PubMatic Inc (PUBM) 2025 Q3 法說會逐字稿

Hello, everyone, and welcome to PubMatic's third quarter 2025 earnings call. My name is Emmanuel, and I will be your Zoom operator today. Thank you for your attendance today. As a reminder, this webinar is being recorded.

大家好，歡迎參加 PubMatic 2025 年第三季財報電話會議。我叫伊曼紐爾，今天我將擔任你們的 Zoom 操作員。感謝各位今天的蒞臨。再次提醒，本次網路研討會正在錄製。

I will now turn the call over to Stacie Clements with The Blueshirt Group.

現在我將把電話轉給藍衫集團的史黛西·克萊門茨。

Good afternoon, everyone, and welcome to PubMatic's earnings call for the third quarter of 2025. This is Stacie Clements with The Blueshirt Group, and I'll be your operator today. Joining me on the call are Rajeev Goel, Co-Founder and CEO; and Steve Pantelick, CFO. Before we get started, I have a few housekeeping items. Today's prepared remarks have been recorded, after which Rajeev and Steve will host live Q&A.(EventInstructions) A copy of our press release can be found on the website at investors.pubmatic.com.

各位下午好，歡迎參加 PubMatic 2025 年第三季財報電話會議。這裡是藍衫集團的史黛西·克萊門茨，今天我將為您接聽電話。與我一同參加電話會議的還有共同創辦人兼執行長 Rajeev Goel 和財務長 Steve Pantelick。在我們開始之前，我還有一些準備工作要做。今天的演講稿已錄製完畢，之後拉吉夫和史蒂夫將主持現場問答環節。 （活動說明）您可以在 investors.pubmatic.com 網站上找到我們的新聞稿副本。

I would like to remind participants that during this call, management will make forward-looking statements, including, without limitation, statements regarding our future performance, market opportunity, growth strategy and financial outlook. Forward-looking statements are based on our current expectations and assumptions regarding our business, the economy and future conditions.

我想提醒各位參與者，在本次電話會議中，管理層將發表前瞻性聲明，包括但不限於有關我們未來業績、市場機會、成長策略和財務前景的聲明。前瞻性陳述是基於我們目前對公司業務、經濟和未來狀況的預期和假設。

These forward-looking statements are subject to inherent risks, uncertainties and changes in circumstances that are difficult to predict. You can find more information about these risks, uncertainties and other factors in our reports filed from time to time with the Securities and Exchange Commission and are available at investors.pubmatic.com, including our most recent Form 10-K and any subsequent filings on Form 10-Q or 8-K.

這些前瞻性陳述存在固有的風險、不確定性和難以預測的情況變化。您可以在我們不時向美國證券交易委員會提交的報告中找到有關這些風險、不確定性和其他因素的更多信息，這些報告可在 investors.pubmatic.com 上獲取，包括我們最新的 10-K 表格以及任何後續提交的 10-Q 表格或 8-K 表格。

Our actual results may differ materially from those contemplated by the forward-looking statements. We caution you, therefore, against relying on any of these forward-looking statements. All information discussed today is as of November 10, 2025, and we do not intend and undertake no obligation to update any forward-looking statement, whether as a result of new information, future developments or otherwise, except as may be required by law.

我們的實際結果可能與前瞻性聲明中設想的結果有重大差異。因此，我們提醒您不要依賴任何此類前瞻性陳述。今天討論的所有資訊均截至 2025 年 11 月 10 日，我們不打算也不承擔任何更新任何前瞻性聲明的義務，無論是由於新資訊、未來發展或其他原因，除非法律要求。

In addition, today's discussion will include references to certain non-GAAP financial measures, including adjusted EBITDA, non-GAAP net income, cash flow from operations and free cash flow. These non-GAAP measures are presented for supplemental informational purposes only and should not be considered a substitute for financial information presented in accordance with GAAP. A reconciliation of these measures to the most directly comparable GAAP measures is available in our press release.

此外，今天的討論還將提及某些非公認會計準則財務指標，包括調整後的 EBITDA、非公認會計準則淨收入、營業活動現金流量和自由現金流量。這些非GAAP指標僅供補充資訊之用，不應被視為取代依照GAAP編製的財務資訊。我們的新聞稿中提供了這些指標與最直接可比較的 GAAP 指標的調整表。

And now I will turn the call over to Rajeev.

現在我將把電話交給拉吉夫。

Thank you, Stacie, and welcome, everyone. We delivered a stronger-than-expected quarter with revenue and adjusted EBITDA ahead of guidance as well as strong cash flow, demonstrating the power of our platform, the continued diversification of our business and our accelerated pace of innovation.

謝謝Stacie，也歡迎各位。本季業績超出預期，營收和調整後 EBITDA 均高於預期，現金流強勁，證明了我們平台的實力、業務的持續多元化以及創新步伐的加快。

CTV significantly outpaced the market rate of growth and grew over 50% year-over-year, excluding political, driven by increased premium supply, continued scaling of agency marketplaces, traction in our live sports marketplace and growth of small and mid-market advertisers. Emerging revenues grew over 80% year-over-year as sell-side targeting and newly launched AI solutions quickly ramped. We also strengthened our end-to-end platform with cutting-edge AI innovations that are deepening our competitive moat and unlocking measurable incremental revenue opportunities.

CTV 的成長速度明顯超過市場平均水平，較去年同期成長超過 50%（不包括政治廣告），這主要得益於優質廣告供應的增加、代理商市場的持續擴張、體育直播市場的蓬勃發展以及中小市場廣告主的成長。隨著賣方目標定位和新推出的 AI 解決方案迅速發展，新興市場營收年增超過 80%。我們也利用尖端人工智慧創新技術加強了端到端平台，從而加深了我們的競爭優勢，並釋放了可衡量的增量收入機會。

The industry is rapidly redefining itself, and we are actively shaping its future. The impending Google Ad tech remedies verdict will very likely shift market share. The prioritization of data targeting and performance is shifting value creation in the ecosystem to the sell side. And over the past few months, we've seen a groundswell of AI-driven innovation reshaping the entire ecosystem. AI is now at the center of every strategic conversation, whether the objective is advertising performance, transparency or automation.

該行業正在迅速重塑自身，我們正在積極塑造其未來。即將出爐的Google廣告技術補救措施判決很可能會影響市場佔有率。數據定位和性能的優先性正在將生態系統中的價值創造轉移到銷售方。在過去的幾個月裡，我們看到人工智慧驅動的創新浪潮正在重塑整個生態系統。無論是廣​​告效果、透明度或自動化，人工智慧現在都處於所有策略對話的核心。

As an early adopter of AI, our leadership is a defining advantage for us, which will grow over time. We continue to innovate with an AI strategy centered around three distinct layers of programmatic advertising, the infrastructure layer, the application layer and the transaction layer. For the infrastructure layer, we own and operate our full technology stack, giving us the efficiencies, control and independence that many of our peers don't have, evidenced by multiple recent public cloud outages.

身為人工智慧的早期採用者，我們的領先地位是我們的決定性優勢，而且隨著時間的推移，這一優勢還會不斷擴大。我們不斷創新，以人工智慧為核心的程序化廣告策略圍繞著三個不同的層面：基礎設施層、應用層和交易層。在基礎設施層面，我們擁有並經營我們自己的完整技術棧，這使我們擁有許多同行所不具備的效率、控制權和獨立性，最近多次公共雲服務中斷就證明了這一點。

Through our technical collaboration with NVIDIA, we are deploying next-generation AI models on the world's most advanced GPU architecture. Five years in the making, this collaboration required three ingredients, physical infrastructure capable of deploying GPUs at scale, massive transaction volume to test and optimize across the full open Internet and technical sophistication to be an early adopter. Today, our infrastructure is a clear differentiator, and we believe years ahead of peers.

透過與 NVIDIA 的技術合作，我們正在世界上最先進的 GPU 架構上部署下一代 AI 模型。歷時五年，這項合作需要三個要素：能夠大規模部署 GPU 的實體基礎設施、用於在整個開放互聯網上進行測試和優化的大量交易量，以及成為早期採用者所需的技術水平。如今，我們的基礎設施是一項明顯的差異化優勢，我們相信我們領先同行多年。

The business impacts are tangible, 5x faster bid responses and 85% fewer auction timeouts, all recovering millions in ad spend. These results close the infrastructure advantage of walled gardens and directly translate into advertiser performance with higher publisher yield.

業務影響是切實可見的，競價響應速度提高了 5 倍，拍賣超時減少了 85%，所有這些都節省了數百萬美元的廣告支出。這些結果縮小了封閉式平台的基礎設施優勢，並直接轉化為更高的廣告商收益，從而提升了廣告主的績效。

Looking ahead, as autonomous AI agents begin planning and negotiating ad campaigns, industry compute requirements are expected to grow dramatically. These early investments only widen our infrastructure advantage as legacy platforms are constrained by cloud and computing limits. This is the monumental shift for open Internet digital advertising that PubMatic has been building for.

展望未來，隨著自主人工智慧代理商開始規劃和協商廣告活動，產業運算需求預計將大幅成長。這些早期投資只會進一步擴大我們的基礎設施優勢，因為傳統平台受到雲端運算和運算能力的限制。這是 PubMatic 一直致力於打造的開放式網路數位廣告的重大變革。

Next, at the application layer, we're deploying some of the most exciting and innovative capabilities we've ever launched, embedding AI directly into our products to power intelligent workflows and decision automation. We launched AI-powered buyer and publisher platforms that now handle more complexity with significantly less manual effort. Our solutions cut campaign setup time by 87% and speed up issue resolution by 70%, translating directly into faster activations, higher productivity and better outcomes for our customers.

接下來，在應用層，我們將部署一些我們推出過的最令人興奮和最具創新性的功能，將人工智慧直接嵌入我們的產品中，以支援智慧工作流程和決策自動化。我們推出了人工智慧驅動的買家和發布商平台，現在可以處理更複雜的操作，同時大幅減少人工幹預。我們的解決方案可將活動設定時間縮短 87%，並將問題解決速度提高 70%，直接轉化為更快的啟動速度、更高的生產力和更好的客戶成果。

Independent agency Butler/Till has been using our AI-powered PubMatic for buyers' platform. Scott Ensign, their Chief Strategy Officer, said, and I quote, historically, systems don't talk to each other. Data sets are disparate, walled garden data is hard to connect. AI allows us to scale human reasoning and run campaigns that truly look across all channels and optimize across them. Working with PubMatic for buyers helps make that possible, end quote.

獨立代理商 Butler/Till 一直在使用我們基於人工智慧的 PubMatic 買家平台。他們的首席策略官史考特‧恩賽恩說（我引用他的原話），從歷史上看，系統之間是無法相互溝通的。資料集分散，封閉的資料平台難以連線。人工智慧使我們能夠擴展人類推理能力，並開展真正跨所有管道進行優化和優化的行銷活動。與 PubMatic 合作，買家就能實現這一點，最終報價。

This same enthusiasm is building on the publisher side. Our newly launched publisher suite already includes 17 operational AI agents guiding yield, diagnostics and creative setup. Customer feedback has been exceptional. One of our largest omnichannel partners, Overwolf, told us that the PubMatic Assistant AI chatbot is unique with respect to the accuracy and speed of execution.

出版商方面也湧現出了同樣的熱情。我們新推出的出版商套件已經包含 17 個可運行的 AI 代理，用於指導收益、診斷和創意設定。客戶反饋非常好。我們最大的全通路合作夥伴之一 Overwolf 告訴我們，PubMatic Assistant AI 聊天機器人在執行的準確性和速度方面是獨一無二的。

And finally, at the transaction layer, we're preparing for the next major step at Agentic AI, where advertisers and publishers AI agents will be able to transact directly through our infrastructure. We are a co-founder of the newly established Ad Context Protocol, or ACP, alongside partners like Yahoo, LG Ad Solutions and Raptive and the first to publish a model context protocol specification for agent-to-agent communication in the programmatic industry. We are establishing the protocols, safety mechanisms and the interoperability standards that will enable AI agents across the entire ecosystem to transact efficiently and securely.

最後，在交易層，我們正在為 Agentic AI 的下一個重大步驟做準備，屆時廣告商和出版商的 AI 代理將能夠透過我們的基礎設施直接進行交易。我們與 Yahoo、LG Ad Solutions 和 Raptive 等合作夥伴共同創立了新成立的廣告上下文協議 (ACP)，並且是第一個發布程序化行業代理間通信模型上下文協議規範的機構。我們正在建立協議、安全機制和互通性標準，使整個生態系統中的人工智慧代理能夠有效率且安全地進行交易。

With this three-layer strategy, infrastructure, application and transaction, we are building the complete system for Agentic AI, the high-performance car, the roads to drive on and the traffic laws to govern it all. While still in early days, we are already seeing material benefits. First, AI is driving increased platform usage. As we roll out new generative AI and Agentic AI features across our platform, our customers are able to launch campaigns and resolve issues faster and improve performance. Validating our leadership in AI, customers have repeatedly said they are not seeing this level of innovation from other companies in the industry.

透過基礎設施、應用和交易這三層策略，我們正在建立完整的智慧體人工智慧系統、高性能汽車、行駛道路以及管理這一切的交通法規。雖然目前還處於早期階段，但我們已經看到了實質的好處。首先，人工智慧正在推動平台使用量的成長。隨著我們在平台上推出新的生成式人工智慧和智慧體人工智慧功能，我們的客戶能夠更快地進行行銷活動、解決問題並提高績效。客戶一再表示，他們在業內其他公司沒有看到這種程度的創新，證明了我們在人工智慧領域的領先地位。

Second, AI solutions are generating new revenue streams. One example is our new AI-based yield optimization solution for publishers, which uses adaptive learning models to automate pricing and improve auction efficiency. This AI solution is driving growth for our publishers, increasing their revenue on average by 10%. Launched just a few months ago, it has already unlocked tens of millions of dollars in incremental revenue for our publishers and in turn, is generating new pragmatic revenue as part of our emerging revenue category.

其次，人工智慧解決方案正在創造新的收入來源。例如，我們為出版商推出的全新基於人工智慧的收益優化解決方案，利用自適應學習模型實現定價自動化並提高競價效率。該人工智慧解決方案正在推動我們出版商的成長，平均使其收入成長 10%。該產品推出僅幾個月，就已經為我們的出版商帶來了數千萬美元的增量收入，反過來，它也正在作為我們新興收入類別的一部分，創造新的務實收入。

And third, AI is improving our operational efficiency and profitability. In the last two months, we deployed a dozen AI agents internally to automate operational workflows, accelerate development and reduce overhead. Our goal is to deploy substantially more agents in the coming quarters to give us measurable margin leverage while we continue to invest and strengthen our long-term moat.

第三，人工智慧正在提高我們的營運效率和獲利能力。在過去的兩個月裡，我們在內部部署了十幾個人工智慧代理，以實現營運工作流程的自動化、加快開發速度並降低營運成本。我們的目標是在接下來的幾個季度部署更多代理商，從而獲得可衡量的利潤槓桿，同時我們將繼續投資並加強我們的長期護城河。

Looking ahead, despite the significant progress we have made, we believe we're just scratching the surface. AI will continue to drive higher usage across our platform, generate incremental revenue streams and improve operational leverage. And because we're investing across all three layers, thematic is poised to lead the next era of Agentic AI advertising. While AI is a powerful driver of our long-term growth strategy, it's equally important that we execute across the four other strategic priorities I outlined last quarter. I'm pleased to report that we're making significant progress in each of those areas.

展望未來，儘管我們已經取得了顯著進展，但我們相信我們只是觸及了冰山一角。人工智慧將繼續推動我們平台的使用率提高，創造更多收入來源，並提高營運效率。因為我們在所有三個層面都進行了投資，主題廣告有望引領下一代智慧AI廣告。雖然人工智慧是我們長期成長策略的強大驅動力，但同樣重要的是，我們要執行我上個季度概述的其他四個策略重點。我很高興地報告，我們在這些領域都取得了顯著進展。

First, we're broadening our demand side ecosystem and accelerating our pipeline. We expanded a top three DSP partnership, introducing programmatic guaranteed deals that streamline execution for advertisers across premium streaming content. This integration reduces friction in deal setup, accelerates time to market for campaigns and unlocks incremental budgets. It's a great example of how our relationships with global DSP partners are becoming more strategic as we diversify beyond the legacy DSPs. We also launched a new partnership with Bliss, an omnichannel DSP that brings high-value demand from leading global brands across automotive, retail and financial services.

首先，我們正在擴大需求面生態系統，並加快專案進度。我們擴大了與前三大 DSP 合作夥伴的合作，推出了程序化擔保交易，簡化了廣告主在優質串流媒體內容上的投放流程。這種整合減少了交易達成過程中的摩擦，加快了行銷活動的上市速度，並釋放了額外的預算。這很好地說明了隨著我們擺脫對傳統 DSP 的依賴，與全球 DSP 合作夥伴的關係變得越來越具有策略性。我們還與全通路 DSP 平台 Bliss 建立了新的合作夥伴關係，該平台為汽車、零售和金融服務等行業的全球領先品牌帶來了高價值需求。

Bliss combines T-Mobile's app engagement data with real-world movement patterns and transaction signals to drive performance-focused campaigns with measurable outcomes from brand awareness to store visits and sales. This partnership expands our reach into premium brand advertisers who prioritize full funnel measurement and offline attribution.

Bliss 將 T-Mobile 的應用程式互動數據與現實世界的行動模式和交易訊號相結合，以推動以績效為導向的行銷活動，並實現從品牌知名度到門市訪問量和銷售額等可衡量的結果。此次合作將我們的業務拓展至那些重視全通路衡量和線下歸因的高端品牌廣告商。

These mid-market focused DSPs like Bliss represent one of the fastest-growing advertising segments. In Q3, ad spend from this segment grew 25%-plus year-over-year, reflecting meaningful progress in our diversification strategy. Second, we're accelerating our investment on the buy side. We're extending our reach with independent agencies and direct advertisers, expanding our focus from the top 20 agencies to the top 150 and from the top 500 advertisers to approximately 1,500. Supply path optimization remains a key growth driver with the majority of this addressable market as greenfield opportunity.

像 Bliss 這樣專注於中階市場的 DSP 平台，是成長最快的廣告細分市場之一。第三季度，該領域的廣告支出年增超過 25%，反映出我們在多元化策略方面取得了實質進展。其次，我們正在加快收購方面的投資。我們正在擴大與獨立代理商和直接廣告商的合作範圍，將重點從前 20 家代理商擴大到前 150 家，將重點從前 500 家廣告商擴大到大約 1500 家。供應鏈優化仍然是關鍵的成長驅動因素，而這部分潛在市場的大部分都是全新的發展機會。

SPO represented over 55% of activity on our platform in Q3. As a pioneer in SPO, we are the leading incumbent offering scale and a rich history of performance and efficiency gains. Building on this momentum, we are onboarding more buyers on to Activate, our direct-to-supply buying platform. Over the first nine months of the year, the number of active campaigns grew more than 4x over last year with a 35% increase in customers.

第三季度，SPO 占我們平台活動總量的 55% 以上。作為SPO領域的先驅，我們是領先的現有企業，擁有規模優勢和豐富的績效及效率提升經驗。乘著這股勢頭，我們正在將更多買家引入我們的直接供貨採購平台 Activate。今年前九個月，活躍的行銷活動數量比去年同期增加了 4 倍多，客戶數量增加了 35%。

Activate is a key solution that enables the ecosystem-wide push for increased performance, transparency and efficiency of programmatic advertising. And this is only the beginning. We've begun integrating AI-powered agent-to-agent workflows into Activate to boost performance and reduce friction, making advertiser adoption even easier. We believe that this new technology could have a massive impact on Activate adoption over the medium-term as we accelerate investment in mid-tail buyers. Finally, we continue to deepen our integration with DSPs to create value beyond real-time bidding transactions.

Activate 是一項關鍵解決方案，它推動了整個生態系統朝向提高程式化廣告的效能、透明度和效率邁進。而這只是個開始。我們已開始將人工智慧驅動的代理間工作流程整合到 Activate 中，以提高效能並減少摩擦，使廣告主更容易採用。我們相信，隨著我們加快對中端買家的投資，這項新技術在中期內可能會對 Activate 的普及產生巨大影響。最後，我們將繼續深化與DSP的整合，以創造即時競價交易以外的價值。

We are the first SSP to integrate the Trade Desk's price discovery and provisioning API, which allows publishers and advertisers to share deal metadata between our platforms to better identify and resolve issues with underperforming deals in real time. Today, over 50% of programmatic deals sit dormant because this information was previously only available offline. We anticipate this innovation will accelerate our share of PMP and PG deals as we drive adoption together with the Trade Desk.

我們是首家整合 Trade Desk 的價格發現和供應 API 的 SSP，該 API 允許發布商和廣告商在我們的平台之間共享交易元數據，以便更好地即時識別和解決表現不佳的交易問題。如今，超過 50% 的程序化交易處於停滯狀態，因為這些資訊以前只能在線下取得。我們預計，隨著我們與交易台共同推動採用，這項創新將加速我們在 PMP 和 PG 交易中的份額。

Third, our momentum in Activate is also fueling our growth in CTV. Excluding political advertising, CTV revenue grew more than 50% year-over-year, and the format remains a primary growth engine for our business. Live sports is an especially exciting category. Buying activity rose more than 150% sequentially from Q2 to Q3 as we scaled our AI-powered live sports marketplace and launched new programmatic guaranteed deals around tentpole events like the US Open for Tennis and Monday Night Football.

第三，我們在 Activate 領域的成長動能也推動了我們在 CTV 領域的成長。不計政治廣告，CTV 收入年增超過 50%，並且該形式仍然是我們業務的主要成長引擎。體育賽事直播是一個特別令人興奮的類別。隨著我們擴大人工智慧驅動的體育賽事直播市場規模，並圍繞美國網球公開賽和周一晚間橄欖球賽等重要賽事推出新的程序化擔保交易，購買活動從第二季度到第三季度環比增長超過 150%。

We also continue to expand our CTV publisher footprint. New deals and expanded partnerships with a number of free ad-supported streaming services, including Tubi, Future Today and Local Now added to a strong roster with over 90% of the top 30 global streamers now on PubMatic. We offer these premium content streamers incremental ad demand that other platforms can't offer because of the scale of our SPO, Activate, Curation and commerce businesses.

我們也在不斷擴大我們的 CTV 出版商覆蓋範圍。PubMatic 與多家免費廣告支援的串流媒體服務商（包括 Tubi、Future Today 和 Local Now）達成了新的合作協議並擴大了合作關係，使其強大的合作夥伴陣容更加強大，目前全球排名前 30 的串流媒體服務商中有超過 90% 都已入駐 PubMatic。由於我們的 SPO、Activate、Curation 和電商業務規模龐大，我們能夠為這些優質內容串流媒體用戶提供其他平台無法提供的增量廣告需求。

For example, Fremantle, one of the world's largest entertainment content creators behind franchises like American Idol, America's Got Talent and the X Factor, generated a 78% increase in incremental programmatic demand across their expanding FAST channel portfolio by partnering with PubMatic.

例如，Fremantle 是全球最大的娛樂內容創作公司之一，旗下擁有《美國偶像》、《美國達人秀》和《X音素》等熱門節目。該公司透過與 PubMatic 合作，使其不斷擴展的 FAST 頻道組合的程序化廣告需求增加了 78%。

This is a remarkable outcome and highlights significant incremental ad revenue our platform generates for our partners. Additionally, we're expanding ad formats on our platform. In collaboration with Dentsu Group, we recently launched pause ads for CTV through Activate. Advertisers can now serve dynamic, contextually relevant ads when viewers pause content, representing a premium brand-safe moment that boosts engagement and yields incremental revenue for publishers. What's more, with $155 billion of ad dollars still in linear television, we believe our CTV business has a long runway for growth given the scale, performance and ad formats now available for buyers on PubMatic.

這是一個了不起的成果，凸顯了我們的平台為合作夥伴帶來的顯著增量廣告收入。此外，我們正在拓展平台上的廣告形式。我們最近與電通集團合作，透過 Activate 平台推出了 CTV 暫停廣告。廣告主現在可以在觀眾暫停內容時投放動態的、與上下文相關的廣告，這代表著一個優質的品牌安全時刻，可以提高參與度並為出版商帶來額外的收入。此外，由於線性電視廣告支出仍高達 1,550 億美元，我們相信，鑑於 PubMatic 平台目前為買家提供的規模、效果和廣告格式，我們的 CTV 業務還有很長的成長之路。

And fourth, we're making significant progress in scaling our emerging revenue streams, which grew over 80% year-over-year in the third quarter. Commerce Media continues to gain momentum. We continue onboarding and scaling with some of the world's leading retailers and transaction-based enterprises as they seek to activate and monetize their first-party audience intelligence. These partnerships are expanding our reach beyond the traditional impression model while generating platform fees and database monetization that accelerate revenue growth.

第四，我們在擴大新興收入來源方面取得了重大進展，第三季年增超過 80%。商業媒體持續發展壯大。我們持續與一些世界領先的零售商和交易型企業合作，幫助他們啟動第一方受眾情報並將其變現，從而不斷擴大合作規模。這些合作關係正在將我們的影響力擴展到傳統展示模式之外，同時產生平台費用和資料庫貨幣化，從而加速收入成長。

Sell-side curation is another fast-growing emerging revenue stream. We expanded partnerships with leading data providers around the world. Nielsen, for example, tapped PubMatic as their exclusive sell-side partner to bring their more than 10,000 audience segments to Australian advertisers and agencies. Together with the previously mentioned AI yield solution for publishers, these initiatives drive incremental high-margin revenue that is scaling quickly.

賣方內容策劃是另一個快速成長的新興收入來源。我們擴大了與全球領先數據提供商的合作關係。例如，尼爾森選擇 PubMatic 作為其獨家銷售方合作夥伴，將其超過 10,000 個受眾群體帶給澳洲的廣告商和代理商。結合前面提到的出版商的 AI 收益解決方案，這些舉措推動了高利潤收入的快速成長。

In closing, our results demonstrate the power of our differentiated business model. We continue to innovate, diversify our business and operate with discipline. We are leading from a position of strength. We're confident that the investments we're making today, particularly across the three layers of our AI strategy, are expanding our competitive advantage while creating sustainable, profitable growth over the mid- to long-term. All of this is happening alongside a once-in-a-generation shift in digital advertising that will likely result in the competitive landscape being reshaped, where even a modest share shift could unlock substantial incremental high-margin revenue for us given our owned and operated infrastructure.

總之，我們的結果證明了我們差異化商業模式的強大力量。我們不斷創新，拓展業務，並以嚴謹的態度運作。我們憑藉實力優勢處於領先地位。我們相信，我們今天所做的投資，尤其是在我們人工智慧策略的三個層面上的投資，正在擴大我們的競爭優勢，同時在中長期內創造可持續的、盈利的成長。所有這一切都發生在數位廣告領域百年一遇的變革之中，這很可能會導致競爭格局的重塑，即使市場份額發生適度的變化，鑑於我們擁有並運營的基礎設施，也能為我們帶來可觀的高利潤收入。

PubMatic is not only positioned to adapt, we're helping define what comes next. We have the technology, the talent and the financial foundation to build a more intelligent, efficient and enduring business, one that creates lasting value for our customers, partners and shareholders.

PubMatic不僅做好了適應變化的準備，我們還在幫助定義未來。我們擁有技術、人才和財務基礎，可以打造一個更智慧、更有效率、更持久的企業，為我們的客戶、合作夥伴和股東創造持久價值。

Let me now turn the call over to Steve.

現在我把電話交給史蒂夫。

Thank you, Rajeev, and welcome, everyone. We exceeded expectations on both revenue and adjusted EBITDA. This outperformance was driven by CTV, combined with stronger-than-expected year-over-year growth for online video and mobile app. We managed expenses, leveraged AI and delivered improved margins and strong free cash flow. Stepping back, it is important to call out that our efforts to transform our business started several years ago as we anticipated the value of the ecosystem shifting to the sell-side.

謝謝你，拉吉夫，也歡迎各位。我們在營收和調整後 EBITDA 方面都超出了預期。這一優異表現主要得益於連網電視 (CTV) 的發展，以及線上視訊和行動應用程式年增超預期。我們控制了支出，利用人工智慧，實現了更高的利潤率和強勁的自由現金流。回顧過去，值得一提的是，我們早在幾年前就開始著手轉型業務，因為我們預見生態系統的價值會轉移到賣方。

We built an end-to-end solution that prioritizes control, performance and transparency while recognizing the need to diversify our business and unlock new paths to monetization. Today, over 40% of our revenue is derived from CTV, mobile app and emerging revenue streams, which represent long-term value for our business, up from less than 30% two years ago. Further, these efforts directly benefit our profitability and enable continued innovation and investment in the business.

我們建立了一個端到端的解決方案，優先考慮控制、效能和透明度，同時認識到需要實現業務多元化並開闢新的獲利途徑。如今，我們超過 40% 的收入來自連網電視、行動應用和新興收入來源，這代表著我們業務的長期價值，而兩年前這一比例還不到 30%。此外，這些努力直接提高了我們的獲利能力，並使我們能夠繼續進行創新和投資。

Turning to the quarterly results, starting with the revenue breakdown. To provide apples-for-apples comparability, year-over-year growth rates for video are adjusted to exclude political ad spend. On that basis, total omnichannel video revenues grew 21%, underscoring the strength of our premium video portfolio and growing adoption of AI-powered optimization across formats. Within this category, CTV once again grew over 50% year-over-year, driven by the success of our live sports marketplace and growth in programmatic guaranteed deals across expanding buyer relationships.

接下來來看季度業績，首先是營收明細。為了提供公平的比較，影片的年成長率經過調整，剔除了政治廣告支出。在此基礎上，全通路視訊總收入成長了 21%，凸顯了我們優質視訊產品組合的實力以及人工智慧驅動的優化技術在各種格式中日益普及。在這個類別中，CTV 再次同比增長超過 50%，這得益於我們的直播體育市場的成功以及透過不斷擴大的買家關係實現的程序化擔保交易的增長。

We monetized CTV inventory from over 90% of the top 30 global streamers. Omnichannel video contributed approximately 38% of total revenue in Q3 2025. Emerging revenue streams continued their high-growth trajectory, growing over 80% year-over-year while scaling to 10% of total revenue in the third quarter. This growth represents incremental durable revenue streams beyond our core sell-side platform capabilities. Most notably, year-over-year revenue from Activate grew over 100% in our Curation and data business Connect grew over 40%.

我們已將全球排名前 30 名的串流媒體平台中超過 90% 的 CTV 廣告資源貨幣化。2025年第三季度，全通路影片貢獻了約38%的總收入。新興收入來源持續保持高速成長勢頭，年增超過 80%，並在第三季佔總營收的 10%。這一成長代表著除我們核心銷售平台功能之外的持續性增量收入來源。最值得注意的是，Activate 的年收入年增超過 100%，而我們的內容策劃和數據業務 Connect 的營收則是年增超過 40%。

Based on the results we are seeing, we will continue to invest for incremental growth opportunities that diversify our revenue like the new AI-driven product capabilities for publishers that are already showing meaningful traction. The growth across these key secular areas helped offset the impact from lower spend by a large DSP we identified last quarter. Following our optimizations and integration adjustments by SPO partners, spend stabilized from this DSP in August and September, resulting in lower but steady run rate. As anticipated, display revenue was down minus 5% year-over-year and was the format most affected by the DSP impact. Excluding this DSP, display grew in the low single-digit percentages.

根據我們目前看到的結果，我們將繼續投資於能夠實現收入多元化的增量成長機會，例如面向出版商的全新人工智慧驅動產品功能，這些功能已經展現出顯著的市場吸引力。這些關鍵長期領域的成長有助於抵消我們上季度發現的一家大型DSP支出減少的影響。經過我們的優化和 SPO 合作夥伴的整合調整，8 月和 9 月來自該 DSP 的支出趨於穩定，從而實現了較低但穩定的運行率。正如預期的那樣，展示廣告收入比去年同期下降了 5%，並且是受 DSP 影響最大的廣告形式。除去DSP，顯示器市場佔有率僅以個位數百分比成長。

With respect to Q3 ad spend across the top 10 verticals, in aggregate, they grew in the single-digit percentages year-over-year. Health and fitness, personal finance and technology each increased over 15%. We saw softer trends in business and automotive, which declined single-digit percentages in the quarter. Ad spend from our mid-tier DSP partners grew over 25% year-over-year in Q3. Importantly, our AI-driven buyer platform is resonating well with performance-focused buyers across CTV, mobile app and e-commerce verticals, and we believe lays the foundation for sustainable growth in the quarters ahead.

就前 10 個垂直領域的第三季廣告支出而言，總體而言，它們同比增長了個位數百分比。健康與健身、個人理財和科技等領域的成長均超過 15%。商業和汽車產業的趨勢較為疲軟，本季這兩個產業的百分比均出現個位數下降。第三季度，我們來自中端DSP合作夥伴的廣告支出較去年同期成長超過25%。重要的是，我們的人工智慧驅動型買家平台在 CTV、行動應用和電子商務等垂直領域中，都受到了注重效果的買家的歡迎，我們相信這將為未來幾個季度的可持續成長奠定基礎。

We anticipate that new buyer relationships like Bliss and Mountain will bring incremental ad demand across our wide portfolio of verticals. Regionally, APAC and EMEA revenues grew plus 12% and plus 7%, respectively, offsetting a minus 14% decline in the Americas, which was primarily due to spend declines from a large DSP buyer.

我們預計，與 Bliss 和 Mountain 等新買家建立的合作關係，將為我們廣泛的垂直領域帶來增量廣告需求。從區域來看，亞太地區和歐洲、中東及非洲地區的收入分別成長了 12% 和 7%，抵消了美洲地區 14% 的下滑，這主要是由於一家大型 DSP 買家的支出下降所致。

Moving on to our operating priorities. We continue to invest and reallocate resources to the highest return areas of the business. As Rajeev noted, we're seeing strong growth across our DSP mix and within Activate as customers increase usage and new products drive incremental revenue. Our focus on generative AI is also improving operational agility, streamlining internal workflows and allowing us to redirect resources towards growth initiatives without adding structural costs.

接下來談談我們的營運重點。我們將繼續投資並將資源重新分配到業務中回報最高的領域。正如 Rajeev 指出的那樣，隨著客戶使用量的增加和新產品帶來的收入成長，我們的 DSP 產品組合以及 Activate 業務都實現了強勁成長。我們對生成式人工智慧的關注也提高了營運靈活性，簡化了內部工作流程，使我們能夠將資源重新分配到成長計畫中，而不會增加結構性成本。

This efficiency has allowed us to expand our sales team focus on buyers and grow spend from existing partners and onboard new partners. We are making great progress integrating with the fast-growing mid- to long-tail segment. And so far in 2025, we've added over 25 new DSP partners. All of these efforts help us counter the near-term headwinds from legacy DSPs and further diversify beyond the top five, as I described last quarter.

這種效率使我們能夠擴大銷售團隊的重點，專注於買家，增加現有合作夥伴的支出，並吸引新的合作夥伴。我們在與快速成長的中長尾市場進行整合方面取得了巨大進展。截至 2025 年，我們已經新增了 25 多個 DSP 合作夥伴。正如我上個季度所描述的那樣，所有這些努力都有助於我們應對來自傳統 DSP 的近期不利因素，並進一步實現前五名以外的多元化發展。

Core to our long-term strategy is being nimble in identifying opportunities and then executing rapidly to capitalize on them. We have achieved this while managing our costs and consistently delivering profits in many different environments. The foundation of this approach is expanding our capacity while driving down our unit costs. We processed approximately 87 trillion gross impressions in Q3, a 24% increase over last year and a 12% sequential gain versus Q2. Nearly 60% of total impressions were from CTV and mobile app inventory, highlighting our increasing focus on high engagement channels.

我們長期策略的核心是靈活地發現機遇，然後迅速執行以抓住機會。我們在控製成本的同時，在各種不同的環境下持續獲利，從而取得了這樣的成就。這種方法的根本在於擴大產能，同時降低單位成本。第三季度，我們處理了約 87 兆次總展示量，比去年同期成長 24%，比第二季度環比成長 12%。近 60% 的總曝光量來自 CTV 和行動應用廣告資源，這凸顯了我們越來越重視高互動管道。

Further, the increase in impressions is highly leveraged over our fixed cost base. Over a trailing 12-month basis, unit costs in the third quarter declined 19% over the comparable prior year period. In terms of operating expenses, our early investments in AI to drive operational efficiency are yielding measurable results. Year-to-date, every quarter, we have successfully kept our total operating expenses roughly flat at $51 million, while realigning investments to the areas that deliver the strongest ROI. This allows us to scale profitably even as we invest ahead of growth.

此外，曝光量的增加對我們的固定成本基礎具有很高的槓桿效應。過去 12 個月，第三季單位成本比去年同期下降了 19%。就營運費用而言，我們早期在人工智慧領域為提高營運效率所做的投資正在產生可衡量的結果。今年迄今為止，每個季度我們都成功地將總營運支出保持在 5,100 萬美元左右，同時將投資重新調整到能夠帶來最強投資回報率的領域。這使我們能夠在成長之前進行投資的同時，實現盈利性規模化發展。

For example, we increased our buyer-focused sales team by 19% in Q3 compared to the prior year, while the overall total headcount was flat. This disciplined approach enabled us to deliver our 38th straight quarter of adjusted EBITDA profitability. This is a track record few companies in our sector can match. Q3 adjusted EBITDA was $11.2 million or 16% margin, which included foreign exchange costs of approximately $1 million due to the weakening US dollar over the quarter. US GAAP net loss was minus $6.5 million or minus $0.14 per diluted share.

例如，與去年同期相比，我們第三季以買家為中心的銷售團隊增加了 19%，而總員工人數保持不變。這種嚴謹的方法使我們能夠連續第 38 個季度實現調整後 EBITDA 獲利。這是業內鮮有公司能夠匹敵的業績。第三季調整後 EBITDA 為 1,120 萬美元，利潤率為 16%，其中包括因本季美元走弱而產生的約 100 萬美元的外匯成本。以美國通用會計準則計算，淨虧損為負 650 萬美元，即每股稀釋虧損 0.14 美元。

Moving to cash and our capital allocation. Our balance sheet remains a core strategic advantage. In the third quarter, we generated $32.4 million in net operating cash flows and free cash flow of $22.8 million. In addition to efficient working capital management, there were two other factors that improved our cash flow for this period. The DSP that made changes in mid-2024 returned to growth in Q3, which favorably improved our DSOs. There was also a reduction in cash taxes paid because of the new federal tax bill that went into effect earlier this year.

轉向現金流和我們的資本配置。我們的資產負債表仍然是我們的核心策略優勢。第三季度，我們產生了 3,240 萬美元的淨經營現金流和 2,280 萬美元的自由現金流。除了高效率的營運資金管理外，還有兩個因素改善了我們這段期間的現金流。2024 年年中進行調整的 DSP 在第三季度恢復了成長，這對我們的 DSO 產生了積極影響。由於今年稍早生效的新聯邦稅收法案，現金繳納的稅款也有所減少。

To underscore our long-term ability to generate cash since the beginning of 2021 through Q3, we have generated over $390 million in net cash from operations and more than $215 million in free cash flow. We ended the quarter with $136.5 million in cash and zero debt.

為了強調我們自 2021 年初至第三季以來的長期現金創造能力，我們已從營運活動中產生了超過 3.9 億美元的淨現金流和超過 2.15 億美元的自由現金流。本季末，我們擁有1.365億美元現金，且無任何債務。

Given the strength, we continue to deploy our capital to maximize shareholder value. Since the inception of our repurchase program in February 2023 through the end of Q3, we have bought back 12.4 million Class A common shares for $180.6 million. We have $94.4 million remaining in our repurchase program authorized for the end of 2026. This program, combined with our ongoing investments in AI innovation reflects a balanced approach to capital allocation and a commitment to long-term shareholder value.

鑑於公司實力雄厚，我們將繼續投入資金，以實現股東價值最大化。自 2023 年 2 月啟動回購計畫以來，截至第三季末，我們已回購 1,240 萬股 A​​ 類普通股，總額達 1.806 億美元。我們截至 2026 年底的股票回購計畫還剩餘 9,440 萬美元。該計劃，加上我們對人工智慧創新領域的持續投資，體現了我們在資本配置方面採取的平衡方法以及對股東長期價值的承諾。

Turning to our Q4 outlook. We anticipate strong growth in secular areas of the business, including double-digit growth for CTV when excluding political advertising and 30%-plus growth for emerging revenues. As a reminder, Q4 2024 political advertising represented about 12% of revenue, nearly 80% of which was via CTV. In terms of latest trends in October, the typical holiday seasonal uptick thus far has been relatively muted for some consumer discretionary ad verticals such as food and drink and art and entertainment. Accordingly, we are taking a prudent approach to guidance based on the latest trends.

接下來展望第四季。我們預計業務的長期成長領域將強勁成長，其中 CTV（不包括政治廣告）將實現兩位數成長，新興收入將成長 30% 以上。提醒一下，2024 年第四季政治廣告收入約佔總收入的 12%，其中近 80% 是透過連網電視 (CTV) 實現的。就 10 月的最新趨勢而言，到目前為止，一些非必需消費品廣告垂直領域（如食品飲料、藝術娛樂等）的典型假日季節性增長相對平緩。因此，我們根據最新趨勢採取謹慎的指導方針。

We expect Q4 revenue to be in the range of $73 million to $77 million. As it relates to the large DSP that declined in July, we are assuming its associated revenue to be flat in Q4 relative to Q3. We anticipate Q4 operating expenses to be similar to Q3's level as AI-driven efficiencies continue to offset selective investments in our sales team. Q4 adjusted EBITDA is projected to be in the range of $19 million to $21 million, which also factors in continued weakness of the US dollar.

我們預計第四季營收將在 7,300 萬美元至 7,700 萬美元之間。至於 7 月出現下滑的大型 DSP，我們假設其第四季的相關營收與第三季持平。我們預計第四季度營運費用將與第三季水準相近，因為人工智慧驅動的效率提升將繼續抵消我們對銷售團隊的選擇性投資。預計第四季度調整後 EBITDA 將在 1,900 萬美元至 2,100 萬美元之間，其中也考慮了美元持續疲軟的因素。

For the full year, we expect revenue to be in the range of $276 million to $280 million and adjusted EBITDA to be in the range of $53 million to $55 million, inclusive of more than $5 million of estimated negative FX impact. We are maintaining our full year CapEx projection at $15 million, which is year-over-year reduction made possible by AI-driven optimization efforts.

我們預計全年營收將在 2.76 億美元至 2.8 億美元之間，調整後 EBITDA 將在 5,300 萬美元至 5,500 萬美元之間，其中包括超過 500 萬美元的預期負面外匯影響。我們維持全年資本支出預期為 1500 萬美元，比前一年有所減少，這得益於人工智慧驅動的優化工作。

In closing, our Q3 results highlighted the durability of our financial model and validate the progress we are making behind our business transformation. Looking ahead, we're confident in our robust multifaceted strategy. Our AI-first end-to-end platform is driving measurable results. We're adding new revenue streams, expanding our SPO relationships and rapidly diversifying our DSPs in line with future growth opportunities in commerce, performance CTV and mobile app.

總之，我們第三季的業績突顯了我們財務模式的穩健性，並驗證了我們在業務轉型方面的進展。展望未來，我們對穩健的多面向策略充滿信心。我們以人工智慧為先導的端到端平台正在帶來可衡量的成果。我們正在增加新的收入來源，擴大我們的 SPO 關係，並根據未來在商業、效果 CTV 和行動應用方面的成長機會，迅速實現 DSP 的多元化。

With respect to potential remedies in the Google AdTech antitrust trial, we continue to believe that any remedies that level the competitive playing field, whether structural, behavioral or both will benefit the open Internet and PubMatic. In 2026 and beyond, as revenue growth reaccelerates, we anticipate margin expansion at both the gross and adjusted EBITDA levels because of our efficient and leveraged business.

關於 Google AdTech 反壟斷審判中可能採取的補救措施，我們仍然認為，任何能夠創造公平競爭環境的補救措施，無論是結構性的、行為性的還是兩者兼而有之的，都將有利於開放的互聯網和 PubMatic。2026 年及以後，隨著營收成長再次加速，我們預計由於我們高效且槓桿化的業務，毛利和調整後 EBITDA 利潤率都將擴大。

Our decade-plus experience owning and operating our global private cloud infrastructure has given us several advantages. First, it has enabled us to expand capacity while at the same time, progressively reduce our rate of CapEx and drive down unit costs through optimization initiatives.

我們在擁有和營運全球私有雲基礎設施方面擁有十多年的經驗，這為我們帶來了許多優勢。首先，它使我們能夠在擴大產能的同時，透過優化措施逐步降低資本支出率並降低單位成本。

Second, it's allowed us to invest early on in next-generation technology with NVIDIA. Today, our infrastructure is a clear differentiator with investments well ahead of our peers that will continue to drive efficiencies and business impact. We are also continuing to leverage AI to drive efficiency and increase our team's productivity.

其次，這使我們能夠及早投資英偉達的下一代技術。如今，我們的基礎設施已成為一項明顯的差異化優勢，我們在基礎設施方面的投資遠遠領先於同行，這將繼續提高效率並帶來業務影響。我們也持續利用人工智慧來提高效率，提升團隊的生產力。

We anticipate total headcount will remain relatively flat in 2026, while investments supporting our fastest-growing areas of our business will increase through internal reallocations. And finally, we are also laser-focused on free cash flow generation and aim to increase cash flow next year supported by further working capital improvements and incremental AI-driven efficiencies. Collectively, we believe our efforts will drive a return to double-digit revenue growth in the future.

我們預計 2026 年員工總數將保持相對穩定，同時，透過內部資源重新分配，對業務成長最快領域的投資將會增加。最後，我們也將重點放在自由現金流的產生上，目標是在明年透過進一步改善營運資本和逐步提高人工智慧驅動的效率來增加現金流。我們相信，透過共同努力，未來公司營收將恢復兩位數成長。

With that, I'll turn the call over to Stacie for questions.

接下來，我將把電話交給史泰西，她可以回答一些問題。

(Event Instructions)

（活動須知）

Andrew Marok, Raymond James.

Andrew Marok，Raymond James。

Rajeev, if you could maybe expand a bit on the topic of SPO and some of the recent moves companies like the Trade Desk have made kind of leaning into OpenPath, launching OpenAds and declaring all SSPs as resellers. But then on the other hand, the two of you are collaborating on that DYD management API. So I guess since we last spoke last quarter, how would you characterize the state of play there? And then I have a follow-up for Steve.

Rajeev，如果你能就 SPO 以及像 Trade Desk 這樣的公司最近採取的一些舉措（例如轉向 OpenPath、推出 OpenAds 並將所有 SSP 都宣佈為經銷商）再詳細闡述一下就好了。但另一方面，你們兩個正在合作開發 DYD 管理 API。所以我想問，自從我們上個季度上次交談以來，您如何評價那裡的現狀？然後我還有一個問題想問史蒂夫。

Sure. Yes. Thanks, Andrew, for the question. So yes, let me first just clarify on the reseller kind of noise that's out there. So the industry standard definition, which has been with the industry for over a decade, is that a reseller is the term for when inventory flows from a publisher to an intermediary and then to an SSP.

當然。是的。謝謝安德魯的提問。所以，首先讓我澄清一下目前市面上關於經銷商的各種傳言。因此，業內標準定義（該定義已在業內沿用十多年）是，經銷商是指庫存從出版商流向中間商，然後再流向 SSP 的過程。

And an intermediary can be another exchange or some sort of aggregator of inventory. In contrast, direct is when inventory flows from the publisher directly to the SSP. PubMatic is a platform for direct inventory monetization. Reselling is not our business. We are a direct connection to publishers, and that's how we're able to provide significant incremental value to publishers and to buyers.

而中間商可以是另一個交易所或某種庫存聚合商。相比之下，直接銷售是指庫存直接從出版商流向SSP。PubMatic是一個直接實現庫存貨幣化的平台。轉售不是我們的業務。我們與出版商直接對接，因此能夠為出版商和買家提供顯著的增值服務。

And I think it's pretty clear we provide values in ways -- value in ways that DSPs do not. Yield optimization is a key example of that. So Trade Desk, in particular, has been very clear that they are not in the yield optimization business. And so a publisher needs to have a yield function in place in order to maximize their revenue, which is core to what we do. At the same time, we're also providing value to Trade Desk and others, right, who, as you noted, Andrew, they're relying on us to improve deals with our price discovery and provisioning API integration announcement.

而且我認為很明顯，我們以數位訊號處理器（DSP）無法做到的方式提供價值。產量優化就是一個典型的例子。因此，Trade Desk 尤其明確表示，他們不從事收益優化業務。因此，出版商需要建立收益管理機制，以最大限度地提高收入，而這正是我們工作的核心。同時，我們也為 Trade Desk 和其他公司創造了價值，正如你所指出的，Andrew，他們依靠我們透過價格發現和供應 API 整合公告來改善交易。

So I think what you can see here is that the ecosystem is multifaceted, certainly complex. Our focus is really on taking that direct integrations with publishers that we have, all of that inventory flowing through our platform, the data the audiences and really creating value for buyers in such a way that they're able to generate increased return on ad spend, increased ROI, which causes them to then spend more on our platform. And then that, in turn, generates higher yield for our publishers. And so that's really the core and the focus that we have. And what we're finding with Activate and other capabilities on our platform, a lot of the focus on AI that we talked about in the prepared remarks is that we have a lot of opportunity coming at it from the sell side of the ecosystem based on the auctions and the data and our close integrations with publishers to be able to add value to both the buyers and to the publishers.

所以我覺得大家可以看出，生態系統是多方面的，當然也很複雜。我們的重點在於利用我們與出版商的直接整合，利用我們平台上的所有廣告資源、數據和受眾，真正為買家創造價值，使他們能夠提高廣告支出回報率，從而促使他們在我們的平台上投入更多資金。而這又會為我們的出版商帶來更高的收益。所以，這才是我們真正的核心與關注。我們發現，透過 Activate 和我們平台上的其他功能，以及我們在準備好的發言稿中提到的許多對人工智慧的關注，我們從生態系統的賣方方面獲得了許多機會，這些機會基於拍賣、數據以及我們與出版商的緊密合作，能夠為買家和出版商都增加價值。

And then maybe, Steve, if I could, on the COGS point, can you just expand a little bit on the ability to drive that unit cost leverage there and how we should think about that line either on an absolute dollar basis or a percentage of revenue going forward?

然後，史蒂夫，如果可以的話，關於銷貨成本這一點，你能否詳細解釋一下如何提高單位成本槓桿，以及我們應該如何考慮這一指標，是以絕對美元為基礎，還是以收入百分比為基礎？

Sure. Well, good to reconnect, Andrew. So from our perspective, we, for many years, have been focused on owning and operating our own infrastructure and have done it very successfully, as you noted, driving down unit costs. We've consistently done that for a decade, often double-digit unit growth reductions. And so from our perspective, '26 is no different.

當然。很高興再次聯繫，安德魯。因此，從我們的角度來看，多年來，我們一直專注於擁有和經營我們自己的基礎設施，正如您所指出的，我們在這方面做得非常成功，並降低了單位成本。十年來，我們一直堅持這樣做，經常出現兩位數的單位成長率下降。所以從我們的角度來看，2026 年也沒有什麼不同。

Other than that, we continue to leverage AI more and more to drive optimizations. As you saw in our prepared comments, we increased the number of gross impressions processed by over 20% in the quarter. We didn't do that by just throwing a lot of CapEx at it. We did it through software and AI. And so that basic process is going to continue, but we have more tools and more ability to do that. So I would expect in the future, as our revenue reaccelerates, we're going to increase our gross margin. It's going to be a function of revenue and managing our costs.

除此之外，我們也將繼續越來越多地利用人工智慧來推動優化。正如您在我們準備好的評論中所看到的，本季我們處理的總曝光量增加了 20% 以上。我們並不是單純地投入大量資本支出就實現了這個目標。我們透過軟體和人工智慧實現了這一點。因此，這個基本過程將會持續，但我們擁有更多的工具和能力來實現這一目標。因此，我預計未來隨著收入的重新成長，我們的毛利率也會提高。這取決於收入和成本控制情況。

Matt Swanson, RBC.

Matt Swanson，RBC。

Obviously, a super strong quarter for CTV growth ex-political. And you called out a lot of the drivers there, right, the great growth in the live sports, maybe some of that expansion of mid-market DSPs and the 90% coverage now top 30 streamers. Could you just give us kind of a little more color maybe over the past year, just what sort of evolution you've seen in the CTV environment and kind of how you're investing to grow or continue to grow next year?

顯然，撇開政治因素不談，CTV（連網電視）本季成長非常強勁。你指出了很多驅動因素，對吧？例如現場體育賽事的蓬勃發展，中端市場數位服務供應商的擴張，以及目前排名前 30 的串流媒體服務商 90% 的覆蓋率。能否請您更詳細地介紹一下過去一年來您在 CTV 領域觀察到的發展變化，以及您明年將如何投資以實現成長或繼續成長？

Sure. Yes. Thank you, Matt, for the question. So CTV has been obviously a very strong growth area for us, and we've seen tremendous scale really building from zero organically several years ago to where we are today. And I think there's a couple of trends that are happening.

當然。是的。謝謝馬特的提問。因此，CTV 顯然一直是我們成長非常強勁的領域，我們看到了巨大的規模，從幾年前的零開始，自然而然地發展到今天的規模。我認為目前存在幾種趨勢。

First of all, we are working with more and more premium publishers, right? So this quarter, we shared over 90% of the top 30 globally. We added some new fast streamers to the Future Today and Local Now. That builds on our base of premium inventory from existing publishers like Roku and Paramount and NBC. At the same time, there's huge growth in the advertiser mix in CTV.

首先，我們正在與越來越多優質的出版商合作，對吧？因此，本季我們在全球前 30 名中佔據了 90% 以上的份額。我們為 Future Today 和 Local Now 添加了一些新的快速串流媒體。這建立在我們現有來自 Roku、派拉蒙和 NBC 等發行商的優質庫存基礎之上。同時，連網電視廣告主的組成也出現了巨大的成長。

So whereas traditional TV has a couple of hundred, maybe 500 advertisers in aggregate that are the lion's share of ad budgets. With streaming TV, what we see is that there's tens of thousands of advertisers reaching now into the hundreds of thousands, and that number is growing very quickly. And as we have focused on going after mid-market focused DSPs, many of them are focused on small and medium businesses like Mountain or TV Scientific where they're focused on performance advertising, again, some of those same folks, there's a lot more dollars that we're able to bring on to the platform. And so that's creating another layer of growth.

因此，傳統電視廣告商加起來可能有兩到五百家，佔了廣告預算的絕大部分。我們看到，隨著串流媒體電視的發展，廣告商的數量從成千上萬增加到數十萬，而且這個數字還在快速增長。由於我們一直專注於瞄準中端市場DSP，其中許多DSP專注於像Mountain或TV Scientific這樣的中小企業，這些企業專注於效果廣告，同樣，我們能夠為這些企業帶來更多的資金。因此，這就創造了另一層成長空間。

And then third is we've really leaned into AI, as you could tell from the past calls and also the prepared remarks today. all three layers of the technology stack. And I think that's really unlocking incremental budgets as well. So live sports is just one example of that. Actually, one of our first generative AI creative tools we launched last year, which was focused on helping publishers bring in more political ad spend by scanning political ad creatives.

第三，我們已經真正開始重視人工智慧，正如你從過去的電話會議和今天準備好的演講稿中看到的那樣。這涵蓋了技術棧的所有三個層面。我認為這確實也釋放了增量預算的潛力。所以，現場體育賽事只是其中的一個例子。事實上，我們去年推出的首批生成式 AI 創意工具之一，旨在透過掃描政治廣告創意來幫助出版商吸引更多政治廣告支出。

And so we're continuing to apply that technology for other use cases. So there's a lot of other verticals like pharma, for instance, health and wellness, where there's sensitive categories, but we're able to generate unlock of spend through AI. So we think it's a really obviously exciting area for us. We're going to continue to focus global investment in this area. And as we bring more buyers onto our platform via Activate, via curation capabilities and now with the ad CP launch, we think there's a long runway of growth ahead of us for CTV.

因此，我們將繼續把這項技術應用於其他用例。因此，還有很多其他垂直領域，例如醫藥、健康和保健，這些領域都存在敏感類別，但我們能夠透過人工智慧釋放消費潛力。所以我們認為這顯然是一個非常令人興奮的領域。我們將繼續把全球投資重點放在這個領域。隨著我們透過 Activate、內容策劃功能以及現在的廣告 CP 的推出，吸引更多買家加入我們的平台，我們認為 CTV 的未來還有很長的成長之路。

Maybe following up on your commentary on GenAI and some of the Agentic. Obviously, we've been hearing a lot from a lot of players in the space around these two themes. Can you just talk a little bit about kind of the right to win and how you help your stakeholders understand the value creation versus the noise around these themes. And maybe as you said, with the AdCP, it's not about winning yourselves or creating a TAM expanding ecosystem around those themes.

或許可以接著你對 GenAI 和一些 Agentic 的評論繼續探討。顯然，我們已經聽到了很多業內人士圍繞這兩個主題的討論。您能否談談“獲勝的權利”，以及您如何幫助利益相關者理解價值創造與圍繞這些主題的各種噪音之間的區別？或許如你所說，對 AdCP 而言，重點不在於贏得市場，也不在於圍繞這些主題創造一個不斷擴大 TAM 的生態系統。

Yes. So I think -- let's take each of those in turn. So first of all, I think the right to win comes from a couple of things. First of all, it's owning all of our own infrastructure. And so what that means is that we're in a unique position to be able to deploy cutting-edge AI technology, AI foundational infrastructure like the partnership that we announced with NVIDIA.

是的。所以我想——讓我們逐一來看。首先，我認為獲勝的權利是來自兩方面。首先，我們要擁有自己所有的基礎設施。這意味著我們擁有獨特的優勢，可以部署尖端的人工智慧技術和人工智慧基礎架構，例如我們與 NVIDIA 宣布的合作關係。

And that came as a result of multiple years of collaboration. So it's not just something you can wake up and do all of a sudden. But to be able to do that, you have to really own and control and be in a position to manage all of that infrastructure. So I think that's the first right to win is our long-term capability set there and expertise. Second is you got to have the transactions and the data flowing on your platform.

這是多年合作的成果。所以這不是你能突然醒來就去做的事情。但要做到這一點，你必須真正擁有和控制所有這些基礎設施，並且能夠管理它們。所以我認為，贏得比賽的首要條件是我們在該領域的長期能力和專業知識。其次，你的平台上必須要有交易和數據流動。

So an AI algorithm is only useful to your customers and only relevant to them if you have the scale and ability to transact. And of course, we have that with 1 trillion or so daily ad impressions flowing on our platform, almost 2,000 publishers, depth in CTV and mobile app in display and so on and so forth. So I think that's the second key for the right to win.

所以，只有當你擁有足夠的規模和能力進行交易時，人工智慧演算法才能對你的客戶有用，才能與他們相關。當然，我們平台每天的廣告展示量約為 1 兆次，擁有近 2000 家發布商，在 CTV 和行動應用程式展示廣告方面擁有深厚的實力等等。所以我認為這是贏得比賽的第二個關鍵。

And then I think the third is a demonstrated ability to innovate and to really build solutions, not just talk about them. And I think what you've seen from us over the last year is that we are significantly ahead of our competition. We have launched 17 agents in our publisher platform already using AI. I think one of our competitors shared that they're planning to build their first agent, right? So you can see that's a one to two year advantage that we have in terms of track record of execution.

我認為第三點是展現出創新能力和真正建立解決方案的能力，而不僅僅是空談。我認為過去一年來，大家已經看到我們遠遠領先競爭對手。我們已在出版商平台上利用人工智慧推出了 17 個代理商。我想我們的一位競爭對手曾透露，他們計劃建立他們的第一個代理系統，對嗎？所以你可以看出，我們在執行記錄方面擁有一到兩年的優勢。

And so what that means is that when we're launching things like AdCP, the ad context protocol, which is managing workflow via AI we are in a position to be able to go talk to both publishers and buyers about here's how you get started. Here are the first use cases to implement, and these things are available on our platform today. And so I think that's a key part of creating value within the ecosystem is being able to participate in those early transactions being on the frontier and then benefiting from the groundswell of growth that we see in front of us.

這意味著，當我們推出 AdCP（廣告上下文協議）等功能時（該協議透過 AI 管理工作流程），我們就可以與發布商和買家討論如何入門。以下是首批需要實現的用例，這些功能目前已在我們的平台上提供。所以我認為，在生態系統中創造價值的關鍵在於能夠參與早期交易，站在前沿，然後從我們眼前看到的成長浪潮中受益。

Shweta Khajuria, Wolfe.

Shweta Khajuria，Wolfe。

I have a follow-up on the first question on OpenPath, not specifically just OpenPath, but are you seeing, Rajeev, any trends that would suggest that perhaps Trade Desk is increasingly going direct and you are getting impacted by -- not only just as a reseller, but in terms of the impressions volume. Are you seeing any impact to your business in general? And is it related to them actually launching that Kokai platform? I would think not, but there is this concern in the industry. So if you could please perhaps comment on that to correct that, that would be great.

關於 OpenPath 的第一個問題，我有一個後續問題。不僅僅是 OpenPath，Rajeev，你是否看到任何趨勢表明 Trade Desk 可能越來越傾向於直接銷售，而你不僅作為經銷商受到了影響，而且在展示量方面也受到了影響。您的企業整體是否受到任何影響？這和他們實際推出 Kokai 平台有關嗎？我個人認為不會，但業界確實有這種擔憂。如果您能就此發表評論並予以糾正，那就太好了。

Sure. Yes, absolutely. So first of all, good news is we are a platform for direct inventory. So regardless of what label anybody wants to put on anything, we know that we are working directly with premium publishers and there is not a more efficient way for buyers to access the inventory than what is coming through our platform. So I think underlying your question, the new Trade Desk platform, Kokai does evaluate and buy media differently from what we have seen.

當然。是的，絕對的。首先，好消息是我們是一個直接庫存平台。所以，無論別人想給任何東西貼上什麼標籤，我們都知道我們正在直接與優質出版商合作，對於買家來說，沒有比透過我們的平台獲取庫存更有效的方式了。所以我覺得，從你的問題來看，新的交易平台 Kokai 對媒體的評估和購買方式與我們以往所見有所不同。

And so we took two important steps over the course of Q3, resulting in spend from them stabilizing in August and September, as Steve mentioned. So the first is that we revised our machine learning algorithms to ensure we're sending an optimal mix of inventory to them. That work is largely behind us. So we're always doing some level of continuous optimization to drive more spend. And then second, we worked with our SPO supply path optimization partners to help them configure their seat in the DSP to ensure that they're getting the benefit of their SPO relationships with us.

因此，我們在第三季採取了兩項重要措施，正如史蒂夫所提到的那樣，使得這方面的支出在 8 月和 9 月趨於穩定。首先，我們修改了機器學習演算法，以確保我們向他們發送最佳的庫存組合。那項工作基本上已經完成了。因此，我們一直在進行一定程度的持續優化，以推動更多消費。其次，我們與 SPO 供應鏈路徑優化合作夥伴合作，幫助他們配置在 DSP 中的席位，以確保他們能夠從與我們建立的 SPO 關係中受益。

And that work is largely completed as well. So we have agency marketplaces set up with pretty much every major Holdco in different parts of the world, in many different parts of the world. And so these agencies had to make changes over the last couple of months to ensure that their marketplaces stay intact and that they continue to get the performance and efficiency that they're seeking. And of course, these marketplaces are critical to the agency's media buying offerings that they provide to their advertiser clients.

這項工作也基本完成了。因此，我們在世界各地與幾乎所有主要的控股公司都建立了代理市場。因此，在過去的幾個月裡，這些機構不得不做出改變，以確保他們的市場保持完整，並繼續獲得他們所追求的績效和效率。當然，這些市場對於代理商向廣告客戶提供的媒體購買服務至關重要。

So we've shared publicly, for instance, we power WPP's premium marketplace. And so that's built on top of our SSP. And so making sure that they're receiving their SPO data, workflow efficiency benefits is key to their being able to continue to offer that in market. Similarly, given we're a leader -- market leader in curation, we work with our curation partners to ensure that their campaigns continue to run via our SSP and that they get the ROI, the transparency and the control that caused them to choose to work with us in the first place. So I think Trade Desk shared that, I believe, 85% of their clients are now up and live on Kokai.

例如，我們已經公開表示，我們為 WPP 的高端市場提供技術支援。所以它是建立在我們SSP之上的。因此，確保他們能夠收到 SPO 數據，從而提高工作流程效率，是他們能夠繼續在市場上提供這些服務的關鍵。同樣，鑑於我們是內容策劃領域的領導者——市場領導者，我們會與內容策劃合作夥伴合作，確保他們的推廣活動繼續透過我們的SSP平台進行，並確保他們獲得投資回報率、透明度和控制權，而這些正是他們當初選擇與我們合作的原因。所以我認為 Trade Desk 分享過，據我所知，他們 85% 的客戶現在已經上線 Kokai 平台。

So we think probably the bulk of this movement is behind us. But that's a key part of why we're also very rapidly diversifying our DSP mix. As the market evolves to a more fragmented DSP landscape, we're finding significant success with 25%-plus year-over-year growth with mid-market DSPs on our platform in Q3.

所以我們認為這場運動的大部分可能已經過去了。但這正是我們正在迅速實現 DSP 產品組合多樣化的關鍵原因之一。隨著市場向更加分散的 DSP 格局發展，我們在第三季度透過我們的平台實現了中端市場 DSP 的顯著成功，年成長超過 25%。

Matt Condon, JMP.

Matt Condon，JMP。

My first one is just there's been a lot of talk of just the impact across the open web on publisher traffic, just given these AI platforms that are taking increased share. Can you just talk about what you're seeing across your publisher base as far as search traffic?

我的第一個問題是，鑑於人工智慧平台正在佔據越來越大的市場份額，人們一直在討論開放網路對出版商流量的影響。能否談談您目前在出版商群體中觀察到的搜尋流量情況？

Yes, absolutely. So we've seen, I would say, a fairly limited impact. And I would say, just stepping back, we believe the exposure overall in our business is limited to a single-digit percentage of revenue if there was zero search traffic going to our publishers, and we took no steps to mitigate it. So that's probably a high-end kind of watermark of potential impact. So we shared in our comments today that roughly 60% of the impressions that we are processing are for CTV and mobile app.

是的，絕對的。所以，我認為，我們看到的影響相當有限。我想說，退一步講，我們認為，如果沒有任何搜尋流量流向我們的出版商，而且我們也沒有採取任何措施來減輕損失，那麼我們業務的整體風險敞口將僅限於收入的個位數百分比。所以這可能是衡量潛在影響力的高階標竿。因此，我們在今天的評論中提到，我們處理的廣告展示量中約有 60% 來自 CTV 和行動應用程式。

And so of course, those are unaffected by AI search. Now of the remaining business, which is browser-based where search is relevant, industry data indicates that search referral traffic is roughly 15% with the rest of publishers' traffic coming from either social or direct navigation. And given that we work in the ahead of the market with the top publishers rather than the long tail, I would expect that 15% number to actually be even a little bit lower. And so if you take, okay, 40% of the impressions, 15% impact at the high end, that gets you to a mid-single-digit percentage potential impact.

因此，這些當然不受人工智慧搜尋的影響。現在，在剩餘的業務中（即基於瀏覽器且與搜尋相關的業務），行業數據顯示，搜尋推薦流量約佔 15%，其餘發布商的流量則來自社交媒體或直接導航。鑑於我們與市場領先的頂級出版商合作，而不是與長尾出版商合作，我預計 15% 這個數字實際上還會更低一些。因此，如果你獲得 40% 的曝光量，最高可達 15% 的影響，那麼你也就只能獲得個位數百分比的潛在影響。

Again, if we had no mitigating steps that we took. But actually, there's a lot that we can do in terms of bringing -- continuing to bring on board more CTV, more mobile app, more commerce impressions and the like. I think the other thing that we're seeing is the offensive opportunity. which is that there's a growing number of AI search experiences that consumers are spending more and more time on. So for instance, Gannett launched a solution where users on their properties can search the archives of Gannett articles and get answers to questions.

再說一遍，如果我們沒有採取任何緩解措施的話。但實際上，在引進更多連網電視、更多行動應用、更多電商展示等方面，我們還有很多工作要做。我認為我們看到的另一個趨勢是進攻機會，那就是越來越多的消費者花越來越多的時間在人工智慧搜尋體驗上。例如，Gannett推出了一項解決方案，用戶可以在其平台上搜尋Gannett文章的存檔，並獲得問題的答案。

And so that, I think, is actually a growing canvas of inventory of opportunity for us as consumers get more and more used to that AI search kind of consumption behavior, then they're looking for that from many of their traditional content partners where they consume content. And so we think that's a new tailwind that is emerging in the business.

因此，我認為，隨著消費者越來越習慣於人工智慧搜尋式的消費行為，他們也開始從許多傳統的內容合作夥伴那裡尋求這種消費方式，這實際上為我們創造了一個不斷擴大的機會空間。因此，我們認為這是商業領域正在出現的一股新的順風。

And then my second one is just on your O&O infrastructure and just the differentiation there and your partnership that you announced with NVIDIA, just the 5x speed improvements in bid response times, just can you just talk about the structural difference that's allowing? And is it allowing just win rates and auctions to be higher? And just talk about the differentiation there and how that can drive sustainable growth as we think about 2026.

我的第二個問題是關於你們的自有基礎設施，以及你們與 NVIDIA 的合作，特別是競價響應時間 5 倍的提升，你們能否談談這種結構性差異是如何實現的？這樣做是否只是提高了得標率和拍賣率？然後談談其中的差異化，以及當我們展望 2026 年時，這種差異化如何能推動永續成長。

Yes, absolutely. So actually, the good news is that there's multiple ways that having this kind of infrastructure cooperation collaboration with NVIDIA can benefit our business. And it really is a collaboration, not just in hardware but also in software. I think NVIDIA themselves says that I think it's over half, maybe over 60% of their revenue comes from software solutions, right? So obviously, they're well known for hardware, but it's really a combination of hardware and software.

是的，絕對的。所以實際上，好消息是，與 NVIDIA 進行這種基礎設施合作可以從多個方面使我們的業務受益。而且這確實是一次合作，不僅體現在硬體方面，也體現在軟體方面。我認為英偉達自己也說過，他們超過一半，甚至可能超過 60% 的收入都來自軟體解決方案，對吧？很顯然，他們以硬體聞名，但實際上，它是硬體和軟體的結合。

So I'll give you kind of three specific examples. So we use NVIDIA GPUs to power real-time ad decisioning in very low latency environments. So think about things like connected TV or live sports, where if we can process ad transactions faster, if we can cut latency, then that leads to fewer time outs, more bids in the auction on our platform and then more opportunities for us to win with the publisher. And so that boosts outcomes, ROI for both advertisers and publishers, also boost our revenue.

所以我給你舉三個具體的例子。因此，我們使用 NVIDIA GPU 來支援在極低延遲環境下的即時廣告決策。想想網路電視或體育直播之類的東西，如果我們能更快地處理廣告交易，如果我們能減少延遲，那麼就會減少超時，在我們的平台上進行更多競價，然後我們就能有更多機會贏得發布商的青睞。這樣一來，就能提高廣告商和出版商的投資報酬率，同時也能提高我們的收入。

The second example is using NVIDIA Triton inference servers, where we use a specific hardware software implementation with them for traffic shaping. And traffic shaping is a very important and critical role that anybody on the sell side plays, which is to figure out which of the roughly 1 trillion ad impressions we have per day we should send to each particular DSP. Some DSPs, we might send tens of billions, some might be single-digit billions, some might be hundreds of billions of ad impressions.

第二個例子是使用 NVIDIA Triton 推理伺服器，我們將其與特定的硬體軟體實作結合使用，以進行流量整形。流量塑造是銷售方每個人都扮演的一個非常重要且關鍵的角色，其目的是確定我們應該將每天大約 1 兆次的廣告展示中的哪些展示發送到每個特定的 DSP。有些DSP平台，我們可能會發送數百億次廣告展示；有些平台可能會發送數十億次廣告展示；有些平台可能會發送數千億次廣告展示。

So it's really important that we pick the right ones. And for every DSP, there's got to be a different decision set of calculus based on the types of advertisers and campaigns that are in their platform. And of course, these decisions have to be made within milliseconds of the publisher requesting a bit from us.

所以，選對人真的非常重要。對於每個 DSP 來說，根據其平台上的廣告主和廣告活動的類型，都必須有一套不同的決策計算方法。當然，這些決定必須在發布商向我們要求資料後的幾毫秒內做出。

And then third is in the reporting area. So we're using an NVIDIA software accelerator for Apache Spark, which allows us to streamline and improve the speed at which we're able to process data. And that, in turn, allows for smarter optimization across a wide array of different use cases. So those are, I think, hopefully just three tangible examples of how our NVIDIA partnership is really leading to the leadership that we talked about in the prepared remarks, in particular, at the infrastructure layer, which then allows us to drive benefit in the application and transaction layers of the AI stack.

第三點是報告領域。因此，我們使用 NVIDIA 軟體加速器來加速 Apache Spark，這使我們能夠簡化流程並提高資料處理速度。而這反過來又使得在各種不同的使用情境中實現更聰明的優化成為可能。所以，我認為，這三個具體的例子可以說明，我們與 NVIDIA 的合作關係是如何真正引領我們走向我們在準備好的發言稿中提到的領導地位的，尤其是在基礎設施層面，這使我們能夠在 AI 堆疊的應用層和交易層帶來效益。

Rob Coolbrith, Evercore.

Rob Coolbrith，Evercore。

Rajeev, I wanted to go back to this 5x faster and big response unlocking optimization strategies previously impossible to programmatic rates. Can you unpack that? And assuming you have a very substantial lead in accelerated computing in this space, are there counterparties on the demand side who can take advantage of that? Or do you think you need to take on a bigger role either in optimizing demand or hosting demands logic and optimization to sort of fully take advantage of those capabilities. There's something more akin to what the walled gardens do maybe in terms of their highly vertically integrated supply chains. And Steve, I just wanted to maybe get a finer point on this Trade Desk issue. Just given current trends, do you see the potential to maybe see growth alongside their growth in the back half of '26?

Rajeev，我想回到以前無法實現的、速度提升 5 倍、響應速度更快、響應更迅速的優化策略，實現程式化價格。你能解釋一下嗎？假設您在加速運算領域擁有非常大的領先優勢，那麼需求方是否有可以利用此優勢的交易對手？或者，您是否認為您需要在最佳化需求或託管需求邏輯和最佳化方面發揮更大的作用，以便充分利用這些功能？就其高度垂直整合的供應鏈而言，這可能更類似於那些封閉式生態系統所做的事情。史蒂夫，我只是想就交易台這個問題再深入探討。就目前的趨勢而言，您認為在 2026 年下半年，貴公司是否有可能與他們的成長同步成長？

Yes. Thanks, Rob. So yes, let me start with your first question, and then I'll hand it over to Steve. So we absolutely do see opportunity to better leverage our infrastructure through vertical integration. And I'll give you two examples of that.

是的。謝謝你，羅伯。好的，那我就先回答你的第一個問題，然後再把問題交給史蒂夫。因此，我們確實看到了透過垂直整合來更好地利用我們基礎設施的機會。我舉兩個例子來說明這一點。

So one is, as we work more and more with mid-market focused DSPs who themselves tend to be smaller, what we find is that there's a lot of opportunity for us to use our platform to help them compete more effectively. And so what I mean by that is we have huge amounts of data on our platform from these 1 trillion daily ad impressions. A typical mid-market DSP, they may see 5% to 10% of the traffic that a very large DSP like a Google DV 360 would see. And so these mid-market DSPs, because they're smaller in nature, they don't have access to all of that same data. At the same time, they also don't have access to all of the performance aspects of the auction that we're running on behalf of the publisher.

因此，一方面，隨著我們與規模較小的、專注於中端市場的DSP合作越來越多，我們發現，我們有很多機會利用我們的平台來幫助他們更有效地參與競爭。我的意思是，我們平台上擁有來自每天 1 兆次廣告展示的大量數據。典型的中型 DSP 的流量可能只有像 Google DV 360 這樣的大型 DSP 的 5% 到 10%。因此，這些中型市場 DSP 由於規模較小，無法取得所有這些數據。同時，他們也無法了解我們代表出版商進行的拍賣的所有績效方面。

And so there's opportunities for us, which we are working very hard on to use our platform to help those DSPs derive better performance, better targeting and ultimately better ROI. And that plays very closely with the infrastructure that we've deployed.

因此，我們有機會利用我們的平台來幫助這些 DSP 獲得更好的績效、更精準的定位，並最終獲得更好的投資報酬率。我們正在為此努力。而這與我們已部署的基礎設施密切相關。

So that's one example, Rob, of where I think vertical integration where we can do more than we have done traditionally for a legacy DSP. And I think these mid-market DSPs because they're growing quickly, they're very hungry for that kind of collaboration and our ability to help them solve problems. And then the second is with our Activate solution, where buyers are buying directly in our SSP. And so here, that faster processing of bids that are inferencing, all of these things, they help make Activate a very effective solution. And we're seeing that play out in terms of the growth, up 4x year-over-year.

所以，羅布，這就是一個例子，我認為垂直整合可以讓我們為傳統 DSP 做更多我們以前從未做過的事情。我認為這些中型市場 DSP 成長迅速，他們非常渴望這種合作以及我們幫助他們解決問題的能力。第二種方式是使用我們的 Activate 解決方案，買家可以直接在我們的 SSP 中購買。因此，更快的投標處理速度以及所有這些因素，都有助於使 Activate 成為一個非常有效的解決方案。我們看到，這種成長動能正在顯現，年增了 4 倍。

Campaign numbers are growing on a similar basis in terms of the number of campaigns run. So we think there's a lot of benefit from vertical integration. And as you said, Rob, that is somewhat akin to what the walled gardens do. And of course, it's no secret that they're very good at driving performance, and we think that vertical integration is a key part of that. I'll turn it over to Steve now.

競選活動的數量也以類似的方式增加。所以我們認為垂直整合有很多好處。正如你所說，羅布，這有點像圍牆花園的作用。當然，他們非常擅長提升業績，這已不是什麼秘密，我們認為垂直整合是其中的關鍵因素。現在我把麥克風交給史蒂夫。

Sure, Rob. Just correct me if I don't have the right question. But in terms of our growth opportunities in '26. So obviously, we're going to come back shortly with the next earnings talk about 2026. But there's a lot of things that we focused on to put us in a very strong position to reaccelerate growth.

當然可以，羅布。如果我的問題問錯了，請糾正我。但就我們在 2026 年的成長機會而言。很顯然，我們很快就會帶著關於 2026 年的下一份收益報告回來。但我們重點關注了很多方面，使我們處於非常有利的地位，能夠重新加速成長。

You heard some of the stats from the third quarter, strong CTV growth, mobile as well as our emerging revenues. So once we work through the current transition that we've identified in the second half of this year, we are very confident that we're going to be growing on a number of different fronts. Just as a reminder, we've been investing in secular growth areas. This has been a long-term strategy, and we're seeing the results of that. We are expanding, diversifying our DSP base.

你們都聽到了第三季的一些統計數據，包括強勁的 CTV 成長、行動裝置成長以及我們新興的收入。因此，一旦我們克服了今年下半年已經確定的當前轉型期，我們非常有信心，我們將在許多不同的方面取得成長。再次提醒大家，我們一直在投資長期成長領域。這是一項長期策略，我們現在看到了它的成效。我們正在擴大規模，使我們的DSP基礎更加多元化。

We're growing very strongly with commerce DSPs, specialized DSPs around pharma. And so the coupling of our focus on secular growth, expanding our buyer base and then innovation around AI, not just in the infrastructure, but certainly in terms of capabilities, functionality. We launched this past quarter, AI-driven publisher products, and that will continue to be sources of growth. So we're very positive about the growth in the second half of '26 as we look at the plethora of opportunities ahead of us.

我們在電商DSP領域，尤其是在醫藥領域的專業DSP領域，發展動能非常強勁。因此，我們將重點放在長期成長上，擴大客戶群，然後圍繞人工智慧進行創新，不僅在基礎設施方面，而且在能力和功能方面也是如此。我們在上個季度推出了人工智慧驅動的出版商產品，這些產品將繼續成為成長的來源。因此，鑑於我們面前湧現出許多機遇，我們對 2026 年下半年的成長前景非常樂觀。

Jacob Armstrong, KeyBanc.

Jacob Armstrong，KeyBanc。

This is Jacob on for Justin. Can you discuss how you believe the role of SSP needs to evolve in the coming years as Agentic AI expands? And what are the key investments needed to ensure PubMatic is best positioned to capitalize off this transition over the next few years?

這是雅各替賈斯汀發言。您能否談談隨著智能體人工智慧的擴展，您認為SSP的角色在未來幾年需要如何演變？為了確保 PubMatic 在未來幾年內能夠充分利用這項轉型帶來的機遇，需要進行哪些關鍵投資？

Sure. Yes. So I think the role of the SSP is going to expand significantly from transaction automation to a much bigger role in workflow automation, particularly around audience and inventory discovery and planning and in measurement. So if we think about where programmatic technologies have been applied so far, maybe for the last 15 or so years, I would say it's been heavily applied at the transaction level. So meaning we have an ad impression, and we want to get a number of advertiser bids on it.

當然。是的。所以我認為 SSP 的角色將從交易自動化大幅擴展到工作流程自動化，尤其是在受眾和庫存發現、規劃和衡量方面。所以，如果我們回顧一下程序化技術迄今為止的應用領域，例如過去 15 年左右的時間裡，我認為它主要應用於交易層面。也就是說，我們獲得了一次廣告展示機會，我們希望獲得一定數量的廣告商競價。

And then we're helping DSPs bid on that individual ad impression. And so there's been a lot of maturation and innovation and focus on that like single atomic impression. But we still have RFPs that advertisers send or agencies send out to publishers via e-mail, fill out the spreadsheet. We're launching this ad campaign. We want to understand what audiences or what inventory you might have available.

然後，我們幫助需求方平台（DSP）競價購買單一廣告展示機會。因此，在單原子印跡等方面，已經取得了長足的進步與創新。但我們仍然會收到廣告商或代理商透過電子郵件發送給出版商的 RFP，請填寫電子表格。我們將啟動這項廣告宣傳活動。我們想了解您可能擁有哪些受眾群體或庫存。

Some human at the publisher fills that out and then they e-mail that back to the agency. The agency collects those and then they decide where they're going to set up and allocate budget. So that's still a largely manual process. Some things have improved, but still a lot of manual approaches.

出版社的某個人填寫完這份表格後，會透過電子郵件將其發回給代理商。該機構收集這些信息，然後決定在哪裡設立機構並分配預算。所以這仍然主要是一個人工操作的過程。有些方面有所改進，但仍有許多人工操作。

So I think there's a huge opportunity to think outside of the pure atomic impression or transaction around the discovery and planning and then after the transaction to the measurement to use AI where an advertiser's agent or an agency's agent can say, hey, I'm launching this product or service, please tell me what you can do for me as a publisher or media owner from an audience and an inventory perspective and we can take a structured response, aggregate that up across many of the publishers that we're working with and then deliver a preconstructed campaign brief to the agency. Agency can begin to buy that using our transaction pipes, and then we can have a feedback loop around measurement with that. And then the agency can revise their campaign.

所以我認為，在發現和規劃階段，以及交易後的衡量階段，存在著巨大的機會跳出純粹的原子式曝光或交易的思維模式，利用人工智能，廣告主的代理人或代理商的代理人可以這樣說：“嘿，我要推出這個產品或服務，請告訴我，作為出版商或媒體所有者，你們能從受眾和庫存的角度為我提供什麼信息？代理商可以透過我們的交易管道開始購買，然後我們可以圍繞此建立一個衡量回饋循環。然後，該機構就可以修改他們的行銷活動。

So I think there's a lot that can be done outside of that single transaction element, and that's really where we are focused with the transaction layer of the stack that I mentioned earlier with ad context protocol, we're working out what exactly should those structured requests and responses look like and then how do they get set up, who owns what data and then how do they get optimized. So I think there's a long runway, Jacob, ahead in that area.

所以我認為除了單一的交易元素之外，還有很多事情可以做，而這正是我們目前關注的重點，也就是我之前提到的廣告上下文協議中的交易層。我們正在研究這些結構化的請求和回應究竟應該是什麼樣子，以及如何設定它們，誰擁有哪些數據，以及如何優化它們。所以我覺得雅各布，在那方面還有很長的路要走。

Ed Alter, Jefferies.

艾德‧奧特，傑富瑞集團。

I wanted to talk about the investments you're making to meet the demand from the mid-market DSPs like Mountain. Where exactly are those going to show up? Is that more headcount type investments? It would be great to talk about the color on that.

我想談談你們為滿足像 Mountain 這樣的中階市場 DSP 的需求而進行的投資。它們究竟會在哪裡出現？這是更多增加人事投入的項目嗎？很想聊聊它的顏色。

Sure. So we've been very focused throughout this year and really leading into this year is to be as efficient as possible in terms of where we deploy our teams. And we made a very conscious decision in the last 18 months to move more and more resources towards the fastest-growing areas, secular growth areas. of which, of course, critical DSPs are part of that. And so we've been increasing investment in the team that is -- goes directly and calls on these DSPs.

當然。因此，我們今年一直非常注重提高團隊部署的效率，並以此為目標展望今年的開端。在過去的18個月裡，我們做出了一個非常明確的決定，將越來越多的資源轉移到成長最快的領域，也就是長期成長領域。當然，關鍵的DSP也是其中的一部分。因此，我們一直在增加對直接與這些DSP對接的團隊的投入。

This year, we increased thus far about 19% in terms of headcount. And we've done that by reallocating team members around the organization. You can see from our results. Our overall headcount is roughly flat. And so we've done a very careful analysis of how we're going to keep on leveraging our existing resources. And of course, all of this is supported by the progress we've made in AI in terms of becoming more efficient just in our daily activities.

今年到目前為止，我們的員工人數增加了約 19%。我們透過重新分配組織內的團隊成員來實現這一點。從我們的結果中就可以看出。我們的員工總數基本上保持不變。因此，我們對如何繼續利用現有資源進行了非常仔細的分析。當然，這一切都得益於我們在人工智慧領域的進步，它使我們在日常活動中變得更有效率。

And so when I look ahead to '26, I'm anticipating we're going to keep our headcount roughly flat, but we're going to keep increasing the resources against those areas that are driving the greatest results. And we're on a great mission to do that. And it's not just on the OpEx side. We are looking at our CapEx, and we don't anticipate increasing our CapEx in '26 based upon sort of all the optimization, the work with NVIDIA, a lot of other things that are in the pipeline around efficiency and optimization.

因此，展望 2026 年，我預計我們的人員數量將大致保持不變，但我們將繼續增加對那些取得最大成果領域的資源投入。我們肩負著完成這項偉大使命的重任。而且這不僅僅是關乎營運支出方面。我們正在審視我們的資本支出，並且基於所有的優化、與 NVIDIA 的合作以及許多其他正在籌備中的有關效率和優化的項目，我們預計 2026 年不會增加我們的資本支出。

And so from our perspective, we're very confident that we are able to move dollars against the right opportunities without burdening the P&L. And so feeling good about the progress and our ability to increase our margins as revenue reaccelerates.

因此，從我們的角度來看，我們非常有信心將資金投入正確的投資機會中，而不會對損益表造成負擔。因此，我們對目前的進展以及隨著收入重新加速成長而提高利潤率的能力感到滿意。

Unfortunately, we are just about out of time. So I'm going to turn the call back over to Rajeev for closing remarks, and we'll talk to you all in your follow-up calls very shortly.

很遺憾，我們時間不多了。那麼，我將把電話轉回給拉吉夫，讓他做總結發言，我們很快就會在你們的後續電話中再聊。

Thank you, Stacie, and thank you all for joining us today. Our results demonstrate the power of our differentiated business model. We continue to innovate, diversify our business and operate with discipline. Our AI innovation is leading the industry with measurable outcomes driving momentum across the ecosystem. Looking to 2026 and beyond, as revenue growth reaccelerates, we anticipate margin expansion at both the gross and adjusted EBITDA levels because of our efficient and leveraged business.

謝謝Stacie，也謝謝大家今天能來參加我們的節目。我們的成果證明了我們差異化商業模式的優勢。我們不斷創新，拓展業務，並以嚴謹的態度運作。我們的人工智慧創新以可衡量的成果引領產業，推動整個生態系統的發展動能。展望 2026 年及以後，隨著營收成長再次加速，我們預計由於我們高效且槓桿化的業務，毛利和調整後 EBITDA 的利潤率都將擴大。

We look forward to seeing many of you at upcoming conferences, including the UBS Technology and AI Conference on December 2, the Wolfe Virtual SMID Cap Conference on December three and Raymond James TMT Conference on December 9. Thanks, everyone, for joining us today. Have a great afternoon.

我們期待在即將舉行的會議上見到你們中的許多人，包括 12 月 2 日的瑞銀科技與人工智慧大會、12 月 3 日的 Wolfe 虛擬中小市值公司大會以及 12 月 9 日的 Raymond James TMT 大會。感謝各位今天蒞臨。祝您下午愉快。