2023年Amazon致股東信：AI創新與客戶至上的成長之路｜見識之旅

導讀

Amazon 2023年致股東信是由公司CEO Andy Jassy撰寫的年度報告，詳細闡述了公司過去一年的業績、創新和未來展望。這封信涵蓋了Amazon各個業務部門的表現，包括零售、雲計算（AWS）和廣告等領域。

Jassy強調了公司在財務表現、客戶體驗改善以及技術創新方面取得的重大進展。

信中重點包括：

• 公司總收入增長12%達到5750億美元，營業收入大幅提升201%
• 零售業務在產品選擇、價格和便利性方面的持續改進
• AWS面對客戶成本優化需求時的靈活應對；以及廣告業務的強勁增長

Jassy還討論了公司在人工智能領域的投資和創新，以及亞馬遜如何運用AI技術改善客戶體驗和提高運營效率。

這封致股東信不僅展示了Amazon在2023年的成就，還描繪了公司對未來的願景和戰略規劃，凸顯了Amazon在競爭激烈的科技和零售行業中保持領先地位的決心。

正文

Dear Shareholders:

親愛的股東們：

Last year at this time, I shared my enthusiasm and optimism for Amazon's future. Today, I have even more. The reasons are many, but start with the progress we've made in our financial results and customer experiences, and extend to our continued innovation and the remarkable opportunities in front of us.

去年此時，我向各位分享了對亞馬遜未來的熱情和樂觀。如今，我的信心更加堅定。這源於多方面的原因，首要是我們在財務業績和客戶體驗方面取得的顯著進展，其次是我們不懈的創新精神，以及擺在我們面前的巨大機遇。

In 2023, Amazon's total revenue grew 12% year-over-year ("YoY") from $514B to $575B. By segment, North America revenue increased 12% YoY from $316B to $353B, International revenue grew 11% YoY from $118Bto $131B, and AWS revenue increased 13% YoY from $80B to $91B.

在 2023 年，Amazon 的總收入同比增長 12％，從 5140 億美元增至 5750 億美元。按部門劃分，北美收入同比增長 12％，從 3160 億美元增至 3530 億美元；國際收入同比增長 11％，從 1180 億美元增至 1310 億美元；AWS 收入同比增長 13％，從 800 億美元增至 910 億美元。

Further, Amazon's operating income and Free Cash Flow ("FCF") dramatically improved. Operating income in 2023 improved 201% YoY from $12.2B (an operating margin of 2.4%) to $36.9B (an operating margin of 6.4%). Trailing Twelve Month FCF adjusted for equipment finance leases improved from -$12.8B in 2022 to $35.5B (up $48.3B).

此外，Amazon 的營業收入和自由現金流（FCF）大幅改善。2023 年的營業收入同比增長 201%，從 122 億美元（營業利潤率 2.4%）增至 369 億美元（營業利潤率 6.4%）。經設備融資租賃調整後的過去十二個月 FCF，從 2022 年的負 128 億美元改善至 355 億美元，增加了 483 億美元。

While we've made meaningful progress on our financial measures, what we're most pleased about is the continued customer experience improvements across our businesses.

雖然我們在財務指標方面取得了顯著進展，但我們最滿意的是我們各項業務中持續改善的客戶體驗。

In our Stores business, customers have enthusiastically responded to our relentless focus on selection, price, and convenience. We continue to have the broadest retail selection, with hundreds of millions of products available, tens of millions added last year alone, and several premium brands starting to list on Amazon (e.g. Coach, Victoria's Secret, Pit Viper, Martha Stewart, Clinique, Lancôme, and Urban Decay).

在我們的商店業務中，客戶熱烈回應我們對選擇、價格和便利性的不懈關注。我們繼續擁有最廣泛的零售選擇，提供數億種產品，僅去年就新增了數千萬種，並且有幾個高端品牌開始在 Amazon 上架（例如 Coach、Victoria's Secret、Pit Viper、Martha Stewart、Clinique、Lancôme 和 Urban Decay）。

Being sharp on price is always important, but particularly in an uncertain economy, where customers are careful about how much they're spending. As a result, in Q4 2023, we kicked off the holiday season with Prime Big Deal Days, an exclusive event for Prime members to provide an early start on holiday shopping. This was followed by our extended Black Friday and Cyber Monday holiday shopping event, open to all customers, that became our largest revenue event ever. For all of 2023, customers saved nearly $24B across millions of deals and coupons, almost 70% more than the prior year.

價格優勢向來重要，但在經濟不確定時期尤為關鍵，因為客戶更加謹慎消費。有鑑於此，我們在 2023 年第四季度以「Prime 大優惠日」揭開假日季序幕。這項專為 Prime 會員設計的獨家活動讓他們能搶先開始假日購物。接著，我們延長了黑色星期五和網路星期一的假日購物活動，並向所有顧客開放，這成為我們有史以來最大規模的收入活動。2023 年全年，客戶透過數百萬筆交易和優惠券共節省近 240 億美元，較前一年增加近 70%。

We also continue to improve delivery speeds, breaking multiple company records. In 2023, Amazon delivered at the fastest speeds ever to Prime members, with more than 7 billion items arriving same or next day, including more than 4 billion in the U.S. and more than 2 billion in Europe. In the U.S., this result is the combination of two things. One is the benefit of regionalization, where we re-architected the network to store items closer to customers. The other is the expansion of same-day facilities, where in 2023, we increased the number of items delivered same day or overnight by nearly 70% YoY. As we get items to customers this fast, customers choose Amazon to fulfill their shopping needs more frequently, and we can see the results in various areas including how fast our everyday essentials business is growing (over 20% YoY in Q4 2023).

我們也持續改善配送速度，打破多項公司紀錄。在2023年，Amazon 為 Prime 會員提供了有史以來最快的配送速度，超過70億件商品實現了當日或次日送達，其中包括美國的40多億件和歐洲的20多億件。在美國，這一成果源於兩個因素。一是區域化的優勢，我們重新設計了網絡架構，使商品儲存地點更接近顧客。二是同日配送設施的擴展，2023年我們將當日或隔夜送達的商品數量同比增加了近70％。隨著我們以如此快的速度將商品送達顧客手中，顧客更頻繁地選擇 Amazon 來滿足他們的購物需求，我們可以在多個領域看到成果，包括日常必需品業務的快速增長（2023年第四季度同比增長超過20％）。

Our regionalization efforts have also trimmed transportation distances, helping lower our cost to serve. In 2023, for the first time since 2018, we reduced our cost to serve on a per unit basis globally. In the U.S. alone, cost to serve was down by more than $0.45 per unit YoY. Decreasing cost to serve allows us both to invest in speed improvements and afford adding more selection at lower Average Selling Prices ("ASPs"). More selection at lower prices puts us in consideration for more purchases.

我們的區域化策略不僅縮短了運輸距離，還有效降低了服務成本。2023年，我們首次自2018年以來在全球範圍內實現了每單位服務成本的下降。僅美國一地，每單位服務成本就同比降低了逾0.45美元。這項成本節省使我們能夠在提升配送速度的同時，以更低的平均售價（「ASPs」）擴大商品選擇。更豐富的選擇加上更具競爭力的價格，讓我們在消費者的購買決策中佔據更有利的地位。

As we look toward 2024 (and beyond), we're not done lowering our cost to serve. We've challenged every closely held belief in our fulfillment network, and reevaluated every part of it, and found several areas where we believe we can lower costs even further while also delivering faster for customers. Our inbound fulfillment architecture and resulting inventory placement are areas of focus in 2024, and we have optimism there's more upside for us.

展望2024年及未來，我們在降低服務成本方面的努力仍在持續。我們對配送網絡中的每一個既定觀念提出質疑，重新評估了每個環節，並發現了幾個可以進一步降低成本同時提升客戶配送速度的領域。2024年，我們將把重點放在入庫配送架構和相應的庫存配置上。我們對此充滿信心，相信這些領域還有巨大的改進空間。

Internationally, we like the trajectory of our established countries, and see meaningful progress in our emerging geographies (e.g. India, Brazil, Australia, Mexico, Middle East, Africa, etc.) as they continue to expand selection and features, and move toward profitability (in Q4 2023, Mexico became our latest international Stores locale to turn profitable). We have high conviction that these new geographies will continue to grow and be profitable in the long run.

在國際業務方面，我們對已建立業務的國家的發展軌跡感到滿意，並在新興市場（如印度、巴西、澳大利亞、墨西哥、中東和非洲）看到顯著進展。這些地區持續擴大商品選擇和功能，並逐步邁向盈利。2023年第四季度，墨西哥成為我們最新實現盈利的國際商店地區，這一成果令人振奮。我們堅信，這些新市場將在長期內持續增長並實現穩健盈利。

Alongside our Stores business, Amazon's Advertising progress remains strong, growing 24% YoY from $38B in 2022 to $47B in 2023, primarily driven by our sponsored ads. We've added Sponsored TV to this offering, a self-service solution for brands to create campaigns that can appear on up to 30+ streaming TV services, including Amazon Freevee and Twitch, and have no minimum spend. Recently, we've expanded our streaming TV advertising by introducing ads into Prime Video shows and movies, where brands can reach over 200 million monthly viewers in our most popular entertainment offerings, across hit movies and shows, award-winning Amazon MGM Originals, and live sports like Thursday Night Football. Streaming TV advertising is growing quickly and off to a strong start.

除了商店業務，Amazon 的廣告業務表現依然亮眼，從 2022 年的 380 億美元大幅增長 24% 至 2023 年的 470 億美元，主要歸功於我們的贊助廣告。我們進一步推出了 Sponsored TV，這是一項創新的自助服務解決方案，讓品牌能在多達 30 多個串流電視平台上輕鬆創建廣告活動，包括 Amazon Freevee 和 Twitch，且無最低消費門檻。最近，我們通過在 Prime Video 的節目和電影中引入廣告，進一步拓展了串流電視廣告業務。品牌現可通過熱門影視作品、屢獲殊榮的 Amazon MGM Originals，以及如 Thursday Night Football 等備受矚目的直播體育賽事，觸及超過 2 億的月活觀眾。串流電視廣告業務增長迅猛，開局強勁，前景可期。

Shifting to AWS, we started 2023 seeing substantial cost optimization, with most companies trying to save money in an uncertain economy. Much of this optimization was catalyzed by AWS helping customers use the cloud more efficiently and leverage more powerful, price-performant AWS capabilities like Graviton chips (our generalized CPU chips that provide ~40% better price-performance than other leading x86 processors), S3 Intelligent Tiering (a storage class that uses AI to detect objects accessed less frequently and store them in less expensive storage layers), and Savings Plans (which give customers lower prices in exchange for longer commitments). This work diminished short-term revenue, but was best for customers, much appreciated, and should bode well for customers and AWS longer-term. By the end of 2023, we saw cost optimization attenuating, new deals accelerating, customers renewing at larger commitments over longer time periods, and migrations growing again.

轉向 AWS，2023 年初我們觀察到大規模的成本優化趨勢，多數公司在不確定的經濟環境中力求節省開支。這種優化主要由 AWS 推動，我們協助客戶更有效地使用雲端並利用更強大、性價比更高的 AWS 功能。這些功能包括 Graviton 晶片（我們的通用 CPU 晶片，比領先的 x86 處理器提供約 40% 更佳性價比）、S3 Intelligent Tiering（利用 AI 檢測較少訪問的對象並存儲在更經濟的存儲層）和 Savings Plans（為長期承諾的客戶提供更優惠價格）。這項工作雖然減少了短期收入，但最有利於客戶，備受讚賞，長遠來看對客戶和 AWS 都有益。到 2023 年底，我們發現成本優化趨勢放緩，新交易加速，客戶以更大承諾和更長期限續約，遷移業務再次增長。

The past year was also a significant delivery year for AWS. We announced our next generation of generalized CPU chips (Graviton4), which provides up to 30% better compute performance and 75% more memory bandwidth than its already-leading predecessor (Graviton3). We also announced AWS Trainium2 chips, which will deliver up to four times faster machine learning training for generative AI applications and three times more memory capacity than Trainium1. We continued expanding our AWS infrastructure footprint, now offering 105 Availability Zones within 33 geographic Regions globally, with six new Regions coming (Malaysia, Mexico, New Zealand, the Kingdom of Saudi Arabia, Thailand, and a second German region in Berlin). In Generative AI ("GenAI"), we added dozens of features to Amazon SageMaker to make it easier for developers to build new Foundation Models ("FMs"). We invented and delivered a new service (Amazon Bedrock) that lets companies leverage existing FMs to build GenAI applications. And, we launched the most capable coding assistant around in Amazon Q. Customers are excited about these capabilities, and we're seeing significant traction in our GenAI offerings. (More on how we're approaching GenAI and why we believe we'll be successful later in the letter.)

過去一年對 AWS 而言是重要的里程碑之年。我們推出了新一代通用 CPU 晶片 Graviton4，較其已領先的前代產品 Graviton3 提供高達 30% 的計算性能提升和 75% 的記憶體頻寬增加。我們還發布了 AWS Trainium2 晶片，為生成式 AI 應用帶來四倍於 Trainium1 的機器學習訓練速度，以及三倍的記憶體容量。

AWS 持續擴展其全球基礎設施，目前在 33 個地理區域內提供 105 個可用區，並即將在六個新區域（馬來西亞、墨西哥、紐西蘭、沙特阿拉伯、泰國和德國柏林）開展業務。

在生成式 AI（「GenAI」）領域，我們為 Amazon SageMaker 新增了數十項功能，簡化了開發人員構建新基礎模型（「FMs」）的流程。我們還創新推出了 Amazon Bedrock 服務，使企業能夠利用現有 FMs 構建 GenAI 應用。此外，我們推出了功能強大的編碼助手 Amazon Q。

這些創新激發了客戶的熱情，我們的 GenAI 產品也因此獲得了顯著的市場吸引力。（我們將在信的後續部分詳細闡述我們的 GenAI 策略以及我們為何深信能在這一領域取得成功。）

We're also making progress on many of our newer business investments that have the potential to be important to customers and Amazon long-term. Touching on two of them:

我們在許多新的業務投資上也取得了進展，這些投資有潛力在長期內對客戶和 Amazon 變得重要。讓我們談談其中兩項：

We have increasing conviction that Prime Video can be a large and profitable business on its own. This confidence is buoyed by the continued development of compelling, exclusive content (e.g. Thursday Night Football, Lord of the Rings, Reacher, The Boys, Citadel, Road House, etc.), Prime Video customers' engagement with this content, growth in our marketplace programs (through our third-party Channels program, as well as the broad selection of shows and movies customers rent or buy), and the addition of advertising in Prime Video.

我們越發確信 Prime Video 有潛力成為一個獨立的大型盈利業務。這份信心源於幾個關鍵因素：我們持續開發引人入勝的獨家內容（如《週四夜足球》、《魔戒》、《鐵拳》、《黑袍糾察隊》、《城堡風雲》、《鬥陣遊戲》等），Prime Video 用戶對這些內容的高度參與度，我們市場計劃的顯著增長（包括第三方 Channels 計劃，以及豐富的租借或購買節目和電影選擇），再加上我們在 Prime Video 中引入廣告服務。

In October, we hit a major milestone in our journey to commercialize Project Kuiper when we launched two end-to-end prototype satellites into space, and successfully validated all key systems and sub-systems—rare in an initial launch like this. Kuiper is our low Earth orbit satellite initiative that aims to provide broadband connectivity to the 400-500 million households who don't have it today (as well as governments and enterprises seeking better connectivity and performance in more remote areas), and is a very large revenue opportunity for Amazon. We're on track to launch our first production satellites in 2024. We've still got a long way to go, but are encouraged by our progress.

2023 年 10 月，我們在商業化 Project Kuiper 的進程中達成重要里程碑：成功將兩顆端到端原型衛星送入太空，並驗證了所有關鍵系統和子系統——這在首次發射中實屬罕見。Kuiper 是我們的低地球軌道衛星計劃，旨在為當前缺乏寬頻連接的 4-5 億家庭提供服務，同時滿足偏遠地區政府和企業對更優質連接的需求。這對 Amazon 而言是一個巨大的收入機會。我們計劃於 2024 年發射首批生產衛星，進展順利。儘管前路漫長，但我們對目前的進展深感鼓舞。

Overall, 2023 was a strong year, and I'm grateful to our collective teams who delivered on behalf of customers. These results represent a lot of invention, collaboration, discipline, execution, and reimagination across Amazon. Yet, I think every one of us at Amazon believes that we have a long way to go, in every one of our businesses, before we exhaust how we can make customers' lives better and easier, and there is considerable upside in each of the businesses in which we're investing.

總的來說，2023年對我們來說是碩果豐碩的一年。我由衷感謝所有為客戶付出努力的團隊。這些成果凝聚了Amazon在創新、協作、紀律、執行和重新構想方面的巨大心血。然而，我相信Amazon的每一位成員都深知，在我們的所有業務領域中，我們還有很長的路要走，才能窮盡所有讓客戶生活更美好、更便捷的可能性。我們投資的每個領域都蘊含著巨大的發展潛力。

===

In my annual letter over the last three years, I've tried to give shareholders more insight into how we're thinking about the company, the businesses we're pursuing, our future opportunities, and what makes us tick. We operate in a diverse number of market segments, but what ties Amazon together is our joint mission to make customers' lives better and easier every day. This is true across every customer segment we serve (consumers, sellers, brands, developers, enterprises, and creators). At our best, we're not just customer obsessed, but also inventive, thinking several years out, learning like crazy, scrappy, delivering quickly, and operating like the world's biggest start-up.

在過去三年的年度致股東信中，我致力於讓股東更深入地了解我們對公司的看法、我們追求的業務、未來的機遇，以及驅動我們前進的動力。儘管我們橫跨多個市場領域，但將 Amazon 凝聚在一起的是我們共同的使命：每天讓客戶的生活變得更美好、更便捷。這一理念貫穿我們服務的所有客戶群體，包括消費者、賣家、品牌、開發者、企業和創作者。在最佳狀態下，我們不僅以客戶為中心，還充滿創新精神，著眼長遠，保持旺盛的學習欲望，精打細算，快速交付，並以世界上最大初創公司的活力運營。

We spend enormous energy thinking about how to empower builders, inside and outside of our company. We characterize builders as people who like to invent. They like to dissect a customer experience, assess what's wrong with it, and reinvent it. Builders tend not to be satisfied until the customer experience is perfect. This doesn't hinder them from delivering improvements along the way, but it drives them to keep tinkering and iterating continually. While unafraid to invent from scratch, they have no hesitation about using high-quality, scalable, cost-effective components from others. What matters to builders is having the right tools to keep rapidly improving customer experiences.

我們投入大量精力思考如何賦能公司內外的創新者。我們將創新者定義為熱衷於發明的人。他們善於解構客戶體驗，識別問題所在，並重新構思解決方案。創新者在客戶體驗達到完美之前絕不輕言滿足。這種追求卓越的態度不僅不妨礙他們持續改進，反而激勵他們不斷調整和迭代。儘管他們勇於從零開始創造，但也樂於採用他人開發的高品質、可擴展、具有成本效益的組件。對創新者而言，關鍵在於擁有合適的工具，以持續快速優化客戶體驗。

The best way we know how to do this is by building primitive services. Think of them as discrete, foundational building blocks that builders can weave together in whatever combination they desire. Here's how we described primitives in our 2003 AWS Vision document:

我們發現實現這一目標的最佳方式是構建原始服務。這些服務可視為獨立的、基礎性的構建模塊，開發者能根據自身需求靈活組合使用。以下是我們在 2003 年 AWS 願景文件中對原始服務的精闢闡述：

"Primitives are the raw parts or the most foundational-level building blocks for software developers. They're indivisible (if they can be functionally split into two they must) and they do one thing really well. They're meant to be used together rather than as solutions in and of themselves. And, we'll build them for maximum developer flexibility. We won't put a bunch of constraints on primitives to guard against developers hurting themselves. Rather, we'll optimize for developer freedom and innovation."

「原始服務是軟體開發者的基本組件或最底層的構建模塊。它們不可再分（若功能上可分，就必須分開），且在特定領域表現卓越。這些服務設計用於組合使用，而非獨立解決方案。我們打造它們時，著眼於最大化開發者的靈活性。我們不會為了防止開發者犯錯而在原始服務上設置諸多限制。相反，我們致力於優化開發者的自由度和創新能力。」

Of course, this concept of primitives can be applied to more than software development, but they're especially relevant in technology. And, over the last 20 years, primitives have been at the heart of how we've innovated quickly.

當然，這種原始服務的理念不僅限於軟體開發，它在整個技術領域都具有重要意義。過去 20 年來，原始服務一直是我們快速創新的核心驅動力。

One of the many advantages to thinking in primitives is speed. Let me give you two counter examples that illustrate this point. First, we built a successful owned-inventory retail business in the early years at Amazon where we bought all our products from publishers, manufacturers, and distributors, stored them in our warehouses, and shipped them ourselves. Over time, we realized we could add broader selection and lower prices by allowing third-party sellers to list their offerings next to our own on our highly trafficked search and product detail pages. We'd built several core retail services (e.g. payments, search, ordering, browse, item management) that made trying different marketplace concepts simpler than if we didn't have those components. A good set of primitives? Not really.

以原始服務的方式思考有諸多優勢，其中之一是速度。讓我舉兩個反例來闡明這一點。首先，在 Amazon 早期，我們建立了一個成功的自有庫存零售業務。我們從出版商、製造商和經銷商處採購所有產品，將它們存儲在自己的倉庫中，並親自負責發貨。隨著時間推移，我們意識到可以通過允許第三方賣家在我們高流量的搜索和產品詳情頁面上列出他們的商品，來擴大選擇範圍並降低價格。我們開發了幾個核心零售服務（如支付、搜索、訂購、瀏覽、商品管理），這使得嘗試不同的市場概念比沒有這些組件時更為容易。但這真的是一組優秀的原始服務嗎？事實並非如此。

It turns out that these core components were too jumbled together and not partitioned right. We learned this the hard way when we partnered with companies like Target in our Merchant.com business in the early 2000s. The concept was that target.com would use Amazon's ecommerce components as the backbone of its website, and then customize however they wished. To enable this arrangement, we had to deliver those components as separable capabilities through application programming interfaces ("APIs"). This decoupling was far more difficult than anticipated because we'd built so many dependencies between these services as Amazon grew so quickly the first few years.

事實證明，這些核心組件缺乏清晰的結構和適當的劃分。我們在 2000 年代初與 Target 等公司合作開展 Merchant.com 業務時，深刻體會到了這一點。當時的設想是 target.com 將 Amazon 的電子商務組件作為其網站的基礎，再根據需求進行定制。為實現這一目標，我們需要通過應用程序編程接口（「APIs」）將這些組件作為獨立功能提供。然而，這種解耦過程遠比預期複雜。原因在於 Amazon 在最初幾年的快速擴張中，我們在各個服務之間建立了過多的相互依賴關係。

This coupling was further highlighted by a heavyweight mechanism we used to operate called "NPI." Any new initiative requiring work from multiple internal teams had to be reviewed by this NPI cabal where each team would communicate how many people-weeks their work would take. This bottleneck constrained what we accomplished, frustrated the heck out of us, and inspired us to eradicate it by refactoring these ecommerce components into true primitive services with well-documented, stable APIs that enabled our builders to use each other's services without any coordination tax.

這種耦合問題進一步被我們稱為「NPI」的繁重操作機制所凸顯。任何需要多個內部團隊協作的新計劃都必須經過 NPI 小組的審查，每個團隊都要詳細說明他們的工作所需人週。這個瓶頸不僅限制了我們的成就，還讓我們感到極度沮喪。然而，這種挫折也激發了我們的創新精神。我們決心通過將這些電子商務組件重構為真正的原始服務來消除這個障礙。這些重構後的服務擁有文檔完善、穩定的 APIs，使我們的開發者能夠無縫使用彼此的服務，徹底消除了協調成本。

In the middle of the Target and NPI challenges, we were contemplating building a new set of infrastructure technology services that would allow both Amazon to move more quickly and external developers to build anything they imagined. This set of services became known as AWS, and the above experiences convinced us that we should build a set of primitive services that could be composed together how anybody saw fit. At that time, most technology offerings were very feature-rich, and tried to solve multiple jobs simultaneously. As a result, they often didn't do any one job that well.

在應對 Target 和 NPI 挑戰的過程中，我們開始構思一套全新的基礎設施技術服務。這套服務不僅能讓 Amazon 更加靈活高效，還能使外部開發者實現他們天馬行空的創意。這就是後來被稱為 AWS 的系統。我們的經歷讓我們堅信：我們應該打造一套可靈活組合的原始服務，以滿足各種需求。當時，大多數技術產品追求功能齊全，試圖一次性解決多個問題。然而，這種做法往往導致它們難以在任何一個領域真正脫穎而出。

Our AWS primitive services were designed from the start to be different. They offered important, highly flexible, but focused functionality. For instance, our first major primitive was Amazon Simple Storage Service ("S3") in March 2006 that aimed to provide highly secure object storage, at very high durability and availability, at Internet scale, and very low cost. In other words, be stellar at object storage. When we launched S3, developers were excited, and a bit mystified. It was a very useful primitive service, but they wondered, why just object storage? When we launched Amazon Elastic Compute Cloud ("EC2") in August 2006 and Amazon SimpleDB in 2007, people realized we were building a set of primitive infrastructure services that would allow them to build anything they could imagine, much faster, more cost-effectively, and without having to manage or lay out capital upfront for the datacenter or hardware. As AWS unveiled these building blocks over time (we now have over 240 at builders' disposal—meaningfully more than any other provider), whole companies sprang up quickly on top of AWS (e.g. Airbnb, Dropbox, Instagram, Pinterest, Stripe, etc.), industries reinvented themselves on AWS (e.g. streaming with Netflix, Disney+, Hulu, Max, Fox, Paramount), and even critical government agencies switched to AWS (e.g. CIA, along with several other U.S. Intelligence agencies). But, one of the lesser-recognized beneficiaries was Amazon's own consumer businesses, which innovated at dramatic speed across retail, advertising, devices (e.g. Alexa and Fire TV), Prime Video and Music, Amazon Go, Drones, and many other endeavors by leveraging the speed with which AWS let them build. Primitives, done well, rapidly accelerate builders' ability to innovate.

我們的 AWS 原始服務從一開始就與眾不同。它們提供重要、高度靈活且功能集中的服務。以 2006 年 3 月推出的 Amazon Simple Storage Service（「S3」）為例，它旨在提供高度安全、極高耐用性和可用性的對象存儲，達到互聯網規模，且成本極低。簡而言之，S3 在對象存儲方面表現卓越。

S3 的推出讓開發者既興奮又困惑。這是一個非常實用的原始服務，但他們不禁疑問：為何僅限於對象存儲？隨後在 2006 年 8 月推出的 Amazon Elastic Compute Cloud（「EC2」）和 2007 年的 Amazon SimpleDB，讓人們意識到我們正在構建一套原始基礎設施服務。這套服務使開發者能更快速、更經濟地實現任何想像，無需預先管理或投資數據中心或硬件。

隨著 AWS 不斷推出這些構建模塊（現已超過 240 個，遠超其他提供商），許多公司迅速在 AWS 上崛起（如 Airbnb、Dropbox、Instagram、Pinterest、Stripe 等）。各行各業也在 AWS 上重塑自己（如 Netflix、Disney+、Hulu、Max、Fox、Paramount 的流媒體服務），甚至關鍵政府機構也轉向使用 AWS（如 CIA 及其他幾個美國情報機構）。

然而，一個常被忽視的受益者是 Amazon 自身的消費者業務。借助 AWS 提供的快速構建能力，Amazon 在零售、廣告、設備（如 Alexa 和 Fire TV）、Prime Video 和 Music、Amazon Go、無人機等眾多領域都實現了驚人的創新速度。

優秀的原始服務能夠顯著加速開發者的創新能力。

So, how do you build the right set of primitives?

那麼，如何構建恰當的原始服務組合呢？

Pursuing primitives is not a guarantee of success. There are many you could build, and even more ways to combine them. But, a good compass is to pick real customer problems you're trying to solve.

追求原始服務並非成功的保證。你可以構建眾多原始服務，組合方式更是不計其數。然而，一個可靠的指南是：專注解決真實的客戶問題。

Our logistics primitives are an instructive example. In Amazon's early years, we built core capabilities around warehousing items, and then picking, packing, and shipping them quickly and reliably to customers. As we added third-party sellers to our marketplace, they frequently requested being able to use these same logistics capabilities. Because we'd built this initial set of logistics primitives, we were able to introduce Fulfillment by Amazon ("FBA") in 2006, allowing sellers to use Amazon's Fulfillment Network to store items, and then have us pick, pack, and ship them to customers, with the bonus of these products being available for fast, Prime delivery. This service has saved sellers substantial time and money (typically about 70% less expensive than doing themselves), and remains one of our most popular services. As more merchants began to operate their own direct-to-consumer ("DTC") websites, many yearned to still use our fulfillment capabilities, while also accessing our payments and identity primitives to drive higher order conversion on their own websites (as Prime members have already shared this payment and identity information with Amazon). A couple years ago, we launched Buy with Prime to address this customer need. Prime members can check out quickly on DTC websites like they do on Amazon, and receive fast Prime shipping speeds on Buy with Prime items—increasing order conversion for merchants by ~25% vs. their default experience.

我們的物流原始服務是一個絕佳範例。Amazon 初期，我們建立了圍繞倉儲、揀選、包裝和快速可靠配送的核心能力。當第三方賣家加入我們的市場時，他們頻繁要求使用這些物流能力。正因為我們已建立了這套初始物流原始服務，我們得以在 2006 年推出 Fulfillment by Amazon（「FBA」）。FBA 允許賣家使用 Amazon 的履行網絡存儲商品，由我們為客戶揀選、包裝和配送，這些產品還能享受快速的 Prime 配送。此服務為賣家節省大量時間和金錢（通常比自營便宜約 70%），至今仍是我們最受歡迎的服務之一。

隨著越來越多商家經營自己的直接面向消費者（「DTC」）網站，許多人渴望繼續使用我們的履行能力，同時訪問我們的支付和身份原始服務，以提高他們網站的訂單轉化率（因為 Prime 會員已與 Amazon 共享這些信息）。幾年前，我們推出了 Buy with Prime 來滿足這一需求。Prime 會員可以像在 Amazon 上一樣在 DTC 網站快速結賬，享受 Buy with Prime 商品的快速 Prime 配送——與商家默認體驗相比，訂單轉化率提高了約 25%。

As our Stores business has grown substantially, and our supply chain become more complex, we've had to develop a slew of capabilities in order to offer customers unmatched selection, at low prices, and with very fast delivery times. We've become adept at getting products from other countries to the U.S., clearing customs, and then shipping to storage facilities. Because we don't have enough space in our shipping fulfillment centers to store all the inventory needed to maintain our desired in-stock levels, we've built a set of lower-cost, upstream warehouses solely optimized for storage (without sophisticated end-user, pick, pack, and ship functions). Having these two pools of inventory has prompted us to build algorithms predicting when we'll run out of inventory in our shipping fulfillment centers and automatically replenishing from these upstream warehouses. And, in the last few years, our scale and available alternatives have forced us to build our own last mile delivery capability (roughly the size of UPS) to affordably serve the number of consumers and sellers wanting to use Amazon.

隨著我們的商店業務蓬勃發展，供應鏈也變得愈發複雜。為了給客戶提供無與倫比的選擇、低價格和極速配送，我們不得不開發一系列新能力。我們已熟練掌握了跨國運輸、清關和存儲的流程。由於運輸履行中心空間有限，無法存儲維持理想庫存水平所需的全部商品，我們建立了一套低成本的上游倉庫，專門用於存儲（不包含複雜的終端用戶、揀選、包裝和運輸功能）。這兩個庫存池的存在促使我們開發了預測算法，能夠預測運輸履行中心何時耗盡庫存，並自動從上游倉庫補充。此外，近年來，我們的規模和可用選擇迫使我們建立了自己的"最後一英里"配送能力（規模相當於 UPS），以便經濟高效地服務於大量使用 Amazon 的消費者和賣家。

We've solved these customer needs by building additional fulfillment primitives that both serve Amazon consumers better and address external sellers' increasingly complex ecommerce activities. For instance, for sellers needing help importing products, we offer a Global Mile service that leverages our expertise here. To ship inventory from the border (or anywhere domestically) to our storage facilities, we enable sellers to use either our first-party Amazon Freight service or third-party freight partners via our Partnered Carrier Program. To store more inventory at lower cost to ensure higher in-stock rates and shorter delivery times, we've opened our upstream Amazon Warehousing and Distribution facilities to sellers (along with automated replenishment to our shipping fulfillment centers when needed). For those wanting to manage their own shipping, we've started allowing customers to use our last mile delivery network to deliver packages to their end-customers in a service called Amazon Shipping. And, for sellers who wish to use our fulfillment network as a central place to store inventory and ship items to customers regardless of where they ordered, we have a Multi-Channel Fulfillment service. These are all primitives that we've exposed to sellers.

我們透過建立額外的履行原始服務來滿足這些客戶需求，不僅能更好地服務 Amazon 消費者，還能應對外部賣家日益複雜的電子商務活動。例如，我們為需要進口產品協助的賣家提供了 Global Mile 服務，充分利用我們在這方面的專業知識。為了將庫存從邊境（或國內任何地方）運送到我們的存儲設施，賣家可以選擇使用我們的自營 Amazon Freight 服務，或通過合作承運商計劃使用第三方貨運合作夥伴。

為了以更低成本存儲更多庫存，確保更高的庫存率和更短的交貨時間，我們向賣家開放了上游的 Amazon Warehousing and Distribution 設施，並在需要時自動補充到我們的運輸履行中心。對於想自行管理運輸的賣家，我們推出了 Amazon Shipping 服務，允許他們使用我們的最後一英里配送網絡將包裹送達終端客戶。此外，我們還提供多渠道履行服務，讓賣家可以將我們的履行網絡作為中心場所存儲庫存並向客戶發貨，無論訂單來自何處。

這些都是我們向賣家開放的原始服務，旨在為他們提供全方位的物流解決方案。

Building in primitives meaningfully expands your degrees of freedom. You can keep your primitives to yourself and build compelling features and capabilities on top of them to allow your customers and business to reap the benefits of rapid innovation. You can offer primitives to external customers as paid services (as we have with AWS and our more recent logistics offerings). Or, you can compose these primitives into external, paid applications as we have with FBA, Buy with Prime, or Supply Chain by Amazon (a recently released logistics service that integrates several of our logistics primitives). But, you've got options. You're only constrained by the primitives you've built and your imagination.

構建原始服務能顯著擴展你的自由度。你可以將這些服務保留內部使用，在其基礎上打造吸引人的功能與能力，讓客戶和業務從快速創新中獲益。你也可以將原始服務作為付費產品提供給外部客戶（就像我們對 AWS 和近期的物流服務所做的那樣）。另一種選擇是將這些原始服務組合成外部付費應用，例如我們的 FBA、Buy with Prime，或者 Supply Chain by Amazon（一項近期推出的整合了多個物流原始服務的解決方案）。總之，你擁有多種選擇。唯一的限制是你所構建的原始服務和你的想像力。

Take the new, same-day fulfillment facilities in our Stores business. They're located in the largest metro areas around the U.S. (we currently have 58), house our top-moving 100,000 SKUs (but also cover millions of other SKUs that can be injected from nearby fulfillment centers into these same-day facilities), and streamline the time required to go from picking a customer's order to being ready to ship to as little as 11 minutes. These facilities also constitute our lowest cost to serve in the network. The experience has been so positive for customers that we're planning to double the number of these facilities.

以我們商店業務中的新型當日履行設施為例。這些設施分佈在美國最大的都會區（目前有58個），不僅存放我們銷量最高的10萬種商品，還能覆蓋數百萬種可從鄰近履行中心調配的商品。我們將從揀選客戶訂單到準備發貨的時間縮短至最快11分鐘。值得一提的是，這些設施在我們的網絡中運營成本最低。客戶對此反應極為熱烈，促使我們計劃將這類設施的數量翻倍。

But, how else might we use this capability if we think of it as a core building block? We have a very large and growing grocery business in organic grocery (with Whole Foods Market) and non-perishable goods (e.g. consumables, canned goods, health and beauty products, etc.). We've been working hard on building a mass, physical store offering (Amazon Fresh) that offers a great perishable experience; however, what if we used our same-day facilities to enable customers to easily add milk, eggs, or other perishable items to any Amazon order and get same day? It might change how people think of splitting up their weekly grocery shopping, and make perishable shopping as convenient as non-perishable shopping already is.

但是，如果我們將這種能力視為核心構建模塊，還能如何運用呢？我們在Whole Foods Market和非易腐商品（如消耗品、罐頭食品、健康和美容產品）方面擁有一個龐大且快速增長的業務。我們一直致力於建立大規模實體店（Amazon Fresh），提供優質的易腐商品體驗。然而，如果我們利用當日設施，讓客戶能輕鬆地將牛奶、雞蛋或其他易腐商品添加到任何 Amazon 訂單並當天收到呢？這可能徹底改變人們對每週雜貨購物的看法，使易腐商品購物變得與非易腐商品一樣便捷。

Or, take a service that some people have questioned, but that's making substantial progress and we think of as a very valuable future primitive capability—our delivery drones (called Prime Air). Drones will eventually allow us to deliver packages to customers in less than an hour. It won't start off being available for all sizes of packages and in all locations, but we believe it'll be pervasive over time. Think about how the experience of ordering perishable items changes with sub-one-hour delivery?

以我們的配送無人機（Prime Air）為例。這項服務雖然受到一些質疑，但正在取得實質性進展，我們認為它是一種極具價值的未來原始能力。無人機終將使我們能在不到一小時內將包裹送達客戶手中。儘管初期無法適用於所有尺寸的包裹和所有地點，我們堅信它終將普及。試想，若配送時間縮短至一小時以內，訂購易腐商品的體驗將會如何徹底改變？

The same is true for Amazon Pharmacy. Need throat lozenges, Advil, an antibiotic, or some other medication? Same-day facilities already deliver many of these items within hours, and that will only get shorter as we launch Prime Air more expansively. Highly flexible building blocks can be composed across businesses and in new combinations that change what's possible for customers.

Amazon Pharmacy 亦是如此。需要喉糖、Advil、抗生素或其他藥物嗎？我們的當日設施現已能在數小時內送達這些物品，隨著 Prime Air 的進一步擴展，配送時間將更加縮短。這些高度靈活的構建模塊不僅可以跨業務整合，還能以創新方式組合，徹底改變客戶體驗的可能性。

Being intentional about building primitives requires patience. Releasing the first couple primitive services can sometimes feel random to customers (or the public at large) before we've unveiled how these building blocks come together. I've mentioned AWS and S3 as an example, but our Health offering is another. In the last 10 years, we've tried several Health experiments across various teams—but they were not driven by our primitives approach. This changed in 2022 when we applied our primitives thinking to the enormous global healthcare problem and opportunity. We've now created several important building blocks to help transform the customer health experience: Acute Care (via Amazon Clinic), Primary Care (via One Medical), and a Pharmacy service to buy whatever medication a patient may need. Because of our growing success, Amazon customers are now asking us to help them with all kinds of wellness and nutrition opportunities—which can be partially unlocked with some of our existing grocery building blocks, including Whole Foods Market or Amazon Fresh.

構建原始服務需要耐心。在我們展示這些構建模塊如何協同工作之前，推出最初幾個原始服務可能會讓客戶（或大眾）感到缺乏連貫性。我曾以 AWS 和 S3 為例，但我們的健康服務也是一個很好的說明。過去 10 年，我們在不同團隊中進行了幾項健康實驗，但這些並非源於我們的原始服務方法。2022 年，情況發生了轉變，我們將原始服務思維應用到全球醫療保健的巨大挑戰和機遇中。我們現已創建了幾個關鍵的構建模塊，旨在改變客戶的健康體驗：通過 Amazon Clinic 提供急性護理、通過 One Medical 提供初級護理，以及一項能夠滿足患者所有藥物需求的藥房服務。隨著我們不斷取得成功，Amazon 客戶開始尋求我們在各種健康和營養領域的幫助——這些需求部分可以通過我們現有的雜貨構建模塊來滿足，如 Whole Foods Market 或 Amazon Fresh。

As a builder, it's hard to wait for these building blocks to be built versus just combining a bunch of components together to solve a specific problem. The latter can be faster, but almost always slows you down in the future. We've seen this temptation in our robotics efforts in our fulfillment network. There are dozens of processes we seek to automate to improve safety, productivity, and cost. Some of the biggest opportunities require invention in domains such as storage automation, manipulation, sortation, mobility of large cages across long distances, and automatic identification of items. Many teams would skip right to the complex solution, baking in "just enough" of these disciplines to make a concerted solution work, but which doesn't solve much more, can't easily be evolved as new requirements emerge, and that can't be reused for other initiatives needing many of the same components. However, when you think in primitives, like our Robotics team does, you prioritize the building blocks, picking important initiatives that can benefit from each of these primitives, but which build the tool chest to compose more freely (and quickly) for future and complex needs. Our Robotics team has built primitives in each of the above domains that will be lynchpins in our next set of automation, which includes multi-floor storage, trailer loading and unloading, large pallet mobility, and more flexible sortation across our outbound processes (including in vehicles). The team is also building a set of foundation AI models to better identify products in complex environments, optimize the movement of our growing robotic fleet, and better manage the bottlenecks in our facilities.

作為建設者，等待構建模塊的完成比直接組合現有組件來解決特定問題更具挑戰性。後者雖然速度更快，但往往會在未來阻礙進展。我們在履行網絡的機器人項目中深刻體會到這種誘惑。為了提高安全性、生產力和成本效益，我們致力於自動化數十個流程。其中一些最大的機遇需要在存儲自動化、操作、分類、大型容器長距離移動以及物品自動識別等領域進行創新。許多團隊傾向於直接採用複雜的解決方案，僅僅融入「足夠」的這些技術來實現協調的解決方案。然而，這種做法無法全面解決問題，難以適應新需求，也不利於在其他需要相似組件的項目中重複使用。

相反，當你像我們的機器人團隊一樣以原始服務的方式思考時，你會優先考慮構建模塊。你會選擇能從每個原始服務中受益的重要項目，同時建立一個工具箱，為未來更複雜的需求提供更靈活、快速的組合方案。我們的機器人團隊已在上述每個領域建立了原始服務，這些服務將成為我們下一代自動化的核心，包括多層存儲、拖車裝卸、大型托盤移動，以及在出站流程（包括車輛）中實現更靈活的分類。此外，該團隊正在開發一套基礎 AI 模型，旨在更準確地識別複雜環境中的產品、優化不斷擴大的機器人隊伍的運動，並更有效地管理我們設施中的瓶頸問題。

Sometimes, people ask us "what's your next pillar? You have Marketplace, Prime, and AWS, what's next?" This, of course, is a thought-provoking question. However, a question people never ask, and might be even more interesting is what's the next set of primitives you're building that enables breakthrough customer experiences? If you asked me today, I'd lead with Generative AI ("GenAI").

有時，人們會問我們：「你們的下一個支柱是什麼？在 Marketplace、Prime 和 AWS 之後，下一步是什麼？」這無疑是一個發人深省的問題。然而，人們從未問過一個可能更有趣的問題：你們正在構建哪些能夠帶來突破性客戶體驗的新原始服務？如果你今天問我這個問題，我會毫不猶豫地回答：生成式 AI（「GenAI」）。

Much of the early public attention has focused on GenAI applications, with the remarkable 2022 launch of ChatGPT. But, to our "primitive" way of thinking, there are three distinct layers in the GenAI stack, each of which is gigantic, and each of which we're deeply investing.

早期的公眾注意力主要集中在 GenAI 的應用上，尤其是 2022 年 ChatGPT 的驚艷亮相。然而，從我們「原始服務」的思維角度來看，GenAI 架構實際上包含三個截然不同的層次。每個層次都具有巨大潛力，而我們正全力投資於這三個層面。

The bottom layeris for developers and companies wanting to build foundation models ("FMs").The primary primitives are the compute required to train models and generate inferences (or predictions), and the software that makes it easier to build these models. Starting with compute, the key is the chip inside it. To date, virtually all the leading FMs have been trained on Nvidia chips, and we continue to offer the broadest collection of Nvidia instances of any provider. That said, supply has been scarce and cost remains an issue as customers scale their models and applications. Customers have asked us to push the envelope on price-performance for AI chips, just as we have with Graviton for generalized CPU chips. As a result, we've built custom AI training chips (named Trainium) and inference chips (named Inferentia). In 2023, we announced second versions of our Trainium and Inferentia chips, which are both meaningfully more price-performant than their first versions and other alternatives. This past fall, leading FM-maker, Anthropic, announced it would use Trainium and Inferentia to build, train, and deploy its future FMs. We already have several customers using our AI chips, including Anthropic, Airbnb, Hugging Face, Qualtrics, Ricoh, and Snap.

底層是為希望構建基礎模型（「FMs」）的開發者和公司設計的。主要的原始服務包括訓練模型和生成推理（或預測）所需的計算能力，以及簡化這些模型構建過程的軟件。談到計算能力，核心在於芯片。迄今為止，幾乎所有領先的 FMs 都在 Nvidia 芯片上訓練，我們提供業內最廣泛的 Nvidia 實例集合。然而，隨著客戶擴展模型和應用，供應短缺和成本問題日益突出。客戶要求我們在 AI 芯片的性價比上有所突破，就像我們對通用 CPU 芯片 Graviton 所做的那樣。為此，我們開發了定制的 AI 訓練芯片（Trainium）和推理芯片（Inferentia）。2023 年，我們發布了 Trainium 和 Inferentia 芯片的第二代版本，性價比顯著優於第一代和其他替代品。去年秋天，領先的 FM 開發商 Anthropic 宣布將採用 Trainium 和 Inferentia 來構建、訓練和部署其未來的 FMs。目前，包括 Anthropic、Airbnb、Hugging Face、Qualtrics、Ricoh 和 Snap 在內的多家客戶已開始使用我們的 AI 芯片。

Customers building their own FM must tackle several challenges in getting a model into production. Getting data organized and fine-tuned, building scalable and efficient training infrastructure, and then deploying models at scale in a low latency, cost-efficient manner is hard. It's why we've built Amazon SageMaker, a managed, end-to-end service that's been a game changer for developers in preparing their data for AI, managing experiments, training models faster (e.g. Perplexity AI trains models 40% faster in SageMaker), lowering inference latency (e.g. Workday has reduced inference latency by 80% with SageMaker), and improving developer productivity (e.g. NatWest reduced its time-to-value for AI from 12-18 months to under seven months using SageMaker).

客戶在構建和部署自己的基礎模型（FM）時，面臨著諸多挑戰。數據組織與優化、建立高擴展性和高效率的訓練基礎設施，以及以低延遲、高成本效益的方式大規模部署模型，這些都是極具挑戰性的任務。正因如此，我們開發了 Amazon SageMaker。這是一個端到端的託管服務，徹底改變了開發者的工作方式。SageMaker 在 AI 數據準備、實驗管理、模型訓練加速、推理延遲降低和開發者生產力提升等方面表現卓越。例如，Perplexity AI 利用 SageMaker 將模型訓練速度提升了 40%，Workday 借助 SageMaker 將推理延遲降低了 80%，而 NatWest 則通過 SageMaker 將 AI 項目的交付時間從 12-18 個月縮短到了不到 7 個月。

The middle layer is for customers seeking to leverage an existing FM, customize it with their own data, and leverage a leading cloud provider's security and features to build a GenAI application—all as a managed service. Amazon Bedrock invented this layer and provides customers with the easiest way to build and scale GenAI applications with the broadest selection of first- and third-party FMs, as well as leading ease-of-use capabilities that allow GenAI builders to get higher quality model outputs more quickly. Bedrock is off to a very strong start with tens of thousands of active customers after just a few months. The team continues to iterate rapidly on Bedrock, recently delivering Guardrails (to safeguard what questions applications will answer), Knowledge Bases (to expand models' knowledge base with Retrieval Augmented Generation—or RAG—and real-time queries), Agents (to complete multi-step tasks), and Fine-Tuning (to keep teaching and refining models), all of which improve customers' application quality. We also just added new models from Anthropic (their newly-released Claude 3 is the best performing large language model in the world), Meta (with Llama 2), Mistral, Stability AI, Cohere, and our own Amazon Titan family of FMs. What customers have learned at this early stage of GenAI is that there's meaningful iteration required to build a production GenAI application with the requisite enterprise quality at the cost and latency needed. Customers don't want only one model. They want access to various models and model sizes for different types of applications. Customers want a service that makes this experimenting and iterating simple, and this is what Bedrock does, which is why customers are so excited about it. Customers using Bedrock already include ADP, Amdocs, Bridgewater Associates, Broadridge, Clariant, Dana-Farber Cancer Institute, Delta Air Lines, Druva, Genesys, Genomics England, GoDaddy, Intuit, KT, Lonely Planet, LexisNexis, Netsmart, Perplexity AI, Pfizer, PGA TOUR, Ricoh, Rocket Companies, and Siemens.

中間層針對希望利用現有基礎模型、用自身數據進行客製化，並運用領先雲服務商的安全性和功能來構建生成式 AI 應用的客戶——這些都以託管服務的形式提供。Amazon Bedrock 開創了這一層，為客戶提供了構建和擴展生成式 AI 應用的最簡便方式，擁有最廣泛的第一方和第三方基礎模型選擇，以及領先的易用性功能，使生成式 AI 開發者能更快獲得更高品質的模型輸出。

Bedrock 在短短幾個月內就取得了驚人的成績，吸引了數萬名活躍客戶。團隊持續快速迭代 Bedrock，最近推出了 Guardrails（保護應用回答的問題範圍）、Knowledge Bases（利用檢索增強生成—— RAG ——和即時查詢擴展模型知識庫）、Agents（完成多步驟任務）和 Fine-Tuning（持續優化模型），這些都大幅提升了客戶應用的品質。

我們還新增了來自 Anthropic（他們新發布的 Claude 3 是全球性能最佳的大型語言模型）、Meta（Llama 2）、Mistral、Stability AI、Cohere 的新模型，以及我們自家的 Amazon Titan 基礎模型系列。

客戶在生成式 AI 的初期階段認識到，要以合適的成本和延遲構建具備企業級品質的生產級生成式 AI 應用，需要進行大量有意義的迭代。客戶不僅僅需要一個模型，他們希望能夠根據不同類型的應用訪問各種模型和模型規模。客戶渴望一個能夠簡化這種實驗和迭代過程的服務，這正是 Bedrock 所提供的，也是客戶對它如此熱衷的原因。

目前使用 Bedrock 的客戶包括 ADP、Amdocs、Bridgewater Associates、Broadridge、Clariant、Dana-Farber Cancer Institute、Delta Air Lines、Druva、Genesys、Genomics England、GoDaddy、Intuit、KT、Lonely Planet、LexisNexis、Netsmart、Perplexity AI、Pfizer、PGA TOUR、Ricoh、Rocket Companies 和 Siemens。

The top layer of this stack is the application layer. We're building a substantial number of GenAI applications across every Amazon consumer business. These range from Rufus (our new, AI-powered shopping assistant), to an even more intelligent and capable Alexa, to advertising capabilities (making it simple with natural language prompts to generate, customize, and edit high-quality images, advertising copy, and videos), to customer and seller service productivity apps, to dozens of others. We're also building several apps in AWS, including arguably the most compelling early GenAI use case—a coding companion. We recently launched Amazon Q, an expert on AWS that writes, debugs, tests, and implements code, while also doing transformations (like moving from an old version of Java to a new one), and querying customers' various data repositories (e.g. Intranets, wikis, Salesforce, Amazon S3, ServiceNow, Slack, Atlassian, etc.) to answer questions, summarize data, carry on coherent conversation, and take action. Q is the most capable work assistant available today and evolving fast.

這個堆疊的頂層是應用層。我們正在為每個 Amazon 消費者業務開發大量的生成式 AI 應用。這些應用涵蓋範圍廣泛，包括：Rufus（我們新推出的 AI 驅動購物助手）、更智能且功能更強大的 Alexa、廣告功能（利用自然語言提示輕鬆生成、定制和編輯高質量圖像、廣告文案和視頻），以及提升客戶和賣家服務效率的應用，還有數十種其他應用。

我們還在 AWS 中開發了幾個應用，其中最引人注目的早期生成式 AI 用例可能是編碼助手。我們最近推出的 Amazon Q 就是一個 AWS 專家，它能編寫、調試、測試和實現代碼，同時還能進行轉換（如從舊版 Java 遷移到新版）。此外，Q 還可以查詢客戶的各種數據庫（如內部網、維基、Salesforce、Amazon S3、ServiceNow、Slack、Atlassian 等），以回答問題、總結數據、進行連貫對話並採取行動。Q 是當今最強大的工作助手，而且正在快速進化。

While we're building a substantial number of GenAI applications ourselves, the vast majority will ultimately be built by other companies. However, what we're building in AWS is not just a compelling app or foundation model. These AWS services, at all three layers of the stack, comprise a set of primitives that democratize this next seminal phase of AI, and will empower internal and external builders to transform virtually every customer experience that we know (and invent altogether new ones as well). We're optimistic that much of this world-changing AI will be built on top of AWS.

雖然我們正在開發眾多 GenAI 應用，但絕大多數最終將由其他公司構建。然而，我們在 AWS 中打造的不僅是引人注目的應用或基礎模型。這些 AWS 服務橫跨堆疊的三個層次，構成了一套原始服務，讓 AI 的下一個關鍵階段更加普及。這將賦予內部和外部開發者能力，幾乎可以改變我們所知的每種客戶體驗，同時創造全新體驗。我們樂觀地認為，大部分這種改變世界的 AI 將建立在 AWS 之上。

(By the way, don't underestimate the importance of security in GenAI. Customers' AI models contain some of their most sensitive data. AWS and its partners offer the strongest security capabilities and track record in the world; and as a result, more and more customers want to run their GenAI on AWS.)

（順帶一提，切勿低估生成式 AI 中安全性的重要性。客戶的 AI 模型包含了他們最敏感的數據。AWS 及其合作夥伴提供全球最強大的安全能力和成功案例；因此，越來越多的客戶選擇在 AWS 上運行他們的生成式 AI。）

===

Recently, I was asked a provocative question—how does Amazon remain resilient? While simple in its wording, it's profound because it gets to the heart of our success to date as well as for the future. The answer lies in our discipline around deeply held principles: 1/ hiring builders who are motivated to continually improve and expand what's possible; 2/ solving real customer challenges, rather than what we think may be interesting technology; 3/ building in primitives so that we can innovate and experiment at the highest rate; 4/ not wasting time trying to fight gravity (spoiler alert: you always lose)—when we discover technology that enables better customer experiences, we embrace it; 5/ accepting and learning from failed experiments—actually becoming more energized to try again, with new knowledge to employ.

最近，有人問了我一個發人深省的問題——Amazon 如何保持韌性？這個問題看似簡單，卻意義深遠，因為它觸及了我們過去和未來成功的核心。答案在於我們對幾個深層原則的堅持：

1. 聘用有動力不斷改進和拓展可能性的建設者；
2. 專注解決真正的客戶挑戰，而非僅追求我們認為有趣的技術；
3. 構建原始服務，以最高速率進行創新和實驗；
4. 不浪費時間對抗不可避免的趨勢（劇透：你總是會輸）——當我們發現能改善客戶體驗的技術時，我們就擁抱它；
5. 接受並從失敗的實驗中學習——這反而激發我們更有動力再次嘗試，運用新獲得的知識。

Today, we continue to operate in times of unprecedented change that come with unusual opportunities for growth across the areas in which we operate. For instance, while we have a nearly $500B consumer business, about 80% of the worldwide retail market segment still resides in physical stores. Similarly, with a cloud computing business at nearly a $100B revenue run rate, more than 85% of the global IT spend is still on-premises. These businesses will keep shifting online and into the cloud. In Media and Advertising, content will continue to migrate from linear formats to streaming. Globally, hundreds of millions of people who don't have adequate broadband access will gain that connectivity in the next few years. Last but certainly not least, Generative AI may be the largest technology transformation since the cloud (which itself, is still in the early stages), and perhaps since the Internet. Unlike the mass modernization of on-premises infrastructure to the cloud, where there's work required to migrate, this GenAI revolution will be built from the start on top of the cloud. The amount of societal and business benefit from the solutions that will be possible will astound us all.

如今，我們繼續在前所未有的變革時代中運營，為我們經營的各個領域帶來非同尋常的增長機會。舉例來說，儘管我們的消費者業務規模接近 5000 億美元，全球零售市場仍有約 80% 集中在實體店。同樣地，雖然我們的雲計算業務年收入已近 1000 億美元，但全球超過 85% 的 IT 支出仍在本地進行。這些業務將持續向線上和雲端遷移。在媒體和廣告領域，內容將不斷從傳統線性格式轉向流媒體。全球範圍內，數億缺乏足夠寬頻接入的人們將在未來幾年獲得這種連接。最後，但同樣重要的是，生成式 AI 可能是自雲計算（本身仍處於早期階段）以來，甚至可能是自互聯網以來最大的技術變革。與將本地基礎設施大規模現代化遷移到雲端不同，這場 GenAI 革命將從一開始就建立在雲端之上。這些潛在解決方案所帶來的社會和商業效益將讓我們所有人驚嘆不已。

There has never been a time in Amazon's history where we've felt there is so much opportunity to make our customers' lives better and easier. We're incredibly excited about what's possible, focused on inventing the future, and look forward to working together to make it so.

在 Amazon 的歷史上，我們從未感到有如此多的機會來讓客戶的生活變得更好、更輕鬆。我們對未來的可能性感到無比興奮，專注於發明未來，並期待與大家一起努力實現這一目標。

Sincerely,
Andy Jassy
President and Chief Executive Officer
Amazon.com, Inc.

謹致問候，
Andy Jassy
總裁兼CEO
Amazon.com, Inc.