AWS AI 營收突破 150 億美元：Amazon 年度股東信揭露 AI 事業三年成長 260 倍 | AWS AI Revenue Hits $15 Billion: Amazon Shareholder Letter Reveals 260x Growth in 3 Years
By Kit 小克 | AI Tool Observer | 2026-04-10
🇹🇼 AWS AI 營收突破 150 億美元：Amazon 年度股東信揭露 AI 事業三年成長 260 倍
AWS AI 營收首次公開，150 億美元代表什麼？
Amazon 執行長 Andy Jassy 在 2026 年度股東信中，首次揭露 AWS AI 服務的年化營收已突破 150 億美元，約佔 AWS 總營收的 10%。這是 Amazon 史上第一次公開 AI 業務的具體數字，也是目前所有雲端大廠中最透明的 AI 營收揭露。
為什麼 Amazon 選在現在公布 AWS AI 營收？
因為數字終於夠好看了。Jassy 在信中把 AI 比喻為早期 AWS 的翻版，三年前 AWS 商業化三年後營收才 5,800 萬美元，而生成式 AI 起飛三年後，AI 營收已經是當年的 260 倍。
這個時間點公布還有另一層意義：Amazon 今年資本支出預計達到 2,000 億美元，市場一直在質疑這筆錢花得值不值得。150 億的營收數字，就是 Jassy 對華爾街的回應——「我們不是在賭博」。
AWS AI 營收的三個關鍵數字
- 150 億美元：Q1 年化營收 run rate，佔 AWS 1,420 億總營收的約 10%
- 200 億美元：自研晶片（Graviton、Trainium、Nitro）年化營收，比年初翻倍
- 2,000 億美元：2026 全年預計資本支出，幾乎全押 AI 基礎建設
Amazon 的 AI 策略跟 Google、Microsoft 有什麼不同？
Amazon 走的是「基礎建設優先」路線，跟 Google 和 Microsoft 不太一樣。Google 靠自家模型 Gemini 衝前端體驗，Microsoft 靠 OpenAI 合作搶企業市場，而 Amazon 則是把重心放在底層：自研晶片 + 雲端平台 + 模型託管。
具體來說，AWS Bedrock 讓企業可以用 API 串接各家 LLM（包括 Anthropic Claude、Meta Llama），不綁定單一模型。加上 Trainium 晶片的成本優勢，Amazon 的策略是讓 AI 訓練和推論都在自家平台上跑，從底層就鎖住客戶。
對開發者和企業來說意味著什麼？
AWS AI 營收的爆發代表企業 AI 部署已經從試驗階段進入正式生產。如果你還在觀望要不要把 AI 工作負載搬上雲端，這個數字告訴你：很多公司已經在花真金白銀了。
- Bedrock 的 model hosting 正在變成企業 AI 的標準選擇
- Trainium 晶片讓訓練成本比 NVIDIA GPU 便宜 30-40%
- SageMaker 和 Q Developer 等工具鏈持續整合 AI Agent 功能
Amazon 股價為什麼漲了 4.5%？
市場反應很直接：AI 投資開始回本了。消息公布當天 Amazon 股價上漲 4.5%，因為投資人終於看到具體的 AI 營收數字，而不只是「我們在 AI 上投了很多錢」的空話。
FAQ
- Q：AWS AI 營收 150 億美元包含哪些服務？
A：主要來自 AWS Bedrock（模型託管）、SageMaker（ML 平台）、以及各類 AI/ML API 服務，是第一季度表現的年化數字。 - Q：Amazon 的 2,000 億資本支出都花在哪？
A：絕大部分投入資料中心建設、AI 加速器採購（包括 NVIDIA GPU 和自研 Trainium 晶片），以及全球雲端基礎建設擴張。 - Q：AWS AI 跟 Azure AI 比誰比較強？
A：Microsoft 尚未單獨揭露 Azure AI 營收，但根據財報暗示規模相當。AWS 的優勢在多模型選擇和自研晶片成本，Azure 的優勢在 OpenAI 獨家合作。 - Q：Trainium 晶片真的比 NVIDIA GPU 便宜嗎？
A：Amazon 宣稱 Trainium 2 在特定工作負載上比 NVIDIA H100 便宜 30-40%，但生態系成熟度和軟體支援仍是 NVIDIA 的強項。
好不好用，試了才知道
🇺🇸 AWS AI Revenue Hits $15 Billion: Amazon Shareholder Letter Reveals 260x Growth in 3 Years
AWS AI Revenue Disclosed for the First Time — What Does $15 Billion Mean?
In his 2026 annual shareholder letter, Amazon CEO Andy Jassy revealed for the first time that AWS AI services have reached an annualized revenue run rate exceeding $15 billion, representing roughly 10% of AWS total revenue. This marks the most transparent AI revenue disclosure from any major cloud provider to date.
Why Did Amazon Choose Now to Reveal AWS AI Revenue?
Because the numbers finally look impressive enough to share. Jassy compared AI growth to the early days of AWS — three years after AWS launched commercially, it had a $58 million revenue run rate. Three years after generative AI took off, the AI business is 260 times larger.
The timing also serves a strategic purpose. Amazon plans to spend roughly $200 billion in capital expenditure in 2026, and Wall Street has been questioning whether the spending is justified. The $15 billion figure is Jassy's answer — as he put it, this investment is "not on a hunch."
Three Key Numbers from AWS AI
- $15 billion: Q1 annualized AI revenue run rate, ~10% of AWS's $142B total
- $20 billion: Custom chip revenue (Graviton, Trainium, Nitro), doubled from earlier this year
- $200 billion: Planned 2026 capex, almost entirely for AI infrastructure
How Does Amazon's AI Strategy Differ from Google and Microsoft?
Amazon takes an "infrastructure-first" approach that differs from its competitors. While Google leads with its Gemini models and Microsoft leverages its exclusive OpenAI partnership, Amazon focuses on the foundation layer: custom chips + cloud platform + model hosting.
AWS Bedrock allows enterprises to access multiple LLMs (including Anthropic Claude, Meta Llama) through a unified API without vendor lock-in. Combined with Trainium's cost advantages, Amazon's strategy is to make AI training and inference run on its platform from the ground up.
What Does This Mean for Developers and Enterprises?
The explosive growth of AWS AI revenue signals that enterprise AI deployment has moved from experimentation to production. If you are still evaluating whether to move AI workloads to the cloud, this number tells you many companies are already spending real money.
- Bedrock model hosting is becoming the default choice for enterprise AI
- Trainium chips offer 30-40% cost savings over NVIDIA GPUs for certain workloads
- SageMaker and Q Developer continue integrating AI Agent capabilities
Why Did Amazon Stock Jump 4.5%?
The market reaction was straightforward: AI investment is paying off. Amazon shares rose 4.5% on the day of the announcement because investors finally saw concrete AI revenue figures instead of vague promises about future AI spending.
FAQ
- Q: What services are included in AWS's $15 billion AI revenue?
A: Primarily AWS Bedrock (model hosting), SageMaker (ML platform), and various AI/ML API services. The figure is an annualized run rate based on Q1 performance. - Q: Where is Amazon spending the $200 billion in capex?
A: The vast majority goes to data center construction, AI accelerator procurement (both NVIDIA GPUs and custom Trainium chips), and global cloud infrastructure expansion. - Q: How does AWS AI compare to Azure AI?
A: Microsoft has not separately disclosed Azure AI revenue, though earnings hints suggest comparable scale. AWS's advantage is multi-model flexibility and custom chip costs; Azure's advantage is its exclusive OpenAI partnership. - Q: Are Trainium chips really cheaper than NVIDIA GPUs?
A: Amazon claims Trainium 2 is 30-40% cheaper than NVIDIA H100 for specific workloads, but NVIDIA still leads in ecosystem maturity and software support.
好不好用，試了才知道
Sources / 資料來源
- Amazon cloud unit AI revenue run rate exceeds $15 billion - BNN Bloomberg
- Andy Jassy defends Amazon $200B spending spree - GeekWire
- Amazon Investing $200 Billion Into AI Capex - NextBigFuture
常見問題 FAQ
AWS AI 營收 150 億美元包含哪些服務？
主要來自 AWS Bedrock 模型託管、SageMaker ML 平台、以及各類 AI/ML API 服務，是第一季度表現的年化 run rate 數字。
Amazon 的 2000 億資本支出都花在哪裡？
絕大部分投入資料中心建設、AI 加速器採購（NVIDIA GPU 和自研 Trainium 晶片），以及全球雲端基礎建設擴張。
AWS AI 跟 Azure AI 比較誰比較強？
Microsoft 尚未單獨揭露 Azure AI 營收。AWS 優勢在多模型選擇和自研晶片成本，Azure 優勢在 OpenAI 獨家合作關係。
Trainium 晶片真的比 NVIDIA GPU 便宜嗎？
Amazon 宣稱 Trainium 2 在特定工作負載上比 H100 便宜 30-40%，但 NVIDIA 在生態系成熟度和軟體支援上仍領先。
延伸閱讀 / Related Articles
- EU AI Act 8 月全面生效：開發者必知的合規倒數與罰則重點 | EU AI Act Full Enforcement in August 2026: What Developers Must Know Before the Deadline
- Visa AI Agent 自動付款上線：你的 AI 助手即將能刷卡購物 | Visa Intelligent Commerce Connect: Your AI Agent Can Now Pay with Your Card
- NVIDIA Physical AI 機器人走出實驗室：GR00T N2、Cosmos 3 與真實部署案例完整解析 | NVIDIA Physical AI Robots Go Real: GR00T N2, Cosmos 3, and Production Deployments Explained
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言