction processing, we are raising our expectations for the full year, with growth expected to be flat.

在交易處理方面，我們上調全年預期，預計成長將持平。

Infrastructure growth accelerated in the third quarter, with significantly stronger program-to-program performance of z17, underscoring the durability of the platform and the value it delivers to clients.

基礎設施成長在第三季加速，z17 在各計畫間（program-to-program）的表現顯著更強，凸顯該平台的持久性以及其為客戶帶來的價值。

Combined with continued strength in distributed infrastructure, we continue to expect low-single-digit Infrastructure revenue growth for the full year.

再加上分散式基礎設施持續強勁，我們仍預期全年基礎設施營收將呈現低個位數成長。

In Consulting, backlog quality and momentum in generative AI continue to support a modest acceleration to low- to mid-single-digit revenue growth for the year.

在顧問服務方面，積壓訂單（backlog）品質與生成式 AI 的動能持續支撐全年營收成長小幅加速至低至中個位數。

In the third quarter, our operational discipline and accelerated productivity initiatives allowed us to enhance margins and deliver free cash flow commitments, while continuing to invest for growth.

第三季，我們的營運紀律與加速的生產力計畫，使我們在持續投資成長的同時，得以提升利潤率並達成自由現金流承諾。

These actions include deploying AI and automation at scale across our company, reducing third-party spend, improving sales and marketing efficiency, using AI to drive more efficient software development, optimizing our supply chain, and enhancing services delivery.

這些行動包括在全公司大規模部署 AI 與自動化、降低第三方支出、提升銷售與行銷效率、運用 AI 推動更高效率的軟體開發、優化供應鏈，以及強化服務交付。

We continue to expect 100 basis points of operating pre-tax margin expansion for the full year, reflecting the flow-through from the third quarter earnings beat and continued operational discipline. And our operating tax-rate expectation for the year remains in the mid-teens.

我們仍預期全年營運稅前利潤率擴張 100 個基點，反映第三季獲利優於預期所帶來的傳導效應，以及持續的營運紀律。此外，我們對全年營運稅率的預期仍維持在十位數中段（mid-teens）。

For the full year, we are raising our expectation for free cash flow growth to more than $1 billion. The primary driver of this growth will be adjusted EBITDA, offset by higher CapEx, higher cash tax, and net interest expense.

就全年而言，我們將自由現金流成長預期上調至超過 10 億美元。此成長的主要驅動因素將是調整後 EBITDA，但會被較高的資本支出、較高的現金稅負與淨利息費用所抵銷。

Our free cash flow continues to be driven by growth in adjusted EBITDA and demonstrates consistent free cash flow realization of about 130% over the last few years.

我們的自由現金流仍由調整後 EBITDA 的成長所驅動，並展現過去幾年約 130% 的穩定自由現金流實現率。

Let me close by saying that our third-quarter performance reinforces our confidence in our portfolio, the resilience of our business model, and our outlook for the year.

最後我想說，我們第三季的表現強化了我們對投資組合、商業模式韌性以及全年展望的信心。

At the same time, we continue to invest in the technologies that will shape IBM's next chapter of growth. As Arvind mentioned, Quantum continues to advance across technology, ecosystem adoption, and scale-up.

同時，我們也持續投資於將塑造 IBM 下一個成長篇章的技術。如 Arvind 所提到的，量子（Quantum）在技術、生態系採用與規模化方面持續推進。

While still early, the progress we're seeing from enterprise engagements to government validation reinforces our confidence in both our roadmap and the long-term opportunity ahead. We look forward to discussing in more detail at our Quantum Day in December.

雖然仍在早期階段，但我們從企業合作到政府端驗證所看到的進展，強化了我們對路線圖以及長期機會的信心。我們期待在 12 月的 Quantum Day 上更深入地討論。

Arvind and I are now happy to take questions. Olympia, let's get started.

Arvind 和我現在很樂意開始回答問題。Olympia，我們開始吧。

Thank you, Jim. Before we begin Q&A, I'd like to mention a couple of items:

謝謝你，Jim。在開始問答之前，我想先提幾點：

First, supplemental information is provided at the end of the presentation.

第一，補充資訊已在簡報末尾提供。

And then, second, as always, I'd ask you to refrain from multi-part questions.

第二，一如往常，我想請各位避免提出多段式問題。

Operator, let's please open it up for questions.

接線員，請開放提問。

Thank you. At this time, we'll begin the question-and-answer session of the conference.

謝謝。現在我們將開始本次會議的問答環節。

(Operator Instructions)

（接線員指示）

Amit Daryanani, Evercore.

Amit Daryanani，Evercore。

Thanks a lot. Good afternoon, everyone. And great to see some nice Software exploration here.

非常感謝。各位下午好。也很高興看到軟體這邊有一些不錯的拓展。

Arvind, I think, last quarter, we discussed a lot about customers prioritizing Infrastructure over Software purchases. This quarter, it seems like transaction processing is at double-digit growth, while Infrastructure has remained very strong for you folks.

Arvind，我想上季我們談了很多客戶在支出優先順序上更偏向基礎設施而非軟體採購。這一季看起來交易處理呈現兩位數成長，而你們的基礎設施也依然非常強勁。

Can you just talk about what changed in customer-spending priorities? Are you primarily just recovering before transactions at this point or are you actually starting to see Infrastructure investments starting to create more incremental demand for IBM Software?

你能否談談客戶支出優先順序有哪些改變？目前主要只是交易在復甦，還是你們其實開始看到基礎設施投資正為 IBM 軟體帶來更多增量需求？

And with Red Hat bookings actually growing at high-teens, is that an only indication that Software can actually accelerate, as you go into '27? Thank you.

另外，Red Hat 的訂單（bookings）實際上以十位數高段（high-teens）成長，這是否是軟體在邁向 2027 年時能夠加速的另一個指標？謝謝。

So, Amit, thanks for that question. And, also, thank you for observing the numbers.

Amit，謝謝你的問題。也謝謝你留意到這些數字。

Let me comment on the overall customer priorities, the tie towards Infrastructure and Red Hat. And I'm going to ask Jim to cover the transaction-processing piece a little bit.

我先評論一下整體客戶優先順序、與基礎設施及 Red Hat 的連動；交易處理的部分我會請 Jim 再補充說明。

So, first, as we look at customer-spending priorities, customer-spending priorities in trying to leverage AI remains strong. But, as you begin to leverage it, how does the average customer get value?

首先，就客戶支出優先順序而言，客戶希望運用 AI 的力道仍然強勁。但當你開始運用它時，一般客戶要如何取得價值？

They have to rewrite an application. They have to modernize an existing application. And you, then, ask the question: Where do those run?

他們必須重寫應用程式。他們必須將既有應用程式現代化。接著你會問一個問題：這些要在哪裡運行？

That tends to be a tailwind for Red Hat OpenShift. And I think we are at the very beginning of the cycle so I expect that to carry on for a long time.

這往往會成為 Red Hat OpenShift 的順風因素。而且我認為我們正處於這個循環的非常初期，因此我預期這股趨勢會持續很長一段時間。

Then, the moment they get to saying, okay, I'll write the application on Red Hat OpenShift, you then get to: How do I unlock the data from all of the enterprise data that I need to power that AI application?

接著，當他們一說「好，我要在 Red Hat OpenShift 上撰寫這個應用程式」的那一刻，你接下來就會面臨：我該如何從我需要用來驅動該 AI 應用程式的所有企業資料中，把資料解鎖出來？

Remember: 90% of enterprise data is not on the cloud. It's very much inside the enterprise. So Confluent tends to be a tailwind.

請記住：90% 的企業資料不在雲端。它很大程度上是在企業內部。因此 Confluent 往往會成為順風因素。

I expect these to carry on for a long time. The automation portfolio will get a benefit. When these applications all get to scale and people start worrying about managing hundreds and thousands of them, at that point, automation will also get a tailwind.

我預期這些趨勢會持續很長一段時間。自動化產品組合將會受惠。當這些應用程式都擴展到規模、而人們開始擔心要管理數百、數千個應用程式時，那時自動化也會迎來順風。

On the point you're making on Red Hat, look, we are very comfortable saying that, based on what we can see, we are comfortable about double-digit growth in the fourth quarter of this year.

就你在 Red Hat 上提出的觀點來看，我們很坦然地說，根據我們所看到的情況，我們對今年第四季的雙位數成長感到有把握。

Right now, we don't really want to comment on '27 growth. But, generally speaking, I can say, when we get these kinds of booking growth, it crosses into good revenue growth, going forward.

目前我們不太想評論 2027 年的成長。但一般而言我可以說，當我們有這種訂單成長時，往後就會轉化為良好的營收成長。

Jim, maybe a comment on [TP]?

Jim，要不要對 [TP] 補充一下？

Yeah. Just to build on your points, Arvind -- and Amit, thank you very much for the comment and the question -- we're very pleased, overall, with our Software performance -- team executed extremely well.

好的。Arvind，我就延伸你剛才的重點——Amit，也非常感謝你的評論與提問——我們整體而言對軟體業務表現非常滿意，團隊執行得非常到位。

4 points sequentially to 9% printed here in third quarter. And it was all organically driven. Pervasively across our portfolio, 3 of our 4 sub-categories accelerated and grew double-digits, overall. And that was a reflection of both.

第三季這裡印出來是 9%，較前一季提升 4 個百分點。而且完全是有機成長所驅動。在我們的產品組合中普遍可見，四個子類別中有三個加速並呈現雙位數成長。而這反映了兩個面向。

We talked a lot last quarter about the 80% high-value recurring revenue growing nicely at high single-digit. But we saw a very nice rebound in our transactional business -- that remaining 20% that grew, actually, double-digits, led by our ELAs that were up north of 20%, with an NRR of north of 130%, overall.

上季我們談了很多，80% 的高價值經常性收入以高個位數穩健成長。但我們也看到交易型業務非常明顯的回升——剩下那 20% 的部分實際上也達到雙位數成長，主要由我們的 ELA 帶動，ELA 成長超過 20%，整體 NRR 超過 130%。

That's a function of, yes, disciplined execution on slip deals but also a function of net new business contribution in the quarter.

這一方面是因為我們對延後成交（slip）的案子執行得很有紀律，另一方面也來自本季淨新增業務的貢獻。

And I would tell you before all of you ask: about a quarter of it is the slip deals with regards to overall Software acceleration and about three-quarters of it is net new business that we are capturing, overall, so to build on Arvind's point.

在你們大家問之前我先說：就整體軟體加速而言，大約四分之一來自這些延後成交的案子，約四分之三來自我們整體所取得的淨新增業務，以呼應 Arvind 的觀點。

Now, on TP: TP, we're obviously pleased -- double-digit growth. I think this instantiates the demand for our mission-critical, strategic mainframe software. A lot of discussions a quarter ago about demand destruction versus timing. I think this puts this to rest right now.

接著談 TP：我們當然很高興——雙位數成長。我認為這具體印證了市場對我們關鍵任務、策略性主機（mainframe）軟體的需求。一季前有很多討論在於需求被破壞（demand destruction）還是只是時點問題。我認為現在這點已經可以告一段落。

Because when you take a look at it, we've been talking all year long about the reinforcement of confidence we have in our mainframe platform and how we monetize the strong z17 performance, right?

因為你仔細看，我們整年都在談我們對主機平台信心的強化，以及我們如何將強勁的 z17 表現變現，對吧？

That model is -- the monetization model -- the more capacity you put in the marketplace, the higher monetization opportunity, the higher price opportunity, the higher value-creation opportunity.

那個模式——也就是變現模式——是：你在市場上投放的容量越多，變現機會越大、定價機會越高、價值創造機會也越高。

z17, today, install MIPS are up, high-teens, six quarters in. [Shipped] MIPS are up over 100%. That is the future monetization opportunity that we feel extremely confident about as to the acceleration potential of TP in the future.

截至目前，z17 安裝的 MIPS 在推出六個季度後仍維持高十幾%的成長。[出貨] MIPS 成長超過 100%。這就是我們對未來 TP 加速潛力極具信心的變現機會。

So appreciate the question and great place to start.

所以謝謝你的提問，這也是一個很好的切入點。

Operator, let's take the next question.

接線員，我們來接下一題。

Wamsi Mohan, Bank of America.

美國銀行的 Wamsi Mohan。

Yeah. Thank you so much.

好的，非常感謝。

Thanks for the color you just shared on TP. But, maybe, if we could just double-click on that:

謝謝你們剛才分享的 TP 相關細節。不過也許我們可以再更深入一點：

Last quarter you were thinking that TP might be down low- to mid-single-digits for the second half. And you called it a 27% growth vector. It grew 10%, instead.

上季你們原本認為 TP 在下半年可能會下滑低到中個位數。而你們也稱它是 27% 的成長向量。但結果它反而成長了 10%。

So, maybe, Jim, as part of those -- like, a quarter of Software upside really came from those push deals. Could you share some color, more specifically, with respect to TP, if that was similar or higher in relation to the deals that closed from those push deals?

所以，Jim，作為其中一部分——例如軟體的上行有約四分之一來自那些「推進成交」的案子。你能否更具體分享一些 TP 方面的細節：TP 是否也有類似比例或更高比例，是來自那些推進成交、最終在本季結案的交易？

And how many of those [low-10s] of slip deals have now closed because you've not really changed your 4% to 5% for the year but your commentary and your numbers would suggest that you're gravitating much more towards the higher end of that? If you could share any more incremental color.

另外，那些（大約低十幾%）的延後成交案子中，有多少現在已經結案？因為你們全年指引仍維持 4% 到 5% 不變，但你們的評論與數字看起來顯示你們更偏向區間高端。如果可以的話，請再多提供一些增量的細節。

And on two-cycle TP performance, is that still something that we should think of as mid-single digit? That would be great. Thank you so much.

最後，關於 TP 兩個週期（two-cycle）的表現，是否仍應視為中個位數？那就太好了。非常感謝。

Wamsi, I know you addressed the question to Jim but I'm going to start and then, ask Jim to color -- give a lot of color on both the slip deals, the TP performance, and the cycle-to-cycle compare, if I paraphrase your question short.

Wamsi，我知道你把問題指向 Jim，但我先開頭，然後再請 Jim 就延後成交案子、TP 表現，以及週期對週期的比較——如果我把你的問題簡化成這三點——提供更多細節。

So let's just talk about TP. The questions everybody had last quarter was: Does this imply that overall demand for mainframe has gone down?

我們先談 TP。上季大家的疑問是：這是否意味著主機整體需求下滑了？

And we were very confident that this was a timing question, not a destruction question. And so, at the macro level, as people deploy all of that hardware, as they deploy that MIPS, as the volumes of the underlying transactions increase, they need this mainframe software to support that mission-critical business. That's why we were confident it would come back.

而我們非常有信心這是時點問題，不是需求被破壞。因此在宏觀層面，當人們部署那些硬體、部署那些 MIPS、底層交易量增加時，他們就需要這些主機軟體來支撐關鍵任務業務。這就是我們有信心它會回來的原因。

Now, to your point of us saying low- to mid-single-digit decline, we want to be cautious and we want to be prudent. We want to see the absolute baseline. And we don't want to call above it.

至於你提到我們說可能會下滑低到中個位數，我們希望保持謹慎、保持審慎。我們想先看到絕對的基準線。我們不想在基準線之上先行喊得太高。

I think what really happened was, we got not only the deferral from 2Q, we got a normal 3Q buying cycle on mainframe so got it both, together.

我認為實際發生的是：我們不僅拿回了第二季遞延的部分，第三季也出現了主機正常的採購週期，所以兩者同時發生。

I would be candid and tell you I don't expect double-digit performance in TP to keep going. But I would expect to see, now, more normalized performance, especially going into next year and, perhaps, moderation from 3Q going into the fourth quarter.

我坦白說，我不預期 TP 的雙位數表現會持續下去。但我預期現在會看到更正常化的表現，尤其進入明年；而且從第三季進入第四季，可能會有所降溫。

But it speaks to the underlying demand for the mainframe around resilience, around the economics, around the architecture that supports [8], [9s] of availability.

但這也反映了主機在韌性、經濟性，以及支援 [8]、[9s] 可用性的架構等方面的底層需求。

Let me point that out: 8, 9s of availability but the transactions in the mainframe tend to be almost the opposite of what happens in AI. Here, you need things that are completely deterministic, that are repeatable, and that you can look at it with a financial cold eye that they're meeting all the requirements of those kinds of transactions.

我補充說明一下：8、9 個 9 的可用性；但主機上的交易往往幾乎與 AI 發生的情況相反。在這裡，你需要的是完全可決定（deterministic）、可重複的東西，而且你可以用冷靜的財務視角去檢視它們是否滿足這類交易的所有要求。

And there's plenty of other things that will happen on distributed infrastructure. I think, in terms of the numerics, I'm going to hand it to Jim to cover with you.

而在分散式基礎架構上，還會發生很多其他事情。至於數字面的部分，我把時間交給 Jim 來跟你說明。

Yeah. Wamsi, thanks for the question, overall. Let me just reiterate a little bit about where I ended the last question to Amit.

好的。Wamsi，謝謝你的提問。我先稍微重申一下我在上一題對 Amit 結尾時提到的內容。

When you look at the health of our Software business that accelerated 4 points quarter to quarter, it was a function of both the high-value recurring revenue -- and, again, the underpinnings of that is the Red Hat strength and acceleration.

當你看我們軟體業務的健康狀況，本季較上季加速了4個百分點，這同時來自高價值的經常性收入——而且再次強調，其根基在於 Red Hat 的強勁表現與加速成長。

It's excellent start to our Confluent acquisition. Its record performances in Hashi, overall. And that part of the business is doing well.

這是我們收購 Confluent 的一個極佳開端。整體而言，Hashi 也創下了紀錄性的表現。而這一部分的業務表現良好。

The transactional is where the big difference changed. Last quarter, it impacted our Software growth by 4 points to 5 points, when we were down double digit, transactional.

差異最大的變化出現在交易型收入。上季在交易型收入呈現兩位數下滑的情況下，它使我們的軟體成長率受到4到5個百分點的影響。

This quarter, we were up double digit, transactional. And I'll remind you: That's 20% of our Software book -- annual books, about $32 billion, give or take.

本季交易型收入則是兩位數成長。我也提醒各位：交易型收入約占我們軟體年度簿記（annual books）的20%——大約是320億美元上下。

And that, today, offers about -- two-thirds to three-quarters of that is ELAs. ELAs are strong incumbency positions that we've earned trust of our clients for decades around mainframe running their core architectures.

而在這之中，今天大約有三分之二到四分之三是 ELA（企業授權合約）。ELA 是我們在大型主機（mainframe）領域長期建立的強勢既有客戶地位；數十年來，客戶信任我們在大型主機上運行其核心架構。

On top of that, we build a moat and we build a software-distributed business model that captures that very attractive economic multiplier of $3 to $5 of software for every dollar we land on mainframe.

在此之上，我們建立了護城河，並打造了以軟體分銷為核心的商業模式，能捕捉到非常具吸引力的經濟倍數效應：我們在大型主機上每落地1美元，就能帶動3到5美元的軟體收入。

Now, within that, TP is disproportionate. But, with that moat, we give clients flexibility on where they want to choose that ELA spend as to where it goes between data, between automation, and between TP.

在這裡面，TP 的占比不成比例地高。但有了這道護城河，我們也讓客戶在 ELA 支出上有彈性，可在資料、自动化與 TP 之間選擇資金配置方向。

Red Hat's not part of our ELAs, overall. And what we saw here, both, is a contribution in the quarter on slip deals.

整體而言，Red Hat 並不包含在我們的 ELA 之中。而我們在這裡看到的，還包括本季對於「延後成交（slip）」案子的貢獻。

Let's put this in perspective: We printed a number, what, $7.6 billion of Software. The slip deals were tens of slip deals so let's put that in a ballpark from 2Q of what, $400 million, which is what we missed Software by. We closed two-thirds of those. That's $250-plus million.

讓我們把這放在脈絡中看：我們公布的軟體營收是76億美元。延後成交的案子是數十筆；以第二季的量級來估，大約是4億美元——也就是我們軟體未達標的金額。我們已經結案了其中三分之二。也就是2.5億美元以上。

That is the 25% of the contribution and the growth that we've had of 4 points. By the way, that was disproportionate to TP. That's why you see TP well up at double digits, overall. We didn't get a lot out of that in data and in automation. And some of that should be fourth quarter.

這大約占了我們成長（加速4個百分點）貢獻的25%。順帶一提，這對 TP 的貢獻不成比例地高；因此你會看到 TP 整體呈現兩位數成長。在資料與自動化方面，我們從這些案子得到的貢獻不多。其中一部分應該會落在第四季。

Now, fourth quarter, we gave a guide, I think, to use Arvind's word, anchored in our philosophy 90 days ago: prudent, cautious. We're anchoring this thing at the low end, approximately 7%.

至於第四季，我們給出的指引——借用 Arvind 的用詞——是以90天前我們的理念為錨：審慎、保守。我們把這件事錨定在區間低端，大約7%。

By the way, put that in perspective, I think the current Street estimates for the year, about $32 billion of revenue, we feel comfortable with that $32 billion of revenue. So that's where we're at.

順帶一提，把這放在脈絡中：我認為目前華爾街對全年營收的共識預估約為320億美元，我們對這320億美元的營收目標感到有把握。所以我們目前就是這樣的狀態。

I think I covered most of your questions.

我想我已經回答了你大部分的問題。

Great. Operator, next question.

很好。接線員，下一題。

Ben Reitzes, Melius Research.

Ben Reitzes，Melius Research。

Hey, guys. Thanks. Appreciate the question.

嗨，各位。謝謝。感謝讓我提問。

I wanted to ask about your long-term -- the short-term puts-and-takes we get. But, at your latest Analyst Day, you laid out that you would approach double digits in Software as a long-term target.

我想問的是你們的長期目標——短期的此消彼長我們理解。但在你們最近一次分析師日，你們提出軟體長期目標是要接近兩位數成長。

Based on what you're seeing, do you still feel confident in that? And how is M&A going to play a role in terms of achieving that?

基於你們目前看到的情況，你們仍然對此有信心嗎？另外，併購（M&A）在達成這個目標上會扮演什麼角色？

Because in order to hit it, you got to accelerate 1 point or 2 points after this year. So a lot of folks are wondering where M&A plays in that. Thanks a lot.

因為要達標的話，今年之後你們必須再加速1到2個百分點。所以很多人都在想，M&A 在其中會如何發揮作用。非常感謝。

Thanks, Ben. I appreciate the question, overall.

謝謝你，Ben。我整體上很感謝這個問題。

Obviously -- and it's not going to surprise you -- we'll spend a lot more time on this question here in January. We've got the team extremely focused already. We're 21 days into, arguably, the most important quarter we have in a fourth quarter, overall.

很明顯——這也不會讓你意外——我們會在一月花更多時間討論這個問題。我們的團隊已經高度聚焦。我們進入第四季才21天，而第四季可以說是我們最重要的一季。

But let me take a step back, first, around IBM and then, as Arvind and I have been talking here, as we're heading into our planning cycle and laying out all of our investment and capital-structure strategies for '27. Talk a little bit about IBM, then hit your question on Software.

不過我先退一步，先談 IBM 的整體情況；接著如同 Arvind 和我剛才所談，當我們進入規劃週期、並為2027年擬定所有投資與資本結構策略時，再回到你關於軟體的問題。先談 IBM，再回答你對軟體的提問。

We've got a lot of confidence in our strategy. And our portfolio is well positioned to deliver sustainable, durable growth on our model. We feel extremely confident in that.

我們對自身策略非常有信心。而我們的產品組合也處於有利位置，能在我們的商業模式下交付可持續、具韌性的成長。我們對此極具信心。

Why? We got an attractive TAM; market backdrop of technology; the diversity of our portfolio, leveraging our hybrid cloud and AI-platform-centric model, our continued portfolio optimization to shift the higher growth end markets, new innovation we continue to bring the market around Lightwell, around Gen. AI, and around, to your question, M&A growth synergies. All of that will play out.

為什麼？我們擁有具吸引力的 TAM（總可服務市場）；技術市場的大環境；我們產品組合的多元性；運用以混合雲與 AI 平台為中心的模式；持續進行產品組合優化以轉向更高成長的終端市場；我們持續帶給市場的新創新——包括 Lightwell、生成式 AI（Gen. AI），以及你提到的併購帶來的成長協同效應。這些都會逐步展現。

But, from a segment perspective, since you asked about Software, I'll give you some early indicators:

不過，既然你問的是軟體，我就從事業部角度給你一些早期指標：

One, we got a high-value recurring revenue stream that's been accelerating all year long, $25 billion ARR book of business, high-single-digit growth. We execute to what our guide is here.

第一，我們擁有高價值的經常性收入來源，且全年都在加速：ARR（年度經常性收入）簿記規模250億美元，成長為高個位數。我們會依照目前給出的指引來執行。

We should be approaching double digits here in the fourth quarter on that 80%, high value. That's one critical indicator.

若依此推進，第四季在那80%的高價值部分，我們應該會接近兩位數成長。這是第一個關鍵指標。

Two, new innovation we're bringing to market. We're extremely excited, to Arvin's point, around Red Hat, the acceleration, the high-teens growth.

第二，我們正在把新創新推向市場。就如 Arvind 所說，我們對 Red Hat 的加速成長非常振奮，成長率達到十幾個百分點（high-teens）。

We've got $700 million worth of pipeline on Lightwell, alone, in front of us, right now, with a $1 trillion TAM opportunity around open-source vulnerabilities. We've got virtualization, a tremendous opportunity. We got hybrid-cloud containerization opportunity.

僅 Lightwell 這一項，我們眼前就有7億美元的管線（pipeline）；而在開源漏洞方面，TAM 機會高達1兆美元。在虛擬化方面，我們也有非常大的機會。我們也有混合雲容器化的機會。

So we feel pretty good about the markets we're in -- the investment in the portfolio that we had built up, overall.

因此，我們對所處的市場——以及我們整體已建立的產品組合投資——感覺相當不錯。

Three, we got -- 2027 is the year of growth synergies, right? 2026 was about integrating and getting Confluent into our mode of operation and go-to-market and scale. '27 is around the growth synergies.

第三，2027年是協同成長的一年，對吧？2026年重點在於整合，讓 Confluent 融入我們的營運模式、上市策略（go-to-market）與規模化能力。2027年則聚焦在成長協同效應。

And, finally, to the TP question, we've got high-teens growth in installed MIPS out in the marketplace. Last five years, our MIPS are up 2x. Last 10 years, our installed MIPS are up 10x. That is the monetization vehicle at TP.

最後，回到 TP 的問題：市場上的已安裝 MIPS 以十幾個百分點（high-teens）的速度成長。過去五年，我們的 MIPS 成長了2倍。過去十年，我們的已安裝 MIPS 成長了10倍。這就是 TP 的變現載體。

And Arvind talked about it. I talked about it. We think that's a growth acceleration back to our model, overall. We'll quantify that in January. But that gives you a little bit of perspective of our early thinking about '27 on Software in IBM.

Arvind 談過這點。我也談過這點。我們認為，這將使整體成長加速回到我們的模型軌道上。我們會在一月量化說明。但以上能讓你對我們目前對 IBM 軟體在2027年的初步思考有一些概念。

Great. Operator, let's take the next question.

很好。接線員，請進行下一個問題。

Raimo Lenschow, Barclays.

Barclays 的 Raimo Lenschow。

Perfect. Two quick ones, if I may. First, congrats on the Software performance.

太好了。如果可以的話，我有兩個簡短的問題。首先，恭喜軟體業務表現出色。

On data, 20% growth. Obviously, some of that is Confluent. But, like, what are you seeing there on upsell, cross-sell on Confluent? We have it for a few quarters now but that looks like it's starting to click for you. So can you double-click on that one?

關於資料（Data）方面，成長 20%。顯然其中一部分來自 Confluent。但你們在 Confluent 的加購（upsell）、交叉銷售（cross-sell）方面看到了什麼？我們已經持有（觀察）幾個季度了，但看起來你們開始抓到訣竅。所以能否再深入說明一下？

And then, one for Jim: Gross margins came up with investors. Can you just give us a comment there what drove the performance there?

另外一題給 Jim：投資人提到毛利率。你能否簡單評論一下，這次表現是由什麼因素驅動的？

Rimo, thanks for the question. And thanks for noticing the data growth.

Rimo，謝謝你的提問。也謝謝你注意到資料業務的成長。

If I look at Confluent -- I actually want to broaden it -- there is Confluent, which is important for moving data in real time, allows data to remain in place but still be available for AI but coupled to that are what are called lakehouses. And we are very excited about the upcoming cross-sell opportunity with Streamhouse, which is also another way to store data that is in flight.

如果我看 Confluent——其實我想把範圍拉大一點——有 Confluent，這對即時移動資料很重要，讓資料可以留在原處但仍可供 AI 使用；同時搭配的是所謂的 lakehouse。我們也非常期待即將到來與 Streamhouse 的交叉銷售機會；Streamhouse 也是另一種用來儲存「在途資料」（in-flight data）的方式。

So the Confluent piece unlocks your data, be it inside the enterprise, be it in an application, be it in a SaaS application, be it on a cloud but still have it completely available for AI agents, as well as inferencing -- really important capability. And you saw that growth pull through.

因此，Confluent 這一塊能解鎖你的資料，不論是在企業內部、在某個應用程式中、在 SaaS 應用中、或在雲端，但仍能讓 AI 代理（AI agents）以及推論（inferencing）完全可用——這是非常重要的能力。你也看到這樣的成長帶動效應。

But lakehouses are also important, as people are beginning to cleanse their data for leverage in AI. And we saw growth there, also.

但 lakehouse 也很重要，因為人們開始清理（cleanse）他們的資料，以便在 AI 上加以運用。我們在那裡也看到了成長。

And then, we are excited about what is going to come with the capability we are about to announce in a couple of weeks around the Streamhouse, which is also, then, going to be directly across-sell and upsell, on top of Confluent.

此外，我們也很期待接下來幾週內將要宣布的 Streamhouse 相關能力；那也將會在 Confluent 之上，直接帶來交叉銷售與加購（upsell）的機會。

So unlocking data for AI, making sure that you don't have to create multiple copies because that creates a lot of extra cost but still have it available as quickly as you need it is the fundamental capability we believe is going to be unlocked in the enterprise, leveraging all of these capabilities. And that contributes to the overall growth there.

所以，為 AI 解鎖資料、確保你不必建立多份拷貝（因為那會帶來大量額外成本），但仍能在你需要的速度下讓資料可用——這是我們相信企業將會被解鎖的核心能力，並會運用上述所有能力來實現。這也促成了整體的成長。

Jim, I'll give it to you for the gross-margins question.

Jim，毛利率的問題我交給你。

Raimo, thank you very much for noticing the operating-leverage engine that we've been driving across this company.

Raimo，非常感謝你注意到我們在全公司推動的營運槓桿引擎（operating-leverage engine）。

You look at, what -- the last four years were probably up, what, about 1,000 basis points of operating leverage across this company.

你看，過去四年我們在全公司大概提升了約 1,000 個基點的營運槓桿。

By the way, that has delivered high-quality, sustainable cash generation over the last four years. Our cash is up roughly $7 billion of growth in 700 basis points of free cash flow margin on that 1,000 basis points of operating leverage.

順帶一提，這在過去四年帶來了高品質、可持續的現金創造。在那 1,000 個基點的營運槓桿之下，我們的現金大約增加了 70 億美元，自由現金流利潤率提升了 700 個基點。

But, when you look underneath it, we continue to improve the structural competitiveness of this business model. We still believe we got a lot of headroom here.

但如果你往下看，我們持續改善這個商業模式的結構性競爭力。我們仍然相信這裡還有很大的提升空間（headroom）。

When you take a look at our profit earnings, up double digits; our operating margins are up 70 basis points, it's delivered by, we are driving in the third quarter 500 basis points of gross productivity in our model.

當你看我們的獲利（profit earnings），以兩位數成長；營業利益率提升了 70 個基點；這是因為我們在第三季於模型中推動了 500 個基點的毛生產力（gross productivity）。

We've always said in our model, about a third is for margin protection, about two-thirds enables us to drive that productivity flywheel, where we reinvest back into the business around R&D, innovation, go-to-market model ecosystem to drive that flywheel of growth and competitive advantage.

我們一直說，在我們的模型裡，大約三分之一用於毛利率防護（margin protection），約三分之二讓我們推動那個生產力飛輪（productivity flywheel）：把資源再投回業務，包括研發、創新、上市策略（go-to-market）模型、生態系，以推動成長與競爭優勢的飛輪。

500 basis points is well above what our model is, overall. How are we doing that? We're running our playbook. That playbook, we believe, has headroom. It's reimagining everything between Arvind and myself about how we execute and run this company.

500 個基點整體而言明顯高於我們模型的水準。我們怎麼做到的？我們在執行我們的作戰手冊（playbook）。我們相信那套手冊還有空間。這是 Arvind 和我之間對於如何執行與經營這家公司所做的全面再想像。

And that gets into leveraging data, a whole new way of workflow reimagination, embedding technology, digitization, AI into our company.

這就涉及以資料驅動、以全新的方式重塑工作流程（workflow reimagination），把技術、數位化與 AI 嵌入到公司之中。

We've been talking about it in prepared marks. How do you reimagine a whole new way for software development? How do we reimagine a whole new way on go-to-market models? How do we redesign the G&A structure -- the way we run our company?

我們在事先準備的發言中也談到過。如何重新想像一種全新的軟體開發方式？如何重新想像一種全新的上市策略（go-to-market）模式？如何重新設計 G&A（一般與行政）架構——也就是我們經營公司的方式？

And I think there's a lot of headroom there. A few years in, we are now north of $5.5 billion of productivity out of this company. That is a great flywheel for investment flexibility and capability.

我認為那裡還有很大的提升空間。幾年下來，我們現在已經從公司中釋放出超過 55 億美元的生產力。這是一個非常好的飛輪，能帶來投資彈性與能力。

So most important thing is, it provides that free cash flow generation engine and it applies that investment flexibility so we can continue to compete, both organically and inorganically, to invest.

所以最重要的是，它提供自由現金流的產生引擎，並帶來投資彈性，讓我們能持續競爭、持續投資，不論是有機或無機（併購）成長。

Operator, next question, please.

接線員，請下一個問題。

Brent Thill, Jefferies.

Jefferies 的 Brent Thill。

Thanks. Arvind, I'm curious, in some of the deals, maybe early in the year, there was a little bit of an AI bright light. I don't know what to do. I need to (inaudible) -- a way and investigate.

謝謝。Arvind，我很好奇，在一些交易中，可能在年初的時候，有一點 AI 的「亮點效應」。我不知道該怎麼做。我需要（聽不清）——找個方式並去調查。

And, maybe, the investigation period went on. And they realized maybe some of the systems couldn't actually achieve what the claims were. And I'm curious if you're maybe seeing some element of this in some of the deals recouping where the claims were big but, perhaps, the reality wasn't matching the claim and you're able to deliver on that?

而且，也許調查期拉長了。他們發現也許有些系統其實無法達到宣稱的效果。我很好奇，你是否在一些交易中看到這種情況：先前宣稱很大，但現實與宣稱不符，而你們能在這裡補位、交付真正的成果？

Is that something you've seen or is that a non-scenario that you've experienced?

這是你們看過的情況嗎？還是你們並沒有遇到這種情境？

So, Brent, let me make sure I parsed the question correctly but I'm going to maybe answer it in my own words.

Brent，我先確認我是否正確理解了你的問題，不過我可能會用我自己的話來回答。

I think what we're seeing is that, at least, larger enterprises are becoming a lot more cognizant and a lot more aware of what is the true ROI of all of their AI spend.

我認為我們看到的是，至少大型企業正變得更能辨識、也更清楚他們所有 AI 支出的真正投資報酬率（ROI）是什麼。

They're also getting worried if some of the AI spending, completely out of control with no real return.

他們也開始擔心某些 AI 支出可能完全失控，卻沒有真正的回報。

But the way I've always characterized it is, there is no question to us -- and this is the advice I give every C-suite member that I meet -- that what we're trying to give people on AI, the productivity, the advantages of AI, the way it's going to change work is very real.

但我一直以來的說法是：對我們而言毫無疑問——這也是我給每一位我見到的 C-level 高層的建議——我們想帶給大家的 AI 生產力、AI 的優勢、以及它將如何改變工作方式，都是非常真實的。

So let me not make any confusion around that. That said, a lot of people are underplaying how much work it is to integrate AI into your workflows, to fundamentally sit back and say, how do I redesign the workflow to take advantage of AI?

所以在這點上不要有任何混淆。話雖如此，很多人低估了把 AI 整合進工作流程需要多少工作量：你必須從根本退一步思考，我要如何重新設計工作流程，才能利用 AI？

It's not a little [RPA] tool that you bolt on and you get something. If you do not spend the time and energy to redesign your workflow, you're not going to get the full benefit.

這不是你外掛一個小小的 [RPA] 工具就能得到成果。如果你不花時間與精力去重新設計工作流程，你就無法獲得完整的效益。

So if you're willing to do that, there's a huge benefit to be had. But that implies you've got to pick a few, you've got to learn how to scale them for full impact across the enterprise, and don't do 100 little experiments where you, then, do not have enough investment to get them to scale.

所以如果你願意這麼做，就能獲得巨大的效益。但這也意味著你必須挑選少數幾個，學會如何把它們擴大到能在全企業產生完整影響；不要做 100 個小實驗，最後反而沒有足夠的投資把它們擴展起來。

That is what I think we are seeing. So I think what we're seeing, Brent, is, now, enterprises, as they're getting mature, as they're stopping the experimentation, are saying, all right, who can help me get through this bigger piece of workflow design, as well as integration into our existing systems?

這就是我認為我們正在看到的情況。所以我認為我們看到的是，Brent，現在企業在逐漸成熟、停止實驗之後，會說，好吧，誰能幫我完成這一大塊的工作流程設計，以及與我們既有系統的整合？

Because you're not likely going to throw out your ERP system so how do you leverage it but still get the benefits of AI? And then, they're picking their partners in terms of which software do they -- are they willing to trust to do this, which people have the knowledge to come do this with them; as well as, can you be really efficient about your token spend, as opposed to leave it completely uncontrolled?

因為你不太可能把你的 ERP 系統丟掉，那要如何在善用它的同時，仍然獲得 AI 的效益？接著，他們會在合作夥伴的選擇上做取捨：哪些軟體是他們願意信任來做這件事的——哪些人具備相關知識能與他們一起完成；以及，你是否能在 token 支出上做到非常有效率，而不是讓它完全失控？

So I agree, I think, with your thesis but, maybe, a slight redirection in terms of why we're seeing that change. And, yes, that is a tailwind for us. That is why -- I am going to repeat what I said at the beginning -- OpenShift, Confluent, mainframe with fraud are all seeing the benefits of that play go on, as we are proceeding down this path.

所以我同意你的論點，但也許在我們看到這個變化的原因上，稍微做一點調整。而且，是的，這對我們是順風。這也是為什麼——我會重複我一開始說的——OpenShift、Confluent、以及具備反詐欺能力的大型主機，都正在看到這套打法帶來的好處，因為我們正沿著這條路往前走。

Great. Operator, let's take the next question.

很好。接線員，請進行下一個問題。

Fatima Boolani, Citi.

Fatima Boolani，花旗。

Good afternoon. Thank you for taking my question.

午安。謝謝讓我提問。

I wanted to go back to the discussion around transaction processing, simply because it was such a remarkable sequential recovery in that franchise. And. Jim, I understand so much of that is piggybacking off of the ELA momentum that you regained in the quarter from the transactional side of the business.

我想回到交易處理（transaction processing）的討論，主要是因為那個業務線出現了非常顯著的連續性復甦。而且，Jim，我理解其中很大一部分，是來自你們在本季從交易型業務端重新找回的 ELA 動能所帶動。

But I'm curious if you can help us unpack a little bit more on that strength. I think you talked about a high-teens growth in MIPS shipped.

但我想請你協助我們再更深入拆解一下這股強勁動能。我記得你提到出貨的 MIPS 有高十幾％的成長。

So wondering if you could unpack and break down for us the quantity and shipped MIPS growth in some of those flip deals because those were existing customers -- and to the extent you've been able to leverage a little bit of pricing power in this situation because I know that's not historically a lever that you've used.

所以想請問你能否幫我們拆解並說明：在一些 flip deals（翻轉型交易）中，出貨量與出貨 MIPS 成長的構成——因為那些是既有客戶——以及在這種情況下，你們是否能在某種程度上運用一些定價能力，因為我知道這在歷史上並不是你們常用的槓桿。

So we'd love to get a little bit more second-order detail on that transaction-processing slope of recovery between quantity and quantity growth, as well as pricing uplift. Thank you.

因此，我們希望能取得更多第二層次的細節，來理解交易處理這條復甦斜率：包含數量與數量成長，以及價格上調的貢獻。謝謝。

Yeah. Thanks, Fatima. Appreciate the question. Let's double-click back down on TP:

好的。謝謝你，Fatima。很感謝這個問題。我們再把焦點拉回 TP：

First, let me just reiterate, what is the monetization model, right? TP is transaction processing that runs a mission-critical software, on top of the most enduring platform, that being mainframe.

首先，我先重申一下變現模式是什麼，對吧？TP 是交易處理，運行在一套關鍵任務（mission-critical）的軟體之上，而其底層是最持久的運算平台，也就是大型主機（mainframe）。

By the way, as we've talked about many times before, mainframe runs 70% of the industry's transaction volumes in terms of value: 45 of the top 50 banks, 4 out to 5 top Airlines, 7 of the top 10 retailers around the world.

順帶一提，正如我們過去多次談到的，大型主機承載了整個產業按價值計算約 70% 的交易量：前 50 大銀行中的 45 家、前 4 到 5 大航空公司、以及全球前 10 大零售商中的 7 家。

I think 75%, 78%, last number I've seen, of the Fortune 100 run mainframe as a core architecture. So TP is the core mission-critical software that enables the most secure, reliable, and scalable platform around.

我記得我看到的最新數字是，財富 100 強中有 75%、78% 以大型主機作為核心架構。所以 TP 是核心的關鍵任務軟體，使得這個平台能做到最安全、最可靠、且最具擴展性。

The way we monetize that value is, the more capacity out there, the more opportunity to monetize value, the more opportunity to monetize price, the more opportunity to monetize value creation of TP.

我們變現這個價值的方式是：外部容量越多，變現價值的機會就越大；變現價格的機會就越大；TP 所創造價值的變現機會也就越大。

So, when you take a look at our last two or three cycles, we run 145 million MIPS around the world. Some of those are standard MIPS on the platform. Some of those are new Linux-specialty MIPS, new cloud-native applications, which is why the mainframe has been a growth vehicle over the last two to three cycles, overall.

所以，當你回顧我們最近兩到三個週期，我們在全球運行約 1.45 億 MIPS。其中一部分是平台上的標準 MIPS。另一部分是新的 Linux 專用（specialty）MIPS、以及新的雲原生應用，這也是為什麼在過去兩到三個週期中，大型主機整體成為成長載體。

That 145 million MIPS, our z17, six quarters in, 125%, the highest recorded program to date; above z16, which was the prior highest program to date.

這 1.45 億 MIPS，我們的 z17 推出至今第六個季度，達到 125%，是迄今為止記錄到的最高專案表現；高於 z16，而 z16 先前是迄今最高的專案表現。

We're up, I think, 18% shipped MIPS through six quarters. We're up over 100% of shipped MIPS. Why do I differentiate each two?

我們在六個季度內，出貨 MIPS 成長約 18%。我們的出貨 MIPS 成長超過 100%。為什麼我要區分這兩個？

Read the first one as consumption, capacity, and workload. We can monetize value immediately. The second is the opportunity. It's the TAM, right, and us driving workloads on that.

第一個可以理解為使用量、容量與工作負載。我們可以立即把價值變現。第二個則是機會。那是 TAM（總可服務市場），以及我們在其上推動工作負載。

So when you take a look at it, that's what's been driving the step-function change and growing meaningful over time. It was only a handful of years that our TP portfolio was down, what, mid-single-digit, high-single-digit growth.

所以當你這樣看，這就是推動階躍式變化（step-function change）並隨時間顯著成長的原因。就在幾年前，我們的 TP 產品組合還是下滑的，對吧，大概是中個位數到高個位數的成長率（指負成長/下滑幅度）。

We then stabilized that growth about four years ago and then, through z16, we saw an acceleration. But I'll reiterate that z16, when you look at the launch year, it was about flat the launch year, then it went up to about mid-single digits, year plus one. And then, it went up to high-single digit, year plus two.

大約四年前我們先把成長穩住，然後在 z16 期間看到加速。但我再重申一次：z16 在上市年度（launch year）大致持平；到了上市後第一年（year plus one）上升到約中個位數；到了上市後第二年（year plus two）則上升到高個位數。

Right now, in z17, we were about flat, launch year. We got different dynamics playing out right now. So we're about prudently cautious calling flat for the year.

目前在 z17，我們在上市年度大致持平。現在有不同的動態正在發生。所以我們在判斷上保持審慎保守，仍以全年持平來看待。

But with 100% shipped MIPS and 18% installed MIPS, I would say that's a pretty good future monetization opportunity in 2027, going forward.

但以出貨 MIPS 成長 100% 以及已安裝（installed）MIPS 成長 18% 來看，我會說，這為 2027 以及往後帶來相當不錯的未來變現機會。

Great. Operator, let's take the next question.

很好。接線員，請進行下一個問題。

Brian Essex, JP Morgan.

Brian Essex，摩根大通。

Hi. Good afternoon. Thank you for taking the question. And congrats on the Software acceleration from me, as well.

嗨，午安。謝謝讓我提問。另外也恭喜你們軟體業務加速成長。

Arvind, I caught the prepared remarks or, at least, some of the news remarks that you had about clients deploying Super Intelligence in ways that transform entire operating models. And I appreciate that that's more than just increasing operating efficiency.

Arvind，我聽到你在事先準備的發言中——或至少是你的一些新聞性談話——提到客戶正在以能夠改變整個營運模式的方式部署「超級智慧」（Super Intelligence）。我也理解那不只是提升營運效率而已。

But the question I have for you is: As your customers assess those transformation efforts, how focused are they on sovereign AI? Is that becoming a more meaningful part of your conversations? And how are you seeing your customers focus on the protection of their data and IP, as they assess how they deploy AI for their transformation efforts?

但我想問的是：當你們的客戶評估這些轉型工作時，他們對主權 AI（sovereign AI）的關注度有多高？這是否正在成為你們對話中更重要的一部分？以及當客戶評估如何部署 AI 以推動轉型時，你們看到他們如何聚焦在資料與智慧財產（IP）的保護上？

Brian, thank you, first, for the question and thank you for noticing the precise words I did use in the prepared remarks.

Brian，首先謝謝你的提問，也謝謝你注意到我在事先準備的發言中所使用的精準措辭。

So to get to sovereign AI, sovereign AI is definitely becoming a much larger conversation. While we have talked about it for the last year and a half, I would not tell you that, in '24, it had much traction. In '25, a few people began to pay some attention. And, in '26, it is now beginning to pick up.

談到主權 AI，主權 AI 的確正在成為更大規模的討論主題。雖然我們在過去一年半一直在談這件事，但我不會告訴你在 2024 年它有很大的牽引力。在 2025 年，有少數人開始注意。而在 2026 年，它現在開始升溫。

And let me explain for everybody else exactly what we mean by that:

我也為其他人解釋一下我們所說的到底是什麼：

If you're going to use a large public LLM, effectively, while they're certainly not stealing your data, though they're implicitly maybe learning what questions are asked, there are traces left in the underlying logs and so on that are not the full but that do give some indication of what all it is being used for, as well as some elements of the data that is being used.

如果你要使用大型的公共 LLM，實際上，雖然它們當然不會竊取你的資料，但它們可能會在某種程度上隱含地學習人們提出了哪些問題；而且在底層日誌等地方會留下痕跡——雖然不是完整內容，但確實會提供一些線索，顯示它被用來做什麼，以及所使用資料中的某些元素。

So sovereign becomes, how do you make sure that none of that happens. So people are turning now to two kinds of answers.

因此當主權化之後，你要如何確保這些事情都不會發生。所以人們現在轉向兩類答案。

That is why I think there is such a big boom in AI infrastructure on-premise. And we can see some of the companies that benefit from that but (inaudible) -- so will we in our mainframe with our Spyre cards, as time goes on.

這也是為什麼我認為企業內部部署（on-premise）的 AI 基礎設施會有如此大的熱潮。我們也看到一些公司因此受惠，但（聽不清）——隨著時間推移，我們在大型主機上搭配 Spyre 卡也會受惠。

The other part of this is, what is driving the boom behind open weight models because, there, you have the permission and the IP rights to run them on either on-premise infrastructure or on private clouds where the enterprise controls what is happening to the underlying logs and data.

另一部分是：是什麼在推動開放權重模型（open weight models）背後的熱潮？因為在那裡，你擁有許可與智慧財產權，可以在內部部署的基礎設施上，或在企業能控制底層日誌與資料如何處理的私有雲上運行它們。

That is definitely what is happening there. And we can see that. And the boom behind some of the companies that are creating open-weight models is also increasing behind that.

那裡確實正在發生這種情況。而我們也看得到。因此，打造開放權重模型的一些公司，其背後的成長動能也在提升。

I'll reflect on a conversation I had with a large-bank CEO. In this case, it was outside the US but it's a very large and friendly country. And their view was, look, we'll do some early experiments on the larger frontier LLMs. But for what we're doing inside the bank, we need to run it in a way that is sovereign; meaning, nobody can take any of the underlying data or underlying logs that are there.

我想回顧一下我和一家大型銀行 CEO 的一段對話。這次是在美國以外，但那是一個非常大且友善的國家。他們的看法是：你看，我們會先在更大型的前沿（frontier）LLM 上做一些早期實驗。但對於我們在銀行內部要做的事情，我們需要以主權方式來運行；也就是說，沒有人可以取得任何底層資料或底層日誌。

I was talking to somebody who does trading and they're on -- we're not going to expose our underlying proprietary models to anything outside but we need to run this in a sovereign way.

我也和一位做交易的人聊過，他們的立場是——我們不會把底層的專有模型暴露給任何外部，但我們需要以主權方式來運行。

I believe this is going to turn into a tailwind for many players in the industry; certainly, the infrastructure and semiconductor players but also for those of us who help enterprises deploy in a secure and in a private way.

我相信這將成為產業中許多參與者的順風（tailwind）；當然包括基礎設施與半導體業者，也包括像我們這些協助企業以安全且私密的方式部署的人。

Great. Operator, let's take one last question.

很好。接線員，我們再接最後一個問題。

Matt Swanson, RBC Capital Markets.

RBC Capital Markets 的 Matt Swanson。

Great. Thank you. And maybe this is just a natural follow-up to what Brent was asking but you had mentioned earlier that what AI costs and who controls it have been, like, the two primary questions that enterprises start off with you.

很好。謝謝。也許這只是對 Brent 問題的自然延伸，但你先前提到，AI 的成本是多少以及由誰控制，幾乎是企業一開始就會和你討論的兩個主要問題。

Who controls it, obviously, fits right into that sovereign-AI conversation. But could you just talk a little bit about how people are addressing the trade-offs, if there are trade-offs between those two segments; and how they're architecturing their networks in order to maximize for both those things; and just how, like, symbiotic they can be?

由誰控制，顯然正好切入主權 AI 的討論。但你能否談談人們如何處理這兩個面向之間的取捨（如果確實存在取捨）；以及他們如何設計其網路架構，以同時最大化這兩件事；還有它們之間能有多「共生」？

So, Matt, thanks for that question and also allowing me to give a bit more detail on this topic.

Matt，謝謝你的問題，也謝謝你讓我能在這個主題上補充更多細節。

If I go back to late '24, the difference between the most advanced model, which, let's acknowledge, is mostly a frontier LLM model and alternate models was a 2:1 or 3:1 gap; meaning, if you ran some assessment, some benchmark and it was broad-based, not narrow, the top model might score a [70] and the open model might score a [20] or a [30].

如果我回到 24 年底，最先進模型——我們承認主要是前沿 LLM 模型——與替代模型之間的差距大約是 2:1 或 3:1；也就是說，如果你做一些評估、一些基準測試，而且是廣泛性的、不是狹窄的，頂尖模型可能得分是 [70]，而開放模型可能是 [20] 或 [30]。

You've got to react with, that's a lot of capability we're giving up. And we all know that performance, in some sense, outcome for the enterprise is highly dependent upon that. That's too big a gap. Nobody's going to accept that.

你的第一反應會是：我們放棄了太多能力。而且我們都知道，某種程度上，企業的表現——也就是結果——高度依賴於此。這差距太大了。沒有人會接受。

In late '25, we began to see that gap shrink where, maybe, the gap was not 70 to 20 or 30, it was 70 to [50]. But, when you looked at those curves, and you looked at the improvement, and you looked upon the sequential improvement, we were very comfortable that we'd get this to probably [100] for the top and [90] for the alternative model.

到了 25 年底，我們開始看到差距縮小；也許不再是 70 對 20 或 30，而是 70 對 [50]。但當你看那些曲線、看改善幅度、看逐期改善，我們非常有信心能把它推到：頂尖模型大概到 [100]，替代模型到 [90]。

So, now, the question becomes: How do you do that? Now, the question that the enterprise struggles with is: Where are my kernel engineers? Where are the people who worry about the underlying network? Who can make sure that the model is clustered on GPUs that are close to each other in network proximity? Who can make sure that you're doing the right caching of results so you get an incredible response?

所以現在問題變成：你要怎麼做到？企業真正苦惱的是：我的核心（kernel）工程師在哪裡？那些會關心底層網路的人在哪裡？誰能確保模型被叢集部署在網路距離彼此很近的 GPU 上？誰能確保你做了正確的結果快取（caching），以便得到驚人的回應速度？

I'm telling you that that gap is probably still there, except for the most sophisticated enterprise. So then, you go to, all right, I got it that the answer is 4 times more expensive but it's the easy answer.

我告訴你，除了最成熟、最精密的企業之外，這個差距大概仍然存在。所以接著你就會說：好，我懂了，答案是貴 4 倍，但它是最容易的答案。

As those of us who work in this world produce much more easy-button software that can give you that capability for the alternate models -- and I would say it's not a huge amount of infrastructure.

當我們這些在這個領域工作的人，產出更多「一鍵式」（easy-button）的軟體，能把替代模型的那種能力交付給你——而我會說，所需的基礎設施其實不算多。

Our competition is that, for an average large enterprise, you can get going with maybe two or three racks, which is not a huge amount of infrastructure but it gives you the ability to run those open-weight models.

我們的競爭在於：對一般大型企業而言，你可能用兩到三個機櫃（racks）就能啟動，這不算龐大的基礎設施，但它能讓你有能力去跑那些開放權重模型。

Currently, anywhere from 25% to 10% of the token cost of the frontier models, that is very attractive, given that they are now 90% as effective.

目前其 token 成本大約是前沿模型的 25% 到 10%，在它們現在已經有 90% 效果的前提下，這非常有吸引力。

It's a simple question for any enterprise: Would you like something that's 10% of the cost, if it's 90% as good as the best? Or do you want to pay that 10 times higher cost?

對任何企業來說，這是一個很簡單的問題：如果有一個方案成本只有 10%，但效果有最好的 90%，你要不要？還是你想支付高 10 倍的成本？

And then, the question is: How many tasks require that? I would tell you some do. But the vast majority -- summarizing an e-mail, examining a contract, answering a customer call, composing an e-mail -- all of those can be done, as well, with the much cheaper answer. And that is what is going to drive a lot of this.

接著問題是：有多少任務真的需要那樣？我會告訴你，有些確實需要。但絕大多數——摘要一封電子郵件、檢視一份合約、回覆一通客戶來電、撰寫一封電子郵件——這些都同樣可以用便宜得多的方案做好。而這就是將推動其中很大一部分的原因。

So, Matt and everybody else, thank you for those questions.

所以，Matt 以及各位，謝謝你們的提問。

Look, we are pleased with our performance this quarter. And all our segments accelerated growth. Our portfolio strength, disciplined execution, and relentless focus on productivity reinforce my confidence in our trajectory.

各位，我們對本季的表現感到滿意。而且我們所有事業部門的成長都在加速。我們的產品組合優勢、嚴謹的執行力，以及對生產力的持續專注，強化了我對我們發展軌跡的信心。

I look forward to sharing our progress with you, as we close out the year. And we'll talk to you in January about that.

我期待在我們結束今年之際，與各位分享我們的進展。我們也會在一月再和各位談這件事。

Thank you, Arvind. Operator, let me turn it back to you to close out the call.

謝謝你，Arvind。接線員，請讓我把話交回給你來結束本次電話會議。

Thank you for participating on today's call. The conference has now ended. You may disconnect at this time.

感謝各位參與今天的電話會議。本次會議到此結束。您現在可以中斷連線。