Cloud Infrastructure

Cloud AI Agents

AWS, Azure, Google Cloud, and Databricks Solutions for 2026

Last updated: January 202618 min read

Key Takeaways

  • AWS Bedrock AgentCore offers consumption-based pricing where I/O wait is free, with 13 pre-built evaluation systems and Policy controls for enterprise security
  • Azure AI Foundry Agent Service reached GA with 10,000+ customers, featuring one-click deployment to Teams and Deep Research capabilities
  • Google Vertex AI Agent Development Kit (ADK) has 7M+ downloads; Wells Fargo became the first major bank to deploy AI agents company-wide
  • Databricks Mosaic AI introduced Agent Bricks for auto-optimized agents and Storage-Optimized Vector Search with 7x lower cost

CLOUD AI AGENTS BY THE NUMBERS (2026)

10,000+
Azure AI Foundry customers
7M+
Google ADK downloads
7x
Databricks Vector Search cost reduction
50%
NTT DATA faster time to market

Sources: Microsoft, Google Cloud, Databricks

Why Cloud Platforms for AI Agents

Cloud AI agents represent a major shift in how enterprises deploy autonomous AI systems. Building production-grade AI agents requires sophisticated infrastructure for orchestration, memory management, tool integration, security, and observability. The major cloud providers—AWS, Microsoft Azure, and Google Cloud—have all launched comprehensive platforms specifically designed for cloud AI agents, while Databricks offers specialized capabilities for data-centric agent workflows.

According to Gartner, 40% of enterprise applications will feature task-specific AI agents by the end of 2026. However, they also predict over 40% of agentic AI projects will be canceled by 2027 due to escalating costs, unclear business value, or inadequate risk controls. This makes platform selection critical for enterprise success. For SaaS-based enterprise platforms like Salesforce Agentforce and ServiceNow, see our guide to enterprise AI agents.

What Cloud AI Agent Platforms Provide

  • Runtime Infrastructure: Secure, serverless deployment with automatic scaling
  • Tool Integration: Unified access to APIs, databases, and enterprise systems
  • Memory Systems: Session context and long-term memory management
  • Security and Governance: Enterprise-grade identity, authentication, and policy controls
  • Observability: Monitoring, debugging, and evaluation capabilities

Platform Comparison Overview

Before diving into individual platforms, here is a high-level comparison of cloud AI agent offerings from the major providers:

PlatformPrimary StrengthBest ForStatus
AWS Bedrock AgentCoreMulti-model access, enterprise securityAWS-native enterprises, regulated industriesGA Dec 2025
Azure AI FoundryMicrosoft 365 integration, Entra governanceMicrosoft-centric organizationsGA May 2025
Google Vertex AIObservability, Gemini integrationGCP users, data-centric workflowsGA Mar 2025
Databricks Mosaic AIData lakehouse integration, MLOpsData teams, RAG applicationsGA
AWS

AWS Bedrock Agents and AgentCore

Amazon cloud AI agent infrastructure

AWS AI agents through Amazon Bedrock enable developers to build applications that respond dynamically by accessing enterprise systems, executing actions, and retaining memory across sessions. At AWS re:Invent 2025, Amazon launched AgentCore—a comprehensive platform for building, deploying, and operating enterprise-scale agents without infrastructure management.

AgentCore Components

  • Runtime: Secure, serverless deployment with consumption-based pricing
  • Gateway: Unified tool access and connections to enterprise systems
  • Memory: Intelligent context retention across sessions
  • Identity: Authentication across AWS and third-party services
  • Browser and Code Interpreter: Enhanced agent capabilities

Enterprise Features

  • Policy: Natural language boundaries for agent actions
  • Evaluations: 13 pre-built evaluation systems for quality scoring
  • Observability: Comprehensive monitoring and debugging
  • Flows: Visual workflow builder for multi-step agents

Unique Pricing Model

Unlike traditional compute services, AgentCore Runtime charges only for active resource consumption. Agentic workloads typically spend 30-70% of time in I/O wait (waiting for LLM responses, API calls). With AgentCore, I/O wait time is free, delivering substantial cost savings compared to pre-allocated instance pricing.

Azure

Azure AI Foundry Agent Service

Microsoft cloud AI agent platform

Azure AI agents through the Foundry Agent Service empower developers to securely design, deploy, and scale AI agents with ease. The platform reached general availability in 2025 and has been adopted by over 10,000 customers. Azure AI Foundry agents work with a rich ecosystem of models from the Foundry model catalog, knowledge sources like Bing, SharePoint, and Azure AI Search, plus 1,400+ action connectors via Azure Logic Apps.

Multi-Agent Workflows

Enable dynamic collaboration among specialized agents with built-in orchestration

One-Click M365 Deployment

Instantly deploy agents to Microsoft 365 Copilot and Teams Chat without manual setup

Computer Use Tool

Perform tasks by interacting with computer systems through their user interfaces

Latest Capabilities (2025-2026)

FeatureDescriptionStatus
Deep ResearchOpenAI's advanced agentic research capability via API and SDKPreview
Browser AutomationReal-world browser tasks via Playwright WorkspacesPreview
Hosted AgentsFully managed agent hosting with built-in memoryGA
MCP and A2A SupportOpen standards for agent interoperabilityGA

Customer Success: NTT DATA leveraged Azure AI Foundry Agent Service to automate routine customer service inquiries, resulting in 50% faster time to market and improved productivity across functions.

Source: Microsoft Tech Community

GCP

Google Vertex AI Agent Builder

Google Cloud AI agent platform

Google AI agents through Vertex AI Agent Builder provide a comprehensive and open platform to build, scale, and govern reliable agents in production. The Python Agent Development Kit (ADK) has been downloaded over 7 million times since Agent Builder's public inception, demonstrating strong developer adoption. Google Vertex AI agents 2025 updates include general availability of Agent Engine, new regions, and the Gemini 2.5 Computer Use model.

Build Capabilities

  • Agent Development Kit (ADK): Open-source framework for multi-agent systems
  • Agent Designer: Low-code visual designer for rapid prototyping
  • Agent Garden: Library of prebuilt sample agents and tools
  • Memory Bank: Topic-based memory using ACL 2025 research

Security and Governance

  • Threat Detection: Built-in Security Command Center integration
  • Agent Identity: IAM-based security and access management
  • Private VPC: Deploy agents in private network environments
  • HIPAA Support: Agent Engine now supports HIPAA workloads

Gemini 2.5 Computer Use

The Gemini 2.5 Computer Use model and tool is now available in Preview. It enables applications to interact with and automate tasks in the browser, including:

  • Automating repetitive data entry or form filling
  • Navigating websites to gather information
  • Performing action sequences in web applications
  • Testing and QA automation workflows
DB

Databricks Mosaic AI Agents

Data-centric AI agent platform

Databricks AI agents through Mosaic AI provide a unified platform for building production-quality AI agent systems integrated with enterprise data. At the Data + AI Summit 2025, Databricks announced major updates including Agent Bricks, MLflow 3.0, and Storage-Optimized Vector Search with 7x lower cost.

Agent Bricks (New)

A new way to build high-quality agents that are auto-optimized on your data. Simply provide a description and connect enterprise data—Agent Bricks handles the rest.

  • Structured information extraction
  • Reliable knowledge assistance
  • Custom text transformation
  • Multi-agent systems

MLflow 3.0

Redesigned from the ground up for Generative AI with monitoring, evaluation, and lifecycle management. Monitor agents deployed anywhere, even outside Databricks.

  • Cross-platform agent observability
  • Agents on AWS, GCP, or on-premise
  • Unified governance controls
  • GenAI-specific evaluation metrics

Platform Integrations

  • MCP Integration: Anthropic's Model Context Protocol integrated directly into the platform with MCP servers hosted on Databricks Apps
  • Vector Search: Storage-Optimized Vector Search scales to billions of vectors with 7x lower cost
  • Mosaic AI Gateway: Generally available for unified model access and governance

Customer Success: Block built an AI agent system using Mosaic AI to automate operations for sellers, such as generating customized menus. Intercontinental Exchange (ICE) built an agent system that securely uses unique financial data to provide highly accurate answers to customer questions.

Source: Databricks Blog

Case Study: Wells Fargo and Google Cloud AI Agents

WF

Wells Fargo Google Cloud AI Agents

First major commercial bank to comprehensively adopt AI agents

In August 2025, Wells Fargo announced an expansion of their strategic relationship with Google Cloud to transform how the bank uses and deploys agentic AI at scale. As an early adopter of Google Agentspace, Wells Fargo became one of the first major banks to deploy AI agents company-wide.

Use Cases Deployed

  • FX post-trade inquiry triage and summarization
  • Policy and procedure navigation across internal systems
  • 250,000 vendor agreement documents queried by custom agent
  • 24/7 customer service with hyper-personalization

Employees Equipped

  • Branch bankers
  • Investment bankers and traders
  • Marketing teams
  • Customer relations and corporate teams
"Wells Fargo's adoption of Google Agentspace marks a bold step forward in making banking simpler and smarter—for our customers and employees."
— Tracy Kerrins, Consumer CIO and Head of Enterprise Generative AI, Wells Fargo

Sources: Google Cloud Blog, Banking Dive, PYMNTS

Pricing Comparison

Cloud AI agent pricing varies significantly by provider and component. Here is a side-by-side comparison of key infrastructure costs:

ComponentAWS AgentCoreAzure AI FoundryGoogle Vertex AI
Runtime (vCPU/hour)$0.0895No additional charge$0.00994
Memory (GiB/hour)$0.00945Pay for underlying services$0.0105
Workflow Transitions$0.035/1KVaries by serviceVaries by usage
Free Tier$200 creditsFree to explore$300 credits + Express Mode
I/O Wait BillingFreeN/AStandard billing

Model Inference Costs Add Up

The prices above cover infrastructure only. Model inference costs (per token/request) are billed separately and often represent the largest portion of AI agent expenses. Claude, GPT-4, and Gemini each have different pricing that varies by context length and model version.

For current pricing, see official documentation: AWS | Azure | Google Cloud

How to Choose Your Platform

Selecting the right cloud platform for AI agents depends on your existing investments, team capabilities, and specific use cases:

Choose AWS Bedrock AgentCore if...

  • Your organization is AWS-first with existing VPC, IAM, and infrastructure investments
  • You need enterprise-grade security controls and compliance certifications
  • You want access to Claude, Llama, and Amazon Nova models in one platform
  • Consumption-based pricing with free I/O wait aligns with your workloads

Choose Azure AI Foundry if...

  • Your organization uses Microsoft 365, Teams, and Copilot extensively
  • You need one-click agent deployment to collaboration tools
  • You want access to both OpenAI and Anthropic frontier models
  • You need 1,400+ pre-built connectors via Azure Logic Apps

Choose Google Vertex AI if...

  • You prioritize observability and want built-in dashboards for agent monitoring
  • You want deep Gemini integration including Computer Use capabilities
  • You prefer low-code visual tools alongside pro-code SDKs
  • You need HIPAA compliance for healthcare workloads

Choose Databricks Mosaic AI if...

  • You have existing data lakehouse investments and need unified governance
  • Your agents need deep integration with enterprise data
  • You want auto-optimized agents via Agent Bricks
  • You need to monitor agents across multiple clouds with MLflow 3.0

Frequently Asked Questions

What are cloud AI agents?

Cloud AI agents are autonomous AI systems hosted on major cloud platforms like AWS, Azure, and Google Cloud. These services provide the infrastructure to build, deploy, and scale AI agents that can plan tasks, use tools, access enterprise data, and execute actions with built-in security and governance. Examples include AWS Bedrock Agents, Azure AI Foundry Agent Service, and Google Vertex AI Agent Builder.

Which cloud platform is best for AI agents?

The best platform depends on your existing infrastructure: AWS Bedrock AgentCore suits AWS-native enterprises needing multi-model access and enterprise security. Azure AI Foundry is ideal for Microsoft 365 organizations with one-click Teams deployment. Google Vertex AI excels in observability and offers a free tier. Databricks Mosaic AI is best for data teams with existing lakehouse investments.

How much do cloud AI agents cost?

Pricing varies by provider: AWS AgentCore charges $0.0895 per vCPU-hour for runtime. Google Vertex AI Agent Engine costs $0.00994 per vCPU-hour. Azure AI Foundry Agent Service has no additional charge—you pay for underlying compute and models. All providers bill separately for LLM inference, which often represents the largest cost component.

What is Azure AI Foundry Agent Service?

Azure AI Foundry Agent Service is Microsoft's platform for building and deploying AI agents at scale. It reached general availability in May 2025 and has been used by over 10,000 customers. Key features include multi-agent orchestration, one-click deployment to Microsoft 365 and Teams, Computer Use tool, Deep Research capabilities, and 1,400+ action connectors via Azure Logic Apps.

How is Wells Fargo using Google Cloud AI agents?

In August 2025, Wells Fargo expanded their Google Cloud partnership to deploy AI agents company-wide using Google Agentspace. Use cases include FX post-trade inquiry triage, querying 250,000 vendor contract documents, policy navigation, and 24/7 personalized customer service. Wells Fargo became one of the first major commercial banks to comprehensively adopt AI agents across operations.

Summary: Cloud AI Agent Platforms

AWS BEDROCK AGENTCORE

Enterprise security with consumption-based pricing. Free I/O wait billing, 13 evaluation systems, and multi-model access including Claude and Llama.

AZURE AI FOUNDRY

10,000+ customers with deep M365 integration. One-click Teams deployment, Deep Research, and 1,400+ Logic Apps connectors.

GOOGLE VERTEX AI

7M+ ADK downloads with strong observability. Gemini 2.5 Computer Use, Memory Bank research, and HIPAA compliance support.

DATABRICKS MOSAIC AI

Data-centric agents with Agent Bricks auto-optimization. MLflow 3.0 cross-platform monitoring and 7x lower cost Vector Search.

Navigate the Cloud AI Landscape

At Planetary Labour, we help organizations evaluate and implement cloud AI agent platforms that deliver measurable business outcomes while managing risk.

Explore Planetary Labour →

Continue Learning