AWS vs Azure: 26% vs 19% Share, 39% vs 48% Growth in Q3 2026

AWS booked roughly $46.0 billion in Q3 2026 cloud revenue, up 39% year over year, while Azure reached about $33.2 billion, up 48%, according to CRN's quarterly earnings face-off. Google Cloud posted $30.7 billion, up 103% — roughly double its $15.2 billion from a year earlier.

Synergy Research Group share data, relayed by CRN, shows the drift. AWS held 26% of worldwide cloud infrastructure spending, down from 29%. Microsoft held 19%, down from 20%. Google rose from 13% to 16%. The total market hit about $168 billion in the quarter, up roughly 55%.

One caution on the Microsoft line: Synergy's 19% covers Microsoft's cloud infrastructure business, not pure Azure. Microsoft reports Azure inside a larger segment and discloses growth rates more readily than dollar totals, so treat $33.2 billion as press-reported.

What does the growth gap change for buyers?

Faster growth funds more data centers, custom chips and discounts. A cloud adding 48% a year builds at a different pace than one adding 39%, even when the slower one holds more customers. Press coverage, including a TechGig report, credits AI workloads with roughly one-third of the global cloud infrastructure market.

CNBC's AWS earnings coverage pegs Azure at about 72% of AWS's size, versus 58% two years earlier — a 14-point structural jump. AWS still runs near a $184 billion annualized rate per TelecomTV. Share loss doesn't mean shrinkage: 39% growth on that base adds more absolute dollars than most rivals' entire cloud units. The useful reading: AWS is no longer the default winner of every workload, and buyers have real leverage.

Pricing: egress, serverless, Kubernetes

All prices come from vendor pages fetched November 14, 2026.

• Egress: Both clouds allow the first 100 GB per month free. Azure charges $0.087 per GB in North America and Europe on its premium network ($0.08 on the ISP route) against AWS's $0.09 — a roughly 3% gap. Azure's regional rates dominate: $0.12 per GB in Asia, Australia, the Middle East and Africa; $0.181 in South America.
• Serverless: Identical free grants — 1 million requests and 400,000 GB-seconds monthly. Lambda lists $0.0000166667 per GB-second (x86) and $0.20 per million requests.
• Object storage: S3 Standard runs $0.023 per GB-month for the first 50 TB in US East (N. Virginia), plus $0.005 per 1,000 PUT and $0.0004 per 1,000 GET requests.
• Kubernetes: EKS charges $0.10 per cluster per hour in standard support, $0.60 in extended support — six times the rate for stale versions. AKS offers a free tier with no SLA (not recommended above 10 nodes), a Standard tier with a financially backed SLA, and Premium adding long-term support.

Custom silicon and AI platforms

Both vendors design their own chips. AWS fields Graviton (Arm CPUs) and Trainium, which Anthropic uses as one of four chip families, per its IPO filing. Microsoft's Cobalt 200 is its second-generation Arm CPU; the company claims 50%+ better performance than Cobalt 100, 33% lower agent-call latency and 23% higher agentic throughput — vendor claims without independent confirmation. Maia 200 serves production workloads including Copilot.

On AI platforms, Microsoft Foundry lists more than 10,000 models and adds about 50 monthly; Amazon Bedrock positions itself as a neutral model marketplace. The largest AI labs split spending across several clouds — OpenAI reportedly sources from five, and its AWS agreement expansion of $100 billion over eight years comes from a secondary source and remains unconfirmed. Capacity, not model exclusivity, is the scarce resource; buyers needing large GPU blocks should request written capacity commitments.

Reliability

The best-documented incident in the window is AWS's October 2025 us-east-1 DynamoDB failure: about 14 hours and 32 minutes (11:48 p.m. PDT October 19 to 2:20 p.m. PDT October 20). A latent race condition in DynamoDB's automated DNS management created an empty DNS record for the regional endpoint; automation failed to repair it. AWS restored DNS by 2:25 a.m., with cached records expiring by 2:40 a.m. Multi-AZ deployments didn't help — the fault sat in a regional DNS record; only tested multi-Region failover would have limited impact. No comparable primary-source Azure outage record exists for the window, though Microsoft 365 failures in November 2026 drew more than 130,000 Downdetector reports and hit four Nevada agencies.

Verdict

A split decision. Microsoft-centric enterprises fit Azure; teams starting from zero or needing the widest ecosystem fit AWS, whose new-customer offer reportedly reaches $200 in credits plus a six-month free plan. Kubernetes-heavy teams should compare AKS's free tier against EKS's flat, predictable fee. Either way: negotiate egress terms, upgrade Kubernetes on schedule, and test second-Region failover before you need it. If projected migration savings run under 15% of annual spend, negotiate with your current vendor instead.