Key takeaways
- All three big clouds reported their September quarter this week, and all three accelerated. Google Cloud revenue roughly doubled to about US$30.7B (+103%), Azure grew 48%, and AWS grew 39% to about US$46B, which Tech Insider reports is its fastest pace in nearly five years.
- AI is the main driver, and it is soaking up capacity. Google says it processes nearly 32 billion tokens per minute; reported backlogs run to hundreds of billions of dollars at Google Cloud and Microsoft.
- Cloud is now very profitable for the providers. AWS operating income was about US$17.8B and Google Cloud's about US$11.1B, a reported margin of around 36–39% for both. That margin is your negotiating room.
- There are concrete, time-boxed price moves right now: Gemini 4 Argon introductory pricing ends 30 November, Microsoft CSP promotions of up to 30% start 2 November, and AWS cut the EMR-on-EKS charge by 30%.
- For Indian teams, the message is: lock in what is stable, keep AI workloads portable, and renegotiate before the year-end. Do not sign a three-year commit on GPU capacity because of headlines.
1. The numbers, side by side
These are the headline figures for the quarter ending 30 September 2026, as reported across Tech Insider, AFP via The Manila Times and FourWeekMBA, which works from the companies' filings.
| AWS | Microsoft Azure | Google Cloud | |
|---|---|---|---|
| Quarterly revenue | ~US$46.0B | ~US$33.2B (reported, from 10-Q table) | ~US$30.7B |
| Year-on-year growth | 39% | 48% | 103% |
| Added in one year | ~US$13.0B | ~US$10.8B | ~US$15.6B |
| Operating income | ~US$17.8B | Not broken out | ~US$11.1B |
| Reported results date | 29 Oct | 28 Oct | 28 Oct |
| India regions | Mumbai, Hyderabad | Central, South, West India | Mumbai, Delhi NCR |
Two points stand out. First, Google Cloud added more new revenue in a year than either rival, so it is no longer the distant third. Second, AWS is still the largest by a clear margin and still growing faster than it has in years. This is not a market where one provider is collapsing. All three are capacity-constrained and pricing from a position of strength.
2. What is driving the growth
AI inference, mostly. The reports point to the same pattern at each provider:
- Google cites token volume of nearly 32 billion per minute, up more than 4x in a year, and a reported Cloud backlog of about US$561B.
- AWS reportedly told investors that Fortune 500 spending on Bedrock rose nearly 7x year on year. Management projected AWS growth "over 40%" for 2027 and 2028, as reported by FourWeekMBA.
- Microsoft reported commercial remaining performance obligations of US$677B, up 73%. That figure covers all commercial business, not just Azure.
Capital spending is following. Amazon expects about US$220B of cash capex in 2026, mostly for AWS, and Microsoft nearly doubled its quarterly property-and-equipment spending to about US$39B (AFP).
What this means for you: GPU and accelerator capacity will stay tight and expensive, while ordinary compute, storage and databases are commodity products the providers compete on. Treat the two differently in your budget.
3. Price moves you can use this month
The earnings week came with several concrete changes. None of these is a reason to migrate, but each is worth checking against your bill.
| Change | Provider | Detail | Deadline |
|---|---|---|---|
| Gemini 4 Argon intro price | Google Cloud | US$2 / 1M input, US$10 / 1M output tokens; rises to US$4 / US$20 from 1 Dec (reported) | Ends 30 Nov |
| CSP promotions | Microsoft | Up to 30% on Azure database modernisation, up to 20% on Fabric capacity, up to 15% on App Platform and data services (reported) | From 2 Nov |
| EMR on EKS | AWS | EMR charge cut 30% per vCPU-hour and per GB-hour; EKS pricing unchanged (AWS) | Live |
| GPU sharing on ECS | AWS | NVIDIA MPS lets several ECS tasks share one GPU (AWS) | Live |
| Aurora Serverless (PostgreSQL) | AWS | Max scale raised to 384 ACUs from 256 (AWS) | Live |
The Gemini deadline matters most for Indian startups testing AI features. If you are running evaluations on Gemini 4 Argon, the price doubles on 1 December. Budget for the standard price now, not the trial price. As an illustration at ₹88 per US dollar, a workload of 100 million input tokens and 20 million output tokens a month costs about ₹35,000 at the intro price and about ₹70,000 at the standard price, before GST and before caching discounts.
The Microsoft CSP promotions reach most Indian SMBs through a partner, because many buy Azure through a reseller rather than directly. Ask your CSP partner which promotions apply to your tenant and whether they will pass them through.
4. The India angle
Data residency is solved for standard services, not for frontier AI. All three providers run at least two regions in India, which covers most DPDP Act and sector-regulator needs for databases, storage and compute. But many of the newest AI models are still served from outside India or through global routing. Check each model's region list before you send customer data to it.
Rupee costs move with the dollar. All three bill Indian customers in INR, but list prices are set in US dollars, so a weaker rupee raises your bill even when nothing changes. A one-year commitment priced in INR through a partner can reduce that risk; ask for it explicitly.
The SMB push is real. Google Cloud held Cloud Day India this week with more than 600 founders and executives, focused on agents for tasks such as reconciling e-way bills against vendor invoices and connecting Tally or SAP data to BigQuery (The420.in). When providers court a segment this hard, credits and pilot funding are usually negotiable. Ask.
Hiring follows spend. Cloud and FinOps skills tied to AI workloads (GPU scheduling, token cost monitoring, Bedrock/Vertex/Foundry governance) are where demand is moving. If you are an engineer, a cost-optimisation project on your own employer's bill is a strong portfolio piece.
5. Six moves before your Q4 renewal
- Split your bill into "commodity" and "AI". Tag compute, storage and databases separately from GPU instances and model API calls. You negotiate them differently.
- Commit on the commodity side only. One-year Savings Plans, Reserved Instances or committed-use discounts on stable workloads are low-risk. Avoid long GPU commitments unless you have a contract to match them.
- Re-price AI calls at standard rates. Any business case built on intro or preview pricing needs a second column at full price. Gemini's 1 December step-up is the obvious example.
- Use caching and smaller models first. Cached input is heavily discounted on most platforms, and a mid-tier model often does the job. Measure quality on your own data before paying for the flagship.
- Ask for credits in writing. With providers this profitable and this eager for AI workloads, startup and migration credits are easier to get. Get them in the order form, not an email.
- Keep AI code portable. Use a thin gateway layer so you can switch between Bedrock, Vertex AI and Foundry. The same models increasingly run on all three, so the switching cost is mostly in your code.
6. What to watch next
- Microsoft Ignite, 17–20 November: expect Azure AI pricing and Copilot licensing changes. New Microsoft 365 Copilot purchases move to new offers from 1 December (reported).
- AWS re:Invent (usually late November to early December): historically where new instance types and price cuts land.
- Indian region availability for new models: the single most important item for regulated Indian businesses.
The short version: the providers are winning big on AI, and they have room to give you a better deal on everything else. Use the year-end to take it.
Want a second opinion on your AWS, Azure or Google Cloud bill before renewal? See our data centre and cloud services or estimate costs with our free tools.