In the earlier article "Global Cloud Giants Q1: AI and Non-AI Both Soar, Compute Still Insufficient," it was noted that cloud AI revenue continued to surge, non-AI demand recovered above expectations, and compute remained supply-constrained. This quarter, the three major clouds began to show AI winners (Azure, GCP) and a loser (AWS), and compute supply constraints were projected to persist into early next year. But as AWS data center build-outs completed and volumes ramped, AWS staged a "comeback" to become a winner. Under severe AI compute supply-demand imbalance, the speed of AI data center construction directly determines cloud growth rates.
1
Amazon AWS
Global Cloud Leader
AWS Q3 revenue $33B, up 20% year over year, with sequential growth acceleration of 3 percentage points, an 11-quarter high, up 7% sequentially. Operating income $11.4B, up 9% year over year, operating margin 35%, down 3 percentage points year over year.
AWS Q3 revenue represented 18% of Amazon total revenue, contributing 66% of Amazon operating income.
AWS Q3 revenue run rate $132B. Beyond sustained AI demand growth, enterprises have resumed migrating from on-premises to cloud. Supply constraints expected to persist for several quarters but improve each quarter. Added 3.8GW of power capacity over the past 12 months, more than any other CSP. Q4 will add another 1GW. Capacity expected to double again by 2027. Industry-wide, the near-term bottleneck may be power; at some point it may shift to chips. Bringing significant compute online; the speed of compute bring-up determines monetization speed. Custom Trainium2 chip compute orders up 150% sequentially. Project Rainier cluster built for Anthropic with 500K Trainium2 chips dedicated to Claude training, expected to reach 1M Trainium2 chips by year-end, a multi-billion dollar revenue business, up 150% sequentially. Trainium3, expected to enter volume production next year, will begin onboarding more customers, though AWS continues to order large volumes of NVIDIA chips. AWS coding agent Kiro has processed trillions of tokens to date, with rapid weekly active user growth. In migration and transformation, the Transform agent has saved customers 700K hours of manual work. For enterprise customers, the recently launched Quick Suite AI toolkit has compressed month-long projects into days, saving 80%+ time on complex tasks and achieving 90%+ cost reduction. For contact centers, Amazon Connect handled 12B minutes of AI-driven customer interactions over the past year; Connect Q3 run-rate revenue surpassed $1B. RPO reached $200B, up 22% year over year.
Q3 cash capex $34.2B, mostly allocated to AI. Full-year cash capex guided to $125B, with 2026 capex continuing to increase.
2
Microsoft Azure
Second-Largest Global Cloud
Microsoft Azure Q3 revenue $23.6B, up 40% year over year, with sequential growth acceleration of 1 percentage point, continuing to lead global top-three cloud growth.
Intelligent Cloud (Server + Azure + Enterprise Services) revenue $30.9B, up 28% year over year. Intelligent Cloud gross margin disclosed for the first time at 60%, down 4.1 percentage points year over year, primarily due to depreciation. Total operating income $13.4B, up 27% year over year, operating margin 43%. Intelligent Cloud Q3 revenue represented 40% of Microsoft total revenue, contributing 35% of Microsoft operating income.
This quarter Azure was driven by better-than-expected growth in the core infrastructure business (largest customer), continuing to gain market share. Cloud migration continues to accelerate. Azure AI services revenue met expectations. Even with more compute brought online this quarter, supply remains constrained. Going forward, Azure will continue to balance revenue growth against growing compute demands from first-party applications, AI solutions, internal R&D, and end-of-life server replacements. Azure revenue growth could have been higher. AI compute will increase 80%+ this year and double again over the next two years, reflecting observed demand. Azure AI Foundry has 80K customers, covering 80% of the Fortune 500. Announced the Fairwater global AI data center in Wisconsin, coming online next year with 2GW power capacity, deploying the world's first large-scale NVIDIA GB300 cluster. Phi-family SLMs have been downloaded 60M+ times cumulatively, up 3x year over year.
In data and analytics, Fabric revenue up 60%, 28K paid users. In databases, SQL DB Hyperscale revenue up nearly 75%, Cosmos DB up 50%. GitHub now has 180M+ developers, GitHub Copilot 26M+ users.
Azure compute supply expected to remain a bottleneck through end of FY26. Three Azure growth drivers: cloud migration + cloud-native SaaS scale expansion + AI. Commercial RPO $392B, up 51% year over year. Next quarter Azure revenue guided up 37% year over year. Capex continues to grow sequentially. FY26 overall capex growth rate will exceed FY25, above market expectations.
Market expectations for future Azure revenue diverge significantly from SemiAnalysis. Bloomberg consensus projects sequential Azure growth deceleration, while SemiAnalysis projects acceleration, even overtaking AWS by 2027. Notably, SemiAnalysis previously accurately predicted the AWS growth rebound ahead of the market.
3
Google Cloud
Third-Largest Global Cloud
Google Cloud Q3 revenue $15.2B, up 34% year over year, with sequential growth acceleration of 2 percentage points. Operating income $3.6B, operating margin 24%, up 7 percentage points year over year.
Google Cloud core five businesses: AI infrastructure, Vertex commercial AI platform, BigQuery data platform, AI cybersecurity solutions, Workspace productivity suite. GCP revenue growth continued to exceed overall Cloud growth. Backlog up 46% sequentially to $155B. Monthly token processing grew from 4.8 quadrillion at I/O to 13 quadrillion currently. Gemini MAU exceeds 650M, daily requests up 3x vs Q2. Gemini Enterprise has 2M+ subscribers from 700+ companies. Cloud signed more $1B+ contracts this quarter than in the past two years combined. GCP delivering NVIDIA GB300 instances to cloud customers; TPU v7 Ironwood launched, secured Anthropic order for 1M TPUs. Compute expected to remain tight through Q4 and 2026. Workspace driven by ARPU growth. GCP now has 13 product lines with ARR exceeding $1B.
Management expects compute supply constraints to persist through 2026.2025 capex guidance raised from $85B to $91-93B. 2026 capex expected to grow significantly.
4
Alibaba Cloud
China's Cloud Leader
Q3 revenue RMB 39.8B ($5.594B), up 34% year over year, with sequential growth acceleration of 8 percentage points. Excluding internal consolidation, up 29% year over year, with sequential acceleration of 3 percentage points. Driven by core public cloud including AI, with AI-related revenue achieving triple-digit year-over-year growth for the 9th consecutive quarter. Per Omida, Alibaba Cloud led China's AI cloud market with 35.8% share in H1.
Q3 EBITA profit RMB 3.6B, up 35% year over year, EBITA margin 9%, flat year over year. Q3 capex RMB 31.5B, previous quarter RMB 38.6B.
5
Oracle Cloud
Global Cloud Challenger
Q3 revenue $7.19B, up 28% year over year, with sequential growth acceleration of 1 percentage point. OCI (IaaS) revenue $3.3B, up 55% year over year, with sequential acceleration of 3 percentage points. FY26 OCI revenue guided to $18B, up 77% year over year. FY27/FY28/FY29/FY30 IaaS revenue projected at $32B/$73B/$114B/$166B. OCI gross margin expected between 30%-40%. Q3 capex $8.5B. FY26 capex raised from $25B to $35B, vast majority for data center equipment, not land or buildings.
OCI consumption revenue up 57% year over year, persistently supply-constrained. FY26 consumption growth expected to accelerate further. MultiCloud Database revenue from Azure/GCP/AWS grew sequentially, expected to continue sequential growth. Cloud Database revenue up 32% year over year, annualized revenue $2.8B. Autonomous Database consumption revenue up 43% year over year. RPO exceeds $500B, including $20B from Meta and $45B from orders outside of OpenAI.
Currently, the multicloud footprint spanning Azure, AWS, and GCP has 34 data centers online, with 37 more under construction over the next 12 months. OCI is aggressively entering the AI training and inference markets, bullish on the larger future inference opportunity. Its core advantages today are private databases, fast time-to-live, and fast network data transfer, while also delivering Cloud@Customer at 1% of competitors' entry price.
6
CoreWeave
AI NeoCloud
Q3 revenue was $1.36B, up 134% year over year. Gross margin excluding T&I depreciation was 18.2%, operating margin 4%. Q3 adjusted EBITDA was $840M, EBITDA margin 61%. Q3 capex was $1.85B. Due to a suspected delay in Core Scientific data center power shell delivery, full-year capex guidance was sharply lowered from $20-23B to $12-14B, but still well above revenue. Combined with high leverage and high depreciation, the market remains skeptical of profitability.
Q3 RPO was $55.6B, up 271% year over year and 85% sequentially. The company now operates 41 data centers globally, with ~590 MW of available power (QoQ +120 MW) and ~2.9 GW of contracted power (QoQ +0.7 GW). According to official materials, CoreWeave effectively operates on a 'build-to-order' or 'customer-funded capex' model, where data center capacity expansion is directly tied to secured contracts and prepayments.
7
DigitalOcean
AWS for Everyone
Q3 revenue was $230M, up 16% year over year, with AI revenue doubling year over year for the fifth consecutive quarter. ARR was $920M, up 16% year over year. Gross margin was 60%, flat year over year. Operating margin was 20%, up 8 percentage points year over year. Monthly ARPU is no longer disclosed. The company expects full-year 2026 revenue growth of 18-20% year over year.
Q3 EBITDA was $99.79M, up 15% year over year. EBITDA margin was 43%, down 1 percentage point year over year. Net dollar retention rate was 99% this quarter, flat sequentially. RPO was $47M (vs. $58.3M last quarter).
DigitalOcean's customer base is predominantly SMBs, making it highly sensitive to macroeconomic cycles. Previously noted as slow to act amid the rise of AI cloud challengers such as CoreWeave and Nebius, it only opened H100 instances to all customers on October 1 last year and launched its first AI product, GenAI Platform, in early November. It now offers H100, H200, MI300X, and MI325X instances. A positive is its relatively low leverage compared to high-profile neoclouds at this stage.
Conclusion
Cloud capex surged again this quarter. FAMG plus Oracle Q3 capex topped $120B, up 81% year over year, with growth accelerating again. As hyperscalers' compute demand rises sharply, 2026 capex guidance for the five giants continues to call for large increases. Capex concentration on GPU servers will keep rising, yet the market persistently views this as overinvestment. Amazon has repeatedly emphasized this resembles the early days of cloud computing, but argues cloud is currently the highest-ROI scenario for AI, bar none. Another overlooked phenomenon: as AI startups scale, their demand for traditional cloud services also grows.
