Verdict: Microsoft Azure is the default AI in cloud choice for enterprises inside the Microsoft stack, and the gap is widening. Bedrock stays the pick when model breadth decides — it alone serves OpenAI and Anthropic models side by side. Google's Agent Platform wins on Gemini-native agents and the lowest entry price. Azure crossed 100 billion dollars in annual revenue while growing 43 percent in the final quarter of fiscal 2026 (Microsoft), and now hosts Claude models directly on its own infrastructure (Foundry).
The sharper finding: the model layer has stopped differentiating these platforms, so the runtime is what you are actually choosing. Claude now sells on all three clouds, Google's Model Garden carries 200 plus models, and Bedrock's marketplace tops 100 (AWS). Our own article-planning test scored both a Gemini and a Claude model 17 of 17 on machine-checked constraints (n=6, measured 2026-10-10) — same bar, 23 versus 67 seconds median wall time via the Antigravity CLI. Pick the platform for economics, governance, and where your data lives.
TL;DR
- Azure leads AI in cloud for Microsoft-standardized enterprises: 100B dollars annual revenue, Agent 365 near 40 million registered agents; Claude GA (Foundry).
- Bedrock wins the model-breadth tiebreak: 100 plus models, OpenAI and Anthropic via one API.
- Google's Agent Platform is Gemini-native, with a pricing cliff after December 2026 to plan around.
How the three platforms split the market
| Microsoft Azure (Foundry) | Amazon Bedrock | Google Agent Platform | |
|---|---|---|---|
| Scale proof | Azure revenue passed 100B dollars, +43% Q4 (Microsoft) | Independent tracker lists 116 models across 31 regions (Bedrock Explorer) | 6T tokens/month through the ADK (Google) |
| Model layer | Claude GA hosted on Azure; 11,000 plus models claimed in catalog | 100 plus in Bedrock Marketplace; OpenAI Chat Completions and Anthropic Messages compatible endpoints | 200 plus models in Model Garden, Claude included |
| Claude availability | GA, hosted on Azure infrastructure | GA in 25 plus regions, global routing | Listed in Model Garden |
| Agent story | Foundry Agent Service, Agent 365 near 40M registered agents | Bedrock Agents, Flows, Guardrails | Gemini Enterprise app, A2A v1.0 default |
| Best fit | Microsoft-standardized enterprise | Model diversity and region control | Gemini-native agents, lowest entry tier |
That middle row is the story. The Foundry GA post states Claude runs inside Azure on NVIDIA Blackwell Ultra with a zero-data-retention option, and the model lineup runs Haiku to Fable at 1M context. When the same flagships appear on every rack, the differentiator is the model card alone no more — procurement, residency, and identity trust decide.
Why Azure's lead is now structural
The fiscal 2026 results put commercial remaining performance obligation at 678 billion dollars, up 84 percent. Foundry reached 100,000 customers with revenue more than doubling YoY, and trillion-token-run-rate customers grew 4x (Futurum); Copilot passed 30 million paid seats. No other platform here carries that revenue base.
Second reason: Anthropic. The GA post describes the production path enterprises asked for — Azure-native governance, simplified procurement. A platform hosting OpenAI and Anthropic frontier models on its own rails removes the last excuse to multi-cloud for AI.
The cost lever: Foundry's model router routes each query to the right Claude model and saves customers up to 50 percent (Foundry). For teams where token spend is the biggest line item, routing plus caching — not loyalty discounts — moves the budget, as our cloud-cost analysis shows.
Where Bedrock still wins
Bedrock Marketplace offers 100 plus models (AWS); the independent Bedrock Explorer catalogs 116 across 31 regions as of September 15, 2026, including 18 Anthropic models plus OpenAI, Mistral, Meta, DeepSeek and Qwen. No other cloud serves OpenAI and Anthropic flagships through one API.
It also wins on granularity: global, geo, or in-region routing, with regional endpoints at a 10 percent premium (Anthropic Bedrock docs). Mumbai and Hyderabad sit in the Claude region tables — real India residency options. The open-weight roster (DeepSeek, GLM, Kimi K2.5, Qwen3 Coder, Gemma 4) keeps cost-sensitive workloads inside AWS's compliance perimeter. Choosing between models first? Start with our frameworks comparison.
Google's bet, and the December 2026 pricing cliff
Google retired the Vertex AI name on April 22, 2026 and folded its roadmap into the Gemini Enterprise Agent Platform: one 200 plus model catalog, A2A v1.0 as the default inter-agent protocol, agent identity built in, and Claude in the same Model Garden as Gemini. Google says more than six trillion tokens flow monthly through the Agent Development Kit — the highest-throughput agent runtime of the three.
The sharp edge is the pricing calendar. Gemini API pricing lists 3.8 Flash at 0.75 dollars input and 3.75 output per million tokens through December 31, 2026, then 1.50 and 7.50 from January 1, 2027 — a doubling. Price the January 2027 rate in before you standardize, or see Qwen vs Claude vs Kimi for vendor-side economics.
FAQ
Q: Which cloud has the most AI models?
A: Amazon Bedrock serves the broadest menu in practice (100 plus marketplace models; an independent tracker lists 116), Google claims 200 plus in Model Garden including open models (Google), and Azure claims 11,000 plus in its catalog — catalog counts are not comparable, so test the specific models your workload needs (AWS).
Q: Can I run Anthropic Claude on all three clouds?
A: Yes — GA and hosted on Azure infrastructure (Microsoft), GA across the Bedrock region list (Anthropic), and Claude models are listed in Google's Model Garden with Anthropic as publisher (Google).
Q: Is Vertex AI shutting down?
A: No. It is renamed: all Vertex AI services and roadmap updates now ship exclusively under the Gemini Enterprise Agent Platform brand (Google), and existing endpoints keep working.
Q: Which cloud AI platform is cheapest?
A: Google today, on list prices — Gemini 3.8 Flash at 0.75 dollars per million input tokens through December 31, 2026 (Google pricing) — but that doubles on January 1, 2027, so budget the January 2027 rate before you commit.
Q: What changed in cloud AI during 2026?
A: The catalogs converged — Claude now sells on all three clouds (Microsoft) — so the platforms compete on runtime economics, governance, and residency instead of model exclusivity.
Corrections log: none yet. Methodology and AI-use disclosure: How We Work.
We may earn a commission if you purchase through links on this page.
Top comments (0)