Last Updated on April 27, 2026 by Arnav Sharma
Enterprise AI services across the three major cloud providers — Updated April 2026
| Category | Microsoft Azure | Amazon Web Services (AWS) | Google Cloud Platform (GCP) |
|---|---|---|---|
| Unified AI Platform |
Microsoft Foundry Rebrand Formerly Azure AI Studio / Azure AI Foundry. Unified PaaS combining agents, models, tools, and governance under a single portal with Entra RBAC and private networking. |
Amazon SageMaker Unified Studio Integrates SageMaker AI, Bedrock, EMR, Glue, Athena, and Redshift into a single workspace. Built on Apache Iceberg lakehouse architecture. |
Gemini Enterprise Agent Platform Rebrand Formerly Vertex AI. Announced at Cloud Next 2026. Absorbs Vertex AI and Agentspace into a unified platform for building, scaling, governing, and optimising agents. |
| Foundation Model Access |
Foundry Models Azure OpenAI models (GPT-5.5, GPT-5.4, o3, o4-mini), plus DeepSeek, Meta Llama, Mistral, Cohere, xAI Grok, Phi-4, NVIDIA Nemotron. Standard & provisioned deployments. |
Amazon Bedrock Serverless access via single API. Anthropic Claude, Meta Llama, Mistral, Cohere, Stability AI, Amazon Nova 2 models. 18+ open-weight models added April 2026. |
Model Garden 200+ models including Gemini 3.1 Pro, Gemini 3.1 Flash, Gemma 4, Anthropic Claude, Meta Llama, GLM 5, and third-party models. First-party and partner models. |
| Agent Framework |
Foundry Agent Service GA Multi-agent orchestration, supports LangChain, LangGraph, CrewAI, LlamaIndex. Built on OpenAI Responses API. Visual multi-agent workflows in portal. |
Bedrock Agents Autonomous agents with action groups connected to Lambda, APIs, and knowledge bases. Supports Strands Agents framework. Bedrock AgentCore for deployment. |
Agent Studio + Agent Development Kit (ADK) Low-code visual builder (Agent Studio) plus full SDK (ADK). Supports Agent2Agent (A2A) protocol for cross-platform agent communication — live at 150 orgs. |
| RAG & Knowledge |
Foundry IQ (powered by Azure AI Search) Agentic RAG engine with user access permissions. Supports hybrid search, semantic ranking, and vector indexing. CMK encryption at service level. |
Bedrock Knowledge Bases Fully managed RAG pipeline: data ingestion from S3, vector store creation, retrieval, and prompt augmentation. Built-in session context and citations. |
Vertex AI Search + Vector Search 2.0 GA Collections unifying data and vectors, auto-embeddings, hybrid search with built-in semantic re-ranking. Private Service Connect and VPC-SC support. |
| Safety & Guardrails |
Azure AI Content Safety Content filtering, prompt shield (jailbreak detection), groundedness detection, PII redaction. Integrates via middleware with LangChain. |
Bedrock Guardrails Six safeguard policies: content filters, denied topics, PII redaction, word filters, contextual grounding, and Automated Reasoning checks (99% accuracy). Model-agnostic via ApplyGuardrail API. |
Responsible AI Toolkit + Model Armor Safety filters, grounding checks, and evaluation tooling. Gen AI evaluation service supports partner model safety assessments. |
| Custom Model Training |
Azure Machine Learning Part of Foundry. AutoML, custom training with managed compute, Foundry fine-tuning CLI, serverless training pipelines. SDK v2 with YAML-first job definitions. |
Amazon SageMaker AI Full ML lifecycle: training, fine-tuning (SFT, DPO, RLVR, RLAIF), model monitoring, and deployment. Serverless customisation with agent-guided workflows (new 2026). |
Vertex AI Training AutoML + custom training. Serverless training and dedicated training clusters on TPU/GPU. Fine-tuning via Vertex AI Studio. Integration with BigQuery and Dataproc Spark. |
| Edge / Local Deployment |
Foundry Local Run models on-device and at the edge. Expanded from SLMs to large multimodal models on NVIDIA GPU hardware (Feb 2026). Zero cloud connectivity option. |
SageMaker Edge + Inferentia Deploy models to edge devices with SageMaker Edge Manager. AWS Inferentia3 chips for cost-efficient inference at scale. |
Vertex AI on GDC / Edge TPU Google Distributed Cloud for on-prem deployments. Edge TPU for low-power inference on edge devices. |
| Interoperability Protocols |
MCP Server (cloud-hosted at mcp.ai.azure.com)Native MCP support with Entra auth. Integration with VS Code, Visual Studio, and the Foundry portal. 1,400+ tool integrations. |
MCP Support in Bedrock Bedrock Responses API supports server-side tools. Integration with Strands Agents and third-party MCP servers. |
MCP + Agent2Agent (A2A) Managed MCP servers for Maps, BigQuery, GKE, Cloud Run, and more. Apigee acts as an MCP bridge. A2A protocol enables cross-vendor agent orchestration. |
| Speech & Language APIs |
Foundry Tools (formerly Cognitive Services) Speech-to-text, text-to-speech, translation, language understanding, document intelligence, and computer vision APIs. |
Amazon AI Services Transcribe (speech-to-text), Polly (text-to-speech), Translate, Comprehend (NLP), Textract (document extraction), Rekognition (vision). |
Cloud AI APIs Speech-to-Text (125 languages), Text-to-Speech (220+ voices), Translation, Natural Language API, Document AI, Vision AI. |
| No-Code / Low-Code |
Foundry Portal + Copilot Studio Visual agent builder in Foundry portal. Copilot Studio for business-user agent creation across M365 ecosystem. |
Bedrock Console + PartyRock Visual agent builder in Bedrock console. PartyRock for no-code generative AI app prototyping. |
Google Workspace Studio New No-code agent builder for Gmail, Docs, Sheets, and third-party services (Salesforce, Jira). Available for Business, Enterprise, and Education tiers. |
| Pricing Model | Platform is free to explore. Pay per deployment: token-based for inference, compute-based for training. Standard and provisioned throughput options. | Serverless pay-per-token for Bedrock. SageMaker billed by compute instance hours. Provisioned throughput and on-demand options. Reserved instances available. | Pay-per-use for API calls. Free tier available for Agent Engine Runtime. Training billed by compute hours (TPU/GPU). Committed use discounts available. |
| Custom AI Silicon | Partners with NVIDIA (A100, H100, H200, B200). Azure Maia AI Accelerator (custom chip, limited rollout). | AWS Trainium (training) + Inferentia (inference). Trainium2 and Inferentia3 in rollout. 40–60% cost savings over GPU for supported workloads. | Cloud TPU (v5e, v6e, Ironwood). Purpose-built for Gemini model family. Trillium TPU architecture announced at Cloud Next 2026. |
Arnav Sharma
Microsoft MVPMCT
Microsoft Certified Trainer · Cloud · Cybersecurity · AI
I help organisations secure their cloud infrastructure and stay ahead of evolving cyber threats. Microsoft MVP and Certified Trainer, author of Mastering Azure Security, and founder of arnav.au — a platform for practical Cloud, Cybersecurity, DevOps and AI content.