Google: The Model-First Enterprise Strategy

Part 5 of the series "The Enterprise AI Vendor Landscape"

Can Gemini Enterprise Unite Google's Fragmented AI Portfolio?

The Consolidation Signal

Google has the strongest foundation models in the enterprise AI market and the weakest enterprise platform story. That tension has defined Google Cloud's competitive position for years, and the Gemini Enterprise rebrand is the most direct attempt yet to resolve it.

The rebrand happened in two phases. In October 2025, Google renamed Agentspace to Gemini Enterprise, consolidating its enterprise search and AI-powered work capabilities under the Gemini brand. Then at Cloud Next 2026 in April, Vertex AI became the Gemini Enterprise Agent Platform, absorbing all of Vertex AI Agent Builder's services into a unified development surface. The consumer-facing and builder-facing sides of Google's AI portfolio now share a single brand identity: Gemini Enterprise for consuming AI capabilities, and the Agent Platform for building them.

The rebrand is not just marketing. It signals a strategic pivot from Google's historical pattern of launching separate products for adjacent problems (and frequently killing them) toward a consolidated enterprise platform that can compete with Microsoft's governance-first approach and Salesforce's application-layer strategy. Google says existing Agent Builder customers required no migration; billing rolled from Vertex AI line items to Gemini Enterprise in mid-2026.

The financial trajectory supports the ambition. Google Cloud revenue reached $24.8 billion in Q2 2026, an 82 percent year-over-year increase that outpaced both Azure (43 percent growth) and AWS (37 percent growth in the same quarter). Google Cloud's market share in global cloud infrastructure climbed to 15 percent, up from 13 percent a year earlier. The backlog stands at $514 billion in committed but unrecognized revenue.

The question is whether Google can convert model leadership and revenue momentum into the kind of enterprise platform coherence that Microsoft and AWS have already established. Google's challenge is not technology. It is execution, trust, and the organizational discipline to stop building new products and start building a platform.

What Gemini Enterprise Includes

The pre-merger portfolio had two distinct components, each addressing a different enterprise need.

Agentspace (now Gemini Enterprise) was the consumption layer: Google-quality multimodal search across company data, with connectors to Drive, SharePoint, Jira, Confluence, Salesforce, Box, and other enterprise repositories. It provided enterprise access controls with role-based permissions and identity provider integration, and offered tiered licensing (Enterprise for search, Enterprise Plus for assistant features). Its most significant capability was a custom agent marketplace where organizations could publish, discover, and govern agents centrally.

Vertex AI Agent Builder (now the Gemini Enterprise Agent Platform) was the developer layer. It provided Agent Studio for building and testing agents, the Agent Development Kit (ADK) for programmatic agent construction, and Agent Engine (now Agent Runtime) for managed deployment. The platform included RAG and grounding capabilities that connected agent responses to enterprise data sources.

The merger brings these under one umbrella, but the architectural challenge remains. Consumption and development are different activities with different users, different pricing models, and different success metrics. Microsoft solved a similar integration problem with Copilot Studio and Agent 365 by layering governance across both surfaces. Salesforce solved it by anchoring everything in the CRM data layer. Google's integration strategy is to anchor everything in the model layer, using Gemini's capabilities as the connective tissue. Whether model-layer integration creates the same stickiness as governance-layer or data-layer integration is the central strategic question for Google's enterprise AI business.

A2A: Google's Interoperability Play

Google's most significant strategic contribution to the broader enterprise AI ecosystem is not a product. It is the Agent-to-Agent (A2A) protocol, which addresses the coordination gap between agents running on different platforms.

Google unveiled A2A in April 2025, and the protocol's trajectory since then has been one of the fastest standards adoption stories in enterprise technology. By June 2025, Google donated A2A to the Linux Foundation. In August 2025, IBM's Agent Communication Protocol merged into A2A, consolidating two competing approaches. In December 2025, Anthropic's Model Context Protocol (MCP) joined the same umbrella organization, the Agentic AI Foundation (AAIF). By August 2026, A2A formally transferred into the AAIF, and the foundation had grown from 49 founding members to over 250 organizations. The platinum signatories include AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI.

The relationship between A2A and MCP is complementary, not competitive. MCP provides the protocol for tool and data connectivity: an agent discovers and invokes external tools through MCP servers. A2A provides the protocol for agent-to-agent communication: agents discover each other, negotiate capabilities, delegate tasks, and coordinate execution. Both are now separate projects under the AAIF, each with its own technical steering committee. The Agent Coordination Gap Part 2 documented MCP's reach (97 million monthly SDK downloads, over 10,000 active public servers) but noted that 86 percent of implementations remained experimental. A2A addresses the layer above tool connectivity, the coordination layer where agents must resolve conflicts, delegate work, and maintain coherent multi-agent workflows.

For Google, A2A serves a dual purpose. It positions Google as a steward of open standards in a market where Microsoft and Salesforce have built proprietary coordination layers. And it addresses Google's own fragmentation problem: if agents built on Gemini Enterprise can coordinate with agents built on Bedrock or Azure through an open protocol, the model layer becomes the differentiator rather than the platform. Google wins when the platform wars are fought on model quality, not on proprietary coordination mechanisms.

The risk is that A2A's governance model is now outside Google's control. The AAIF is a neutral foundation with multi-stakeholder governance. Google originated the protocol but no longer owns it. If the foundation's direction diverges from Google's enterprise strategy, Google loses the ability to steer A2A toward outcomes that favor its competitive position. That is the trade-off of open standards: credibility in exchange for control.

The Model Advantage and the Multi-Model Hedge

Google's model capabilities are the foundation of its enterprise strategy, and the company is making a substantial financial bet that model leadership is the durable competitive advantage in the agentic era.

Gemini is the core offering, and Google reports 1,302 generative AI use cases across its cloud customer base, with approximately 75 percent of Google Cloud customers leveraging AI products. Named enterprise wins include Home Depot (Gemini Enterprise shopping agents), Merck, DBS, and Mattel. These are real deployments at meaningful scale, not proof-of-concept pilots.

But Google's model strategy is not purely proprietary. The company's investment in Anthropic tells a more nuanced story. At Cloud Next 2026, Google announced up to $40 billion in investment in Anthropic ($10 billion in cash plus $30 billion tied to performance targets) at a $350 billion valuation. Total Google commitment to Anthropic now reaches approximately $43 billion. The deal includes 5 gigawatts of compute capacity over five years and up to 1 million TPU access. Claude is distributed through Vertex AI to thousands of enterprise customers alongside Google's own Gemini models.

This creates an unusual competitive dynamic. Google is simultaneously the builder of Gemini and the largest investor in Gemini's most capable competitor. Anthropic distributes Claude via AWS Bedrock, Microsoft Azure Foundry, and Google Cloud Vertex AI, making Claude a multi-cloud model that competes with Gemini on Google's own platform. Industry analysts describe a circular deal structure where cloud providers invest in AI companies and earn revenue selling them compute.

The strategic logic is defensive. If enterprise customers want Claude, Google would rather serve Claude on its own infrastructure than lose those customers to AWS or Azure. The multi-model approach acknowledges that no single model wins every use case, and that enterprises will adopt model portfolios rather than committing to a single provider. By hosting both Gemini and Claude (and other third-party models), Google positions Vertex AI as a model marketplace rather than a single-model platform.

The tension is real: every Claude deployment on Google Cloud validates Anthropic's independence from Google, and every Anthropic advance makes the investment relationship more valuable while simultaneously making Gemini less differentiated. Google is betting that platform economics (compute margin, infrastructure stickiness, enterprise relationships) outweigh model exclusivity. That bet is rational, but it requires Google to maintain model competitiveness even as it finances a competitor's research. If Gemini falls behind in capability, the multi-model strategy becomes less a hedge and more a concession.

The Sunset Problem

Google's enterprise credibility problem has nothing to do with technology. It has everything to do with trust.

The catalog at Killed by Google tracks 299 product retirements since 2006, with an average product lifespan that enterprise buyers find difficult to reconcile with the multi-year commitments cloud contracts require. Recent examples reinforce the pattern: Google VPN was discontinued in June 2024, Jamboard ended the same year, the Google Checks privacy compliance platform is closing in September 2026, and even the consumer GitHub version of Gemini Code Assist was shut down (though the enterprise version continues).

The messaging product history is particularly damaging. Google Talk, Allo, Duo, and Hangouts each launched as a primary communications platform, gained adoption, and were killed or replaced by a successor that was itself eventually discontinued. For enterprise buyers evaluating a multi-year commitment to Gemini Enterprise, that history is not ancient. It is a pattern.

No formal enterprise survey data measures how Google's sunset history affects buyer confidence at a granular level. But the concern surfaces consistently in practitioner conversations and analyst commentary. One analyst reviewing Cloud Next 2026 announcements noted that Google's customer growth claims "does not settle the argument for me as to what AI capability will drive a shift in new customer signups." AWS treats customer trust as worth the cost of keeping legacy products alive. Microsoft maintains backward compatibility as a core operating principle. Google's culture of shipping fast and deprecating faster is a strength in consumer products and a liability in enterprise sales.

The Gemini Enterprise rebrand is partly an attempt to address this. By consolidating multiple products under a single brand, Google reduces the surface area for future sunsets. If Gemini Enterprise is the platform rather than a collection of products, individual feature deprecations become platform evolution rather than product kills. At Cloud Next 2026, Google pitched the Anthropic partnership partly as a trust signal: committing $43 billion to a single AI partner is not the behavior of a company that abandons its bets lightly.

Whether this narrative shift works depends on execution over years, not months. Enterprise trust is earned through consistency, not announcements. Google needs to demonstrate that Gemini Enterprise will still exist, still be supported, and still be competitive in 2030. No rebrand can substitute for that track record. It has to be built.

The Coordination Position

Google's competitive position in the agent coordination landscape is unusual: the company originated the leading open protocol for agent-to-agent communication but has not yet built the strongest coordination infrastructure on its own platform.

The Agent Coordination Gap Part 6 documented the platform lock-in dynamics that emerge when enterprises build their coordination layers on proprietary vendor infrastructure. Google's open-protocol approach through A2A addresses this risk directly: if coordination happens through a vendor-neutral protocol under independent governance, no single vendor captures the coordination layer. That is good for the ecosystem and good for Google's credibility, but it means Google does not control the coordination layer the way Microsoft controls governance through Agent 365 or Salesforce controls the application layer through Atlas.

Gemini Enterprise Agent Platform provides agent development and deployment capabilities, but the coordination features are still maturing. The agent marketplace in Gemini Enterprise provides discovery and governance within the Google ecosystem. Cross-platform coordination, the ability for a Gemini Enterprise agent to discover, negotiate with, and delegate to an agent running on Bedrock or Azure, depends on A2A adoption across the ecosystem. Google has the protocol. It needs the implementation maturity.

The ADK (Agent Development Kit) is Google's answer at the developer level, and it integrates A2A natively. But developer tooling is necessary, not sufficient. Enterprise coordination requires governance, audit, conflict resolution, and human-in-the-lead decision authority across agent boundaries. These are the capabilities that the Agent Coordination Gap series identifies as the hardest unsolved problems, and they are the capabilities that no vendor, including Google, has shipped in a production-ready form.

The Four-Layer Map

The first four articles in this series identified three strategic layers in the enterprise agentic AI market. Google adds a fourth.

Microsoft owns the governance and control plane layer, with Agent 365 providing identity, security, lifecycle management, and policy enforcement. The lock-in mechanism is governance dependency.

Salesforce owns the application and domain layer, with Agentforce 360 and the Atlas Reasoning Engine providing customer-facing agent capabilities anchored in CRM data. The lock-in mechanism is data gravity.

AWS owns the infrastructure layer, with Bedrock AgentCore providing framework-agnostic compute, managed runtimes, and agent payment infrastructure. The lock-in mechanism is operational gravity.

Google owns the model and research layer. Gemini provides the foundation model capabilities, A2A provides the open coordination protocol, and the Anthropic partnership provides multi-model optionality. The lock-in mechanism is model capability and research velocity: the ability to ship new model capabilities faster than competitors can replicate them.

The model layer is both the most powerful and the most fragile competitive position. Models improve rapidly, and today's leader can be tomorrow's runner-up. Google's $43 billion investment in Anthropic is an acknowledgment that model leadership alone is not a durable moat. The Gemini Enterprise rebrand is an attempt to build platform-layer stickiness around model-layer excellence. Whether Google can hold both positions, model leader and enterprise platform, determines whether it becomes a full-stack competitor to Microsoft and AWS or remains the best research lab with the weakest enterprise sales motion.

For enterprise leaders, Google's position creates a specific evaluation question: does your organization value model capability (and the multi-model optionality that Google provides through its Anthropic partnership) more than governance coherence (Microsoft) or infrastructure neutrality (AWS)? Organizations that are model-sensitive, that need the best available reasoning for complex agent workflows, have a reason to build on Gemini Enterprise. Organizations that prioritize operational predictability and trust in their vendor's long-term commitment need to weigh that model advantage against Google's enterprise track record.


Strategy Playbook

1. Test Model Quality Against Your Actual Workloads. Google's competitive advantage is model capability, so test it empirically. Run the same agent workflows on Gemini, Claude (via Vertex AI), and at least one Bedrock-hosted model. Measure reasoning accuracy, latency, token efficiency, and cost per completed task across your specific use cases. If Gemini consistently outperforms on your workloads, the platform has a defensible value proposition for your organization. If model performance is roughly equivalent across providers, the platform decision should be driven by governance, infrastructure, and coordination criteria instead.

2. Assess the Consolidation Risk. The Gemini Enterprise rebrand is less than a year old. Before committing significant workloads, map the current feature set against the pre-merger Agentspace and Agent Builder capabilities. Identify any features that were available in the separate products but are not yet available in the consolidated platform. Establish a quarterly checkpoint to verify that the consolidation roadmap is delivering on schedule. Google's enterprise trust problem is a track record problem, and the only remedy is consistent execution over time. Build your evaluation timeline accordingly.

3. Evaluate A2A Readiness for Cross-Platform Coordination. If your organization runs agents on multiple cloud platforms (or plans to), A2A is the most promising open protocol for cross-platform agent coordination. Assess the current A2A implementation maturity on each platform you use. Test agent discovery, capability negotiation, and task delegation between a Gemini Enterprise agent and agents on other platforms. The protocol is under active development, so evaluate the gap between the specification and the production-ready implementation, and plan for iterative adoption rather than a single deployment event.

4. Build a Sunset Contingency Into Your Architecture. This is specific to Google and reflects the company's historical pattern. If you deploy on Gemini Enterprise, design your agent architectures with portability in mind. Use open protocols (A2A for coordination, MCP for tool connectivity) wherever possible. Minimize dependencies on Google-proprietary APIs that do not have equivalents on other platforms. Build data export and agent migration runbooks before you need them. The goal is not to avoid Google. The goal is to adopt Google's capabilities while maintaining the architectural optionality to move if the platform's trajectory changes. Enterprise leaders who built on previous Google platforms without exit plans learned this lesson the hard way.


Arion Research advises enterprise leaders on AI strategy and the shift to a digital workforce.

Michael Fauscette

High-tech leader, board member, software industry analyst, author and podcast host. He is a thought leader and published author on emerging trends in business software, AI, generative AI, agentic AI, digital transformation, and customer experience. Michael is a Thinkers360 Top Voice 2023, 2024 and 2025, and Ambassador for Agentic AI, as well as a Top Ten Thought Leader in Agentic AI, Generative AI, AI Infrastructure, AI Ethics, AI Governance, AI Orchestration, CRM, Product Management, and Design.

Michael is the Founder, CEO & Chief Analyst at Arion Research, a global AI and cloud advisory firm; advisor to G2 and 180Ops, Board Chair at LocatorX; and board member and Fractional Chief Strategy Officer at SpotLogic. Formerly Michael was the Chief Research Officer at unicorn startup G2. Prior to G2, Michael led IDC’s worldwide enterprise software application research group for almost ten years. An ex-US Naval Officer, he held executive roles with 9 software companies including Autodesk and PeopleSoft; and 6 technology startups.

Books: “Building the Digital Workforce” - Sept 2025; “The Complete Agentic AI Readiness Assessment” - Dec 2025

Follow me:

@mfauscette.bsky.social

@mfauscette@techhub.social

@ www.twitter.com/mfauscette

www.linkedin.com/mfauscette

https://arionresearch.com
Next
Next

The Platform Lock-In Trap