Three hyperscalers, one converging playbook, and the single question that actually decides where your AI runs. 

The “who has the best model” race is both permanent and unwinnable; the leapfrogging never stops. What actually decides where your AI lives in 2026 is gravity: where your data, identity, and people already sit. Microsoft, Google, and Amazon now sell the same five-layer stack. The product names differ, the integration story differs, and that difference is the whole game. 

If you’ve tried to compare the three clouds’ AI offerings in the last six months, you’ve felt the whiplash. Services get renamed, bundled, and re-bundled faster than anyone can document them. So let’s strip away the noise and look at what is structurally true. 

What we’re seeing: convergence, then divergence 

Eighteen months ago, the buying decision was “which foundation model?” That question has become secondary.^[6] Each platform has matured from a model-hosting service into a full-stack AI development environment covering everything from model catalog, build tools, agent orchestration, retrieval pipelines, and governance, all under one roof. Put plainly: the choices look almost identical. 

The divergence shows up one layer down, in the integration story. Microsoft’s pitch is that AI should meet your people where they already work, inside Teams, Outlook, and SharePoint, and that identity and governance should be the same plane you already run.^[5] Google’s pitch is focused on data gravity: if your analytics and your workplace already live in BigQuery and Workspace, the shortest path to value runs through Google. Amazon’s pitch is breadth and neutrality: the widest model marketplace of the three, sitting next to the infrastructure you are probably already paying for.^[6] 

When you net this all out, we find that nobody is choosing a platform on model quality alone anymore. They’re choosing the ecosystem their organization is already standing in, and that’s not a bad place to start (or end up). 

The five layers every platform now has 

To compare these systems without drowning in product names, it helps to think in layers. Each of the three clouds now offers the same five: a foundation-model platform for building and running AI; an enterprise assistant for end users; applied AI services for targeted tasks like vision, speech, and document extraction; a safety and guardrails layer; and a governance and identity plane wrapped around all of it. Match the layer first, then the vendor’s name for it. That is the trick to keeping this straight. 

Layer-by-layer mapping — current as of June 2026 

The key signal: it’s all about agents now (and the names are moving) 

The most important shift this year isn’t a better chatbot, and it’s not just the move from assistants that respond to agents that act; systems that plan, call tools, and execute multi-step work across your applications.[6] Microsoft put down the loudest marker this week at Build (June 2), unveiling Scout, pitched not as another Copilot but as an “Autopilot”: an always-on agent with its own Entra identity that works across Teams, Outlook, and the Windows desktop, and takes action without waiting to be asked.[12] All three vendors have re-architected around this, and the branding now reflects it. 

Watch this one closely, because it is a live trap. Google retired the “Vertex AI” name and rebranded the whole platform as the Gemini Enterprise Agent Platform, folding in its Agentspace product. As of mid-2026, “Vertex AI” no longer appears in the Google Cloud console; model training and registry are now sub-features under an agent-first hierarchy.^[2][4] On the Microsoft side, “Azure AI Studio” became Foundry, and “Azure AD” became Entra ID. And if you’re pursuing certification across these suites and reading older documentation or training material, you are reading legacy names.  

Where this matters: governance is going cross-cloud 

Here is the genuinely unconventional move, and it is worth more attention than the model benchmarks get. The governance layer is breaking out of its own cloud. Microsoft Purview can now extend its data-loss-prevention policies to AI agents running on Amazon Bedrock and Google Cloud, establishing one central policy engine governing AI that runs anywhere.^[9] Amazon’s Bedrock Guardrails apply the same safety controls across third-party and other clouds’ models, not just AWS-hosted ones.^[3] 

With this, the “pick one cloud” framing is already loosening at the control plane. The strategic bet for serious organizations is not a single vendor, it is a single governance and identity spine that can reach across all of them. That matters even more with the EU AI Act’s high-risk obligations applying from August 2026, which turns “we have a policy document” into “show me the enforcement log on every prompt and response.”^[10] 

So how should you actually choose? 

When evaluating a platform choice, skip the feature-checklist bake-off; on paper, they will tie. Decide on gravity and lock-in instead. If your workforce lives in Microsoft 365, Copilot and Foundry give you the shortest time-to-value and the cleanest identity story through Entra.^[11] If your data warehouse is BigQuery and your people are in Workspace, Gemini Enterprise removes the most friction. If you are AWS-native, or you want the widest model choice with vendor-neutral framing, Bedrock plus Amazon Q is the pragmatic pick.^[5] Whichever you choose, make the governance plane a first-class, cross-cloud decision, not an afterthought bolted on later. 

Choosing the wrong platform does not just cost money. It locks you into an SDK ecosystem, a compliance boundary, and a set of model constraints that can take 6–12 months or more to unwind.^[6]  And it will play havoc with your adoption and transformation efforts.  

In conclusion, stop shopping for the smartest model and start mapping your own gravity. Inventory where your data, identity, and people already sit; pick the platform that bends toward them; and stand up a governance spine that does not care which cloud the AI happens to run in. Do that, and the vendor question mostly answers itself. 

References 

1. Microsoft. Learn about Microsoft Purview. Microsoft Learn. learn.microsoft.com/en-us/purview/purview 

2. Google Cloud. Introducing Gemini Enterprise Agent Platform. Google Cloud Blog, Apr. 2026. cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise-agent-platform 

3. Amazon Web Services. Amazon Bedrock Guardrails. aws.amazon.com/bedrock/guardrails 

4. RoboRhythms. Google Retired Vertex AI for Agent Platform in May 2026. May 2026. roborhythms.com/gemini-enterprise-agent-platform-launch 

5. EPC Group. Copilot vs Gemini vs AWS Q: Enterprise AI 2026. epcgroup.net/blog/microsoft-copilot-vs-gemini-vs-aws-q-enterprise 

6. Bits Lovers. Amazon Bedrock vs Azure AI Foundry vs Google Vertex AI: 2026 Deep Comparison. Apr. 2026. bitslovers.com/bedrock-vs-azure-ai-foundry-vs-vertex-ai 

7. Arnav. Cloud AI Platform Comparison: Azure vs AWS vs GCP. Apr. 2026. arnav.au/2026/04/27/cloud-ai-platform-comparison-azure-vs-aws-vs-gcp 

8. Methuku, A. Azure vs AWS vs Google Cloud: Search and Vision AI Services. Medium, Mar. 2026. medium.com/@anjaiahspr 

9. Microsoft. Extend Microsoft Purview data protection to Amazon Bedrock agents. Microsoft Community Hub, June 2026. techcommunity.microsoft.com/…/4525984 

10. Maxim AI. Top 5 AI Guardrails Platforms for Responsible Enterprise AI in 2026. Apr. 2026. getmaxim.ai/articles/top-5-ai-guardrails-platforms-for-responsible-enterprise-ai-in-2026 

11. Kovil AI. Azure AI Foundry vs AWS Bedrock vs Google Vertex AI — Enterprise Platform Comparison 2026. kovil.ai/azure-ai-foundry/compare/vs-aws-bedrock 

12. Microsoft. Introducing Microsoft Scout: Your always-on personal agent. Microsoft Build 2026 newsroom, June 2026. news.microsoft.com/build-2026